GA4GH Cancer Community

Rolling Agenda and Actions

Presentation Scheduler

Leadership Rolling Agenda

2026-07-28 Structural Variant Aggregation

2025-03-10 - CANDEL

2025-02-10 - PMATCH

2024-12-9 - WAYFIND-R Program

2024-11-18 - ZERO Childhood Cancer Beacon

2024-09-16 - GA4GH Connect: Cancer Data Commons

August 12, 2024 - CSIRO Serverless Beacon Implementation

July 8, 2024 - Variation Representation Specification (VRS)

April 24, 2024 - GA4GH Connect: Cancer Beacon part 2

March 11, 2024 - Cancer Beacon part 1

February 12, 2024 - CancerModels.org

January 8, 2024 - CRDC + Velsera

December 11, 2023

September 19, 2023: GA4GH Connect

August 17th, 2023: ELIXIR Cancer Group & GA4GH Cancer Community

June 12th, 2023

May 8th, 2023

February 13th, 2023

January 9th , 2023

September 12 , 2022

June 13, 2022

April 11, 2022 [CANCELED]

March 14, 2022

February 14, 2022

January 10, 2022

December 13, 2021

November 8, 2021

October 18, 2021

September 13, 2021

August 16, 2021

July 12, 2021

June 4, 2021

March 2, 2021

January 25, 2021

October 1, 2020

2026-07-28 Structural Variant Aggregation

Meeting Recording LINK

Passcode: 67*#!A+E

Slides: Structural Variant Meeting - July 28, 2026

Attendees: Alex Wagner, Bernie Pope, Wayne Clarke, Senkei Umehara, Kaitlyn Brethour, Melissa Cline, Christopher Preusch, Emma Tudini, Sarah Chan, Mrinal Thomas, Tim Hefferon, Timothee Cezard, CBullen, Stuart A. Scott, Tiffany Leung, Susan Walker, Christine Von Raesfeld, Ellie Yang, Nora Thomas, Gabriela Repetto, Jamie, Marla Mendes, Marsha M. Wheeler, Ting Chan, Jeroen Van Den Akker, Raymond Chan, Hannah Snow, Ronak Patel, Tristan Nelson, Kejian Zhang, Amanda Spurdle, Worawich, Jennifer Hart, Kevin Chau, Tanya Nelson, Akanchha Kesari, Fred Zhou, Christa Lese Martin, Ramona Rego, J. (Gus) Gustafson, Hugh French, Catherine Procknow, Daniel Cameron, Nadezda Volkova, Melissa Landrum, Bob O’Connor

Overview: The purpose of the meeting will be to briefly frame the SV aggregation landscape, discuss key challenges, and gather input on use cases across the GA4GH community. The goal is ultimately to develop effective matching strategies and statistical inference from cohort studies involving structural variants. While this initial call is hosted by the cancer community, we encourage everyone to attend and share their experiences!

Next steps:

• Kaitlyn - Send out a survey to all participants to self-register interest, specify their role (tool provider, data resource, data consumer), and describe their interest and ability to contribute to the group.
• All participants - Complete the survey promptly and provide through email on areas of interest, use cases, and specific scenarios or examples that highlight challenges in structural variant aggregation.

Summary:

This meeting focused on discussing structural variant aggregation challenges and exploring the development of standards and guidelines for handling structural variants (SVs) across different genomic resources and applications. Alex from the GA4GH VRS product team and GKS Work stream provided an overview of the Variation Representation Specification (VRS) standard, including its ability to represent structural variants through adjacency models and how it spans both small variants and complex structural rearrangements. The discussion highlighted key challenges in aggregating SVs, particularly the difficulty in determining equivalency between different structural variants, especially in complex cases like chromothripsis where genomic rearrangements are random and highly variable. Participants shared their experiences with SV aggregation, including Christa Martin's work with ClinGen on copy number variants and rare diseases, Melissa Landrum's insights from ClinVar's evolving approaches to structural variant curation, and Daniel Cameron's technical perspective on the lack of consensus ontologies for describing SV events. The group discussed the need for clear definitions around what constitutes an SV event, how to handle complex events, and whether to focus on genomic rearrangement descriptions versus functional impact. The conversation ended with plans to send out a survey to gather more detailed information about participants' interests, use cases, and ability to contribute to developing standards and guidelines for SV aggregation.

1. Introduction to the topic

The meeting focused on discussing structural variant aggregation, particularly in the context of cancer research but with broader applications across human health. Bernie Pope, co-lead of the cancer community, introduced the topic, explaining its importance for comparing variants across individuals and large cohorts, as well as for identifying previously classified pathogenic variants. The discussion highlighted challenges in defining and coordinating structural variants, especially with short-read sequencing data.

2. Landscape overview

Alex provided an overview of the Variation Representation Specification (VRS), explaining its purpose of creating interoperable standards across different genomic data structures. He detailed how VRS 2.0 enables representation of both small variants and structural variants, including adjacency events. The discussion covered VRS’ modular JSON structure and its ability to describe sequence variants across DNA, RNA, and protein types, with specific examples of how it represents sequence locations, alleles, and descriptive labels.

Alex discussed the challenges of aggregating structural variants (SVs) and highlighted the need for guidelines on what constitutes equivalent SVs for aggregation purposes. He explained that while small variants can be easily aggregated based on the variant itself, structural variants present more complexity due to different breakpoints and resultant rearrangements. The key open question identified was determining when two SVs are close enough to be considered the same for discovering recurrence-based significance in cancer and other fields. Alex proposed forming a forum to develop policies and guidelines for SV aggregation, with related topics including data structures, annotation levels, and functional effects of SVs to be considered by the group.

3. Structural Variant Aggregation Discussion

The group discussed challenges in variant curation, particularly around matching similar genetic variations across different knowledge bases and real-world cohorts. Alex highlighted the complexity of chromothripsis and CNV matching, suggesting that guidance and standards would be beneficial. Tristan and others mentioned the specific use case of mapping CNVs detected via assays against ClinVar and ClinGen knowledge bases, particularly focusing on recurrent regions and dosage mapping. Interest in copy number variants (CNVs) and structural variants (SVs), particularly in germline and somatic contexts was highlighted. Christa Martin, a principal investigator for the ClinGen project and clinical cytogeneticist, discussed the challenges of describing CNVs universally, using the 22q11.2 deletion syndrome as an example of the complexity involved in defining unique regions versus segmental duplications for clinical and research purposes.

The group discussed the scope and challenges of defining structural variant (SV) ontologies and standards. Daniel Cameron raised questions about whether the group should focus on event ontologies, complex events, and terminology standardization, while Melissa Landrum shared ClinVar's experience with aggregating structural variants and the challenges of balancing precision with practical needs. The group agreed to determine scope based on input from implementers, tool providers, and data consumers, with next steps including a survey to assess interest and participation in the initiative.

2025-03-10 - CANDEL

Meeting Recording

Presenter Slides

Attendees: Heather Creasy, Beatrice Amos, Ben Kamphaus, Marshall Thompson, Jumi Popoola, Alona Sosinsky, Sharon Plon, Kori Kuzma, Angela Page, Wesley Goar

Presentation Summary: Data harmonization is an expensive and time consuming bottleneck for translational data science in oncology. Harmonized clinical and molecular data occupy a first-class role in data systems, enabling and informing dataset discovery, hypothesis formation, and downstream analysis, while also allowing interoperability and generalizability of approaches and analyses between projects and teams. However, due to the technical complexity of harmonization, it is typically performed manually and only on subsets of available data and metadata, resulting in mostly adequate selection and filtering criteria for common diseases and cancer types, but neglecting the inclusion of metadata that may only be more useful in rare cancers and underrepresented patient groups. In addition, data harmonization tends to target a single snapshot of a data model, which means that invested effort in harmonization only provides a limited time of increased value before the current model becomes obsolete. This has led to the perception that data harmonization is brittle and often not worth it, leaving much historical research data siloed and underutilized for contemporary discovery. To lower barriers to harmonizing large datasets, the Rare Cancer Research Foundation’s (RCRF) current informatics team previously worked with the Parker Institute for Cancer Immunotherapy (PICI) to build CANDEL, an open source software toolkit and platform for oncology research. These tools simplified the process of data harmonization while representing data in both a distributed system for data snapshot recovery with minimal overhead and in a sparse data format for ingesting infrequent data fields without imposing constant, additional storage costs. This system was critical to the success of several scientific studies at PICI, was built and maintained by a small team, and has been maintained as an internal software system at PICI with minimal support intervention. RCRF now proposes to extend the CANDEL open source tools into a community-wide data harmonization and storage toolkit named UnifyBio, which will make efficient and accurate data harmonization a reality for cancer researchers of any discipline across the translational research ecosystem. To enable broad adoption, we will extend these open source tools to interoperate more thoroughly with the tools and data formats used by the larger translational cancer research community, while also enhancing and removing friction from the data harmonization user experience. We will further encourage community use and adoption of this system by creating written and multimedia tutorials and examples, including an open source, canonical configuration and deployment at RCRF designed to increase accessibility of rare cancer datasets. The successful development and adoption of UnifyBio will address the community’s needs for data harmonization and efficient distributed and time-auditable storage and will therefore ensure downstream access to more complete datasets, more consistent data models, more inclusion for underrepresented and underserved populations and cancer types, and, ultimately, insights that lead to more clinical advances for all cancer patients.

2025-02-10 - PMATCH

Meeting Recording

Attendees: Lincoln Stein, Anne Deslattes Mays, Benjamin Haibe Kains, Elif Tuzlali, Malachi Griffins, Trevor Pugh, Zina Perova, Kori Kuzma, Ophira Ginsburg, Sharon Plon, Alona Sosinsky, Catalina Lopez Correa, Mrinal Thomas, Jumi Popoola, Neerjah Skantharajah

PMATCH

• Setting the Scene

• Clinical trials are increasingly complex, involve multiple sites, and technologies, and lack a systematic infrastructure for matching patients
• Challenges that PMATCH aims to solve involve genomic alterations, clinical data, structured clinical trial info and clinical trial status information
• Clinical trials are a cornerstone of precision oncology but the current system is failing patients

• The opportunity: molecular tumor board in a box

• PMATCH brings together and matches these suites of software to provide a clinical trial matching infrastructure
• The ultimate goal is to increase the number of patient matched of trials

• PMATCH interlinks data needed for clinical trial matching and learning

• When trying to make a clinical trial match there are different activities involved

• Extracting eligibility criteria from trial protocols

• Eventually hope this will be facilitated through CTIMS which enables abstract of matching criteria into a standardized, machine readable file, CTML - (clinical trials markup language)

• Pulling out clinical data - entirely manual today
• Genomics data - largely structured and include good standards
• Taking the clinical and genomic data and matching them
• Tie breaking infrastructure to improve matches

• SP: This is assuming all these protocols are open or is there a way for institutions to share only those that are open?
• TP: We have to abstract them depending on the hospital, can’t do it all at once right now
• OG: EHRs are moving towards EPIC and that’s an opportunity here as some hospitals are just now getting on
• Extracting patients' clinical trial data is actually pretty achievable and will be completed in the more short-term, and it’s going well. The hard part is the logic for matching
• CTIMS has been used manually so far and it’s in use at Princess Margaret Cancer Centre
• LS: Even if you can’t get accurate matching in one pass you can take advantage of the interactive nature of LLMs. The person doing the matching can have a dialogue with the LLM that has access to the patient record and clinical trial and refine things. Have you thought about that?

• TP: Yes, that is a pretty feasible future state. The issue is getting the information into a machine readable format.
• LS: You can do a semantic search over. Take the clinical trials and vectorize them and put them into a vector database. Take the vectorized version of the patient record and do the search which will give you similar matches and now you have manageable small number of potential matches that the LLM can reason across
• TP: Do you need ground truth to train that?
• LS: It’s a rag search. Retrieval augmented generation search
• TP: We attempted something similar and it got tripped up at cancer subtypes
• LS: That is where you would extract in real time the structure data and do a match on that. You can then use that for reinforcing training.
• TP: Scale?
• LS: Do this as the months go by, and it gets more and more accurate. The universe of clinical trials is not so large.
• TP: Would like to see PMATCH supporting these types of projects long term. Would be hard to start that right at the beginning without a more structured approach.
• LS: Its amenable to a little pilot to test.
• SP: Oncologists would like to know that they’re doing something to improve it. Sometimes if you use something and the matches aren’t good, you’re frustrated. But if you can provide feedback, you have more buy in.
• TP: Currently it’s a slack thread between curator and clinical fellow but could definitely be more technically savvy.

• Clinicaltrials.gov is an important resource of information about trials, but has some issues

• One problem is the data being stored in there - it’s not the whole protocol
• Mechanism of action is not in there and the eligibility criteria is very mixed across studies
• Required fields are well done, but if you want to get specific there’s a lot of variability across trials
• Specific information isn’t always computationally available
• SP: Not a structure field for cancer type?
• TP: There’s a field but the way it’s used is variable. For example, you could tag it with lung cancer instead of a specific subtype of lung cancer that would have been more accurate

• Evaluating the matching engine on a large retrospective clinical trial

• OCTANE supported by OICR
• Active genomic sequencing study

• Clinical trial data has been standardized across Marathon of Hope fields

• The system being used here is EPIC, but there are still some non-EPIC systems that need to be aligned
• Want to use AI to extract clinical patient data

• Oncographer links clinical data collection to the MOHCCN data standard (140 fields) as a graph database and can be imported into cBioPortal
• MATCHER assembles data and identifies potential trial matches for patient
• Test set - 3800 patient, 135 clinical trial arms
• Some trials matches more than others, some were more restrictive
• Insight from exploring this data:

• Problematic arms that matched on “solid tumors” identified need to better support exclusion criteria (e.g. prior immunotherapy, NOT mutant genes etc.)

• Work in progress to validate matches, enhance matching criteria, and abstract more trials to identify edge case

• There were some trials that didn’t match with anyone
• Are there some people that would never match - if so we shouldn't go through the trouble of approving those arms
• Retrospective data can help with developing eligibility criteria early on
• When a patient matches with multiple trials how do you pick which one is the best? Introducing PRIORITIZER

• PRIORITIZER: ranks therapeutic options based on additional biomarkers and published signatures

Discussion - Bias due to prioritization based on biomarkers

• LS: Is there any concern that you’re biasing who goes into which arm of the trial by changing the pre-test likelihood that the patient will respond to the drug in that arm?
• TP: yes, that’s true it’s not going to be random sampling to the arm. But usually there’s not a communication link between the trials. So when deciding between trials you don’t need to be randomized
• SP: Not always true in pediatrics
• Does that bias the assignment of kids to the trial when ranking the trials
• Even for match arms they definitely had a ranking. If you met multiple arms they had a preset order
• BH: That kind of implicit prioritization has been happening forever. During molecular tumor boards for example the patient can decide to go for trial X vs Y it’s just that here it’s systematic.
• LS: Let’s say we’re looking for a response biomarker by looking at people who went into the arm vs the other arm. If the algorithm is picking up a signature and biasong ppl with that signature ino that arm isn’t that weighting the scale?
• TP :It’s rare that the trials set out for a trial biomarker. But you’re right that it would result in a molecularly purified population
• LS: Selection by the biomarker should then be part of the selection criteria if it’s a positive study. All for good, but adding a layer of complexity
• BH: IF you have a trial that has a very complex treatment regimen, then you'll have a bias of patients due to requirements. If it’s not documented it could be problematic. But if it’s documented it becomes one of many confounding factors.
• TP: The bias should probably be documented somewhere. Should be caught in the correlative analysis of the trial.
• BH: You have full traceability of the data through the matching.
• OG: Is there any way to model what would be expected without PMATCH? Do you have a historical baseline?
• TP: OCTANE is a good example - PMATCH didn’t exist when the trial was ongoing. So we could go back and compare what trials they actually went on vs if it were done with PMATCH

• Everyone has agreed on the problem but approached the solution differently

• Is there a common set of approaches we can agree on, specifically around structuring the clinical trial protocol
• How do we get this information in front of the clinical, molecular tumor board, and patients?

• Clinical Trials matching tab being built within CBioportal with patient view summarizing matched trials including what arms, what fields, and also who did the match (PMATHC, Dana Farber etc.)

•  Not live yet

• Going forward I would like to see a core set of clinical trials that have been extracted in a standardized way and patient data…

Questions

ZP: As you were explaining how variable the info is from clinicaltrials.gov, has there ever been an attempt to improve the registration of clinical trials? It’s surprising that there isn’t even use of controlled vocabulary.

        TP: Not appreciated as a major problem. The matching exercise brought this problem to life.

LS: The existence of PMATCH and lots of similar projects will incentivize people to improve their submissions to clinicaltrials.gov

BH: Yes but ct.gov will never make machine readable, compete eligibility criteria mandatory

LS: clinicaltrials.gov could implement a friendly chatbot to help trialists enter the data.

BH: So I think we still have to rise the bar, come up wit the standard, ver the ct,gov entries and spread the word

LS: Have you thought about patient oriented clinical matches based on patient needs?

TP: We have considered that, especially with regards to distance. What wasn’t clear was where we should interject that. Is that a prioritizer job, is this in the first layer?

LS: A patient might want to ask a chat bot.

OG: At the outset of designing something like this we should have someone with lived experience involved.

OG: National clinical trials network? Let me connect you with Sheila Prindiville NCI

Resources from Chat:

• CTRP: https://www.cancer.gov/about-nci/organization/ccct/ctrp
• TrialGPT: https://datascience.cancer.gov/news-events/news/large-language-models-help-match-patients-clinical-trials
• https://commonsenseoncology.org/
• https://www.sciencedirect.com/science/article/pii/S1470204524004510?dgcid=coauthor
• FYI/ Patient advocates are very involved in this. Common Sense Oncology: outcomes that matter Lancet Oncology 2023 https://www.thelancet.com/journals/lanonc/article/PIIS1470-2045(23)00319-4/abstract

2024-12-9 - WAYFIND-R Program

Meeting Recording

Attendees: Daniel Puthawala, Erika Schirghuber, Yasmeen Kurdi, Neerjah Skantharajah, Zina Perova, Melanie Courtot, Dianne Nicol, Grace Pendlebury, Jaclyn Estrin, Jumi Popoola, Ayaat Hassan, Kori Kuzma, Elif Tuzlali, Grant Wood, Alex Wagner

Minutes 

WAYFIND-R: A global precision oncology registry & evidence generation platform - Erika Schirghuber

The goal of WAYFIND-R is to make data available, interoperable, & shareable for research to advance precision oncology

• Example: When clinicians are deciding if a patient is eligible for a specific therapy, Next Generation Sequencing (NGS) testing is often ordered

• Challenges of NGS in clinical care

• Clinicians need to be informed on types of NGS testing and how results inform decision making
• Clinicians need to know if there is a molecularly matched therapy available or clinical trial
• Understanding how clinical trial results translate to the real world, as trial cohorts can have differences from patient populations

• Evidence gaps in precision oncology 

• Molecular tests are heterogenous, their utility may be dependent on intended use
• Conducted RCTs in rare cancers is challenging > obstacle for evidence generation
• Duration of follow-up in RCTs > limited evidence of long-term prognosis & safety

• WAYFIND-R registry 

• Set up similar to a clinical trial where patients are consented for data collection + secondary/tertiary use
• There's a study protocol that undergoes REB review at each participating hospital site
• Patient + tumor characteristics are collected
• In a longitudinal manner, treatment decisions & patient outcomes are collected
• The variables collected were defined with an expert panel to ensure regulatory & ethical compliance and consideration
• In parallel, patients can be enrolled in a clinical trial if eligible
• This registry follows patients from initial diagnosis to end of life / withdrawal of participation
• 32 countries across 5 continents are participating, including academic, hospital, & private clinic sites with a variety of standards of care
• Currently 4800 participants enrolled in the study representing different cancer types (lung, colon, breast, pancreas, uncommon cancers, ovary, prostate, others)
• Opportunity to conduct epidemiological studies + sub studies with this vast data

• Generating real-world evidence (RWE) from real-world data (RWD)

• RWD: data relating to patient health routinely collected from a  variety of sources outside of traditional controlled clinical interventional trials, under real-life clinical practise
• RWE: the evidence derived from the analysis and/or synthesis of RWD
• Making data research ready involves: broad inclusion, standardised clinical data elements, fit-for purpose data collection, & the ability to analyse complex data sets

• WAYFIND-R Data sharing and collaboration platform 

• Dashboard: available to participating clinicians to allow data exploration & descriptive parameters
• Analytical Research Environment: supports researchers to access anonymized individual participant data for secondary data use studies
• Collaboration platform (future):  supporting ancillary studies
• Data access requirements are built in to the IT infrastructure of the platform including: submitting a data access proposal, review by an independent data access committee, signing a data sharing & use agreement
• The data is mapped to the OMOP common data model to facilitate interoperability
• To see a live demo of the platform, see recording from 19:15 - 27:00

Discussion + Q&A

JP: Are only patients with clinical grade NGS represented in the registry?

ES: Any NGS assay that’s been used in routine care is eligible. This could be commercial end-end tests but also locally developed tests that have been validated for routine care. Any panel size is incorporated as well.

MC: Is the software entirely developed by Roche? Or are you relying on existing software, for example, for visualization?

ES: we’d love to learn from GA4GH and the research communities which tools and knowledge bases would be useful to support your research. Most of our software is developed in house but we’re looking into what’s available in the open source community

ES: If you’re using clinical genomic databases, which kind of tools would you need to have?

MC: I was thinking about CBioPortal which has similar concepts.

ZP: the visualization reminded me of CBioPortal as well. This is an interesting potential collaboration.

ZP: You talked about registered clinicians who have access to the platform. What type of additional collaboration do you envision for this platform, beyond registered users?

ES: the WAYFIND-R registry database is open to any academic researcher worldwide. This group for example could log into the platform and use the query builder to request data access. In terms of future opportunities for collaboration, we are considering supporting federated data infrastructure at some point.

GW: How many users of WAYFIND-R do you have today outside of Roche? What are your requirements to have access?

ES: The WAYFIND-R platform is accessible to any academic researcher. There’s a master data sharing agreement available on our website. In terms of users, we have participants across 32 countries & 119 hospitals, and we’re managing 1000+ user accesses

MC: Would that be an opportunity for WAYFIND-R and GA4GH to align eg on authentication/authorization with a view of supporting at least federated discovery (eg via Beacon)?

ES: Absolutely, we’re happy to have the discussion and get your advice on what is actually used internationally because we want to drive international standards

MC: You mentioned that you’re looking into federation, I think that could be a point of alignment with GA4GH (re: data discovery).

GW: of the 4800 study participants, how many have their tumors sequenced and who is paying for the sequencing?

ES: All of them. It’s an inclusion criteria as per the study protocol. The sequencing and treatment are done as per standard routine care. This is an observational study, so patients are undergoing the norm of their respective jurisdiction (some countries have reimbursement systems, some pay out of pocket).

ZP: you mentioned that when choosing the participants, you try to balance the gender/countries/settings. I’m wondering if there was anyone you approached who did not agree to participate and the reasons.

ES: One might introduce a selection bias in the data collection by only enrolling patients who are consenting and willing to participate. That’s why we implemented a “short” consent form to only collect minimal information from patients who aren’t willing to participate.

BHK: You mentioned that WAYFIND-R contains the eligibility criteria for the Roche trials, so I misunderstood

ES: So WAYFIND-R is a non interventional study. It has a study protocol in which patients are consented for data collection, but also secondary data use. This protocol has to be reviewed by local REBs

BHK: I’m looking for a way to access eligibility protocols for full clinical trials, and I was wondering if that’s something you provide through the system, at least for your trial. Is there a platform that advertises Roche eligibility criteria more broadly?

ES: all our clinical trials are registered on clinicaltrials.gov, but what specifically are you looking for?

BHK: Clinical data, clinical variables, and molecular biomarkers. The problem is that clinicaltrials.gov is a registry, and not a database that’s been designed to store eligibility criteria. Getting access to full protocols is difficult.  

ES: we are also looking into clinical trial matching, particularly for molecularly characterized patients

ES: Is there any experience within the GA4GH community on OMOP and OMOP genomic data vocab and conditions? Or any plans to work with the Odyssey community to drive the standards?

MC: That would probably fall under the Clinical & Phenotypic Data Capture Work Stream. Grant. I know you’re worked with FIHR. I don’t know if we’ve worked with OMOP per se.

GW: No we haven’t worked with OMOP

MC: I will bring that back to the group, I’m sure we’ve talked to them at some point but I can’t remember the status  

ES: That would be super curious, because we also looked into CBioPortal and at the time of development and decided not to go with it before we were aware that many clinicians are using this data format. We moved forward with the OMOP data model. If there’s a request and need to move in this direction, we’d be interested to learn more.

NS: Phenopackets doesn’t depend on OMOP (need to confirm) but there is an outstanding question if other standards like Beacon do depend on it.

Actions

• Melanie to follow up with Moni & Grant re: OMOP
• Yasmeen to loop Erika into clinical trial matching discussions in the new year

2024-11-18 - ZERO Childhood Cancer Beacon

Meeting Recording

Attendees: Marie Wong-Erasmus (ZERO Childhood Cancer), Romina Royo, Robin Navest, Liina Nagirnaja, Zina Perova, Alona Sosinsky, Daniel Barrowdale, Natassa, Kori Kuzma, Beatriz Ruz, Jaclyn Estrin, Yasmeen Kurdi, Neerjah Skantharajah, Beatriz Ruz, Rachel Burnside, Sara Morsy, Mrinal Thomas, Jeff Liu

Minutes

•  ZERO Childhood Cancer program 

• consists of all the pediatric hospitals in Australia + 2 New Zealand, they also supports research institutes around the world
• It’s a precision medicine program using genomic analysis to identify the most effective treatments for children with cancer
• Goal: to improve survival, reduce side effects, & enhance scientific understanding of pediatric cancer real time
• Consists of a comprehensive molecular + drug testing program for kids w/ cancer that started w/ a pilot study called TARGET (2015-17) which recruited kids with the highest risk of cancer of relapse ( <30% chance of survival)
• after this success they moved on to a national trial called PRISM 2017-2022 (children w/ the highest risk cancer)
• After this success, they expanded it even further to more kids with cancer in an ongoing trial called ZERO2 (2023+)
• Zero generates one of the largest real world precision medicine datasets (>2000 patients, >130 M variants), including WGS, RNAseq, & methylation data
• This program also includes a 1) preclinical database with patient derived xenografts (PDX) & high-throughput screening (HTS), 2) clinical data, and 3) biospecimens
• Challenge: there isn’t a way to search through all this data that wouldn’t require programming expertise

• Scaling up precision medicine 

• Major bottleneck is variant curation needed to understand very high risk cancer for which little information is available, MTB, and reporting
• Marie details how the program has moved towards standardization from 2017-2024 (see recording from 07:00), where they now employ a plethora of tools and platforms

• The GA4GH Beacon standard allows users to browse Zero’s catalog of data at a cohort level vs individual patient level
• They currently also employ CAVATICA for their analysis pipeline, which is a 7 Bridges platform built in collab with the Children’s Hospital of Philadelphia

• Towards harmonization of childhood cancer data globally 

• With Australian Biocommons, ZERO had a project called the Human Genomes Platforms Project, with the goal of creating virtual cohorts, where they landed on the Beacon V2 reference implementation, to allow people to browse the catalog of data
• Beacon allows researchers within the same Beacon network to discover data at different levels: ZERO, National, Global
• Beacon also makes data FAIR: findable, accessible, interoperable, & reusable
• ZERO’s Beacon endpoints currently include individual & biosample data

• Variant data is not currently available, but will be soon after legal review
• Data types within the endpoints are mapped out onto ontologies like NCIT, OMIT, EFO
• Data is currently limited to diseases with at least 5+ individuals to maintain patient privacy

• To enable easier querying, ZERO built their own Beacon UI from the ground up (see recording from 13:40 for details)

• This UI is going to be open source soon
• CILogon is built in for user authentication

• ZERO Beacon demo - see recording from 16:40 - 22:25

Q&A for Marie

• Q (Yasmeen):What are the key takeaways from this presentation and do you have any asks of the community or opportunities for people to collaborate?

• A (Marie): Beacon is a nice lightweight framework to install and use but there are limitations to the type of information you can query for and the response it returns. You can’t query for more than one disease at a time.
• A (Zina): This is a topic we discuss often in the Cancer Community. I run the portal called Cancermodels.org and we focus on patient derived models. It would be great to include xenografts, organoids, and cell lines in tools like Beacon, because otherwise this information gets lost or people aren’t aware of it. It would be nice to follow up on the xenografts and cell lines you’re planning to include because we also have pediatric models through CCI
• A (Marie): If you have your Beacon, it would be great to query across that, as we’re intending to have a bio resource Beacon in addition to the ZERO Beacon. If Beacon developers could extend the model to accommodate biospecimens with patients, that would be helpful.

• C (Liina Nagirnaja): I’m from the Beacon team at CRG, part of Jordi’s group, and wanted to comment that we have a new version (still V2) that is production ready which is an evolution of the reference implementation and has added functionality (see resources below)
• Q (Liina Nagirnaja): How did you generate the Beacon summary statistics on your slides?

• A (Marie): Throwing a few different queries (hits to ~2 APIs) and then aggregating results, which as you can see is a little bit slow. If there was a way to join entities, it would make queries faster. We’d love for people to reuse our AI and we’ll have instructions for how you can tweak it to your own use. We’re trying to decouple the CILogon because not everyone will use it.
• C (Zina): once the UI documentation is publicly available, we will share it with the community

• Q (Romina): Let's say I’m a researcher and I use this Beacon to find patients of interest, how would the data access process work? Should I submit a request then? Is all the data available for those that have access to the Beacon?

• A (Marie): in the portal itself, if you go to the FAQ at the top right, there’s a data access request section. Once you click it, a form pops up where you can copy all the IDs and select data types of interest. As long as it’s approved, you’ll get access to the data. We can’t share the files within the Beacon itself, which is another limitation, but we will provide you with a CAVATICA link instead (which requires creating a free account to access the files). It typically takes a couple weeks to get approved.

relevant resources

• GA4GH April Connect Cancer Beacon session summary
• CSIRO Beacon implementation presentation notes
• Beacon V2 reference implementation update GitHub

2024-09-16 - GA4GH Connect: Cancer Data Commons

GA4GH Connect September 2024 - Cancer Community Agenda.docx

August 12, 2024 - CSIRO Serverless Beacon Implementation

Meeting Recording 

Attendees [Name(Affiliation)]: Oliver Hofmann, Yatish Jain, Zina Perova, Marion Shadbolt, Jumi Popoola, Jaclyn Estrin, Xuelu (Jeff) Liu, Yasmeen Kurdi, Neerjah Skantharajah,

Meeting Minutes

• AskBeacon is an implementation built on Serverless Beacon (sBeacon) which uses live language models to make querying across the global Beacon network more accessible as it’s able to process natural language
• The 15 minute AskBeacon demo can be accessed here. A 2 minute pitch version can be accessed here.
• Depending on the level of access a user has, they will have access to different levels of granularity (one of the query parameters in the Beacon schema). Not everyone can have record level access which is a user management safety precaution

• Admin access is dependent on how your institution uses Beacon and who is designated part of the data access committee

• Q: how do end questions get translated into Beacon? The results will always be an “or”

• LLMs are able to understand the Beacon schema so when an end question is posed, the LLM will suggest a 2 step approach

• Following the demo, Yatish went more in depth with a live example of a query, which can be viewed in the meeting recording starting at 23:45

Opportunities to collaborate

• The CSIRO team is open to collaboration and encourages feedback from the community on 1) the AskBeacon tool, 2) Severless Beacon, 3) what types of data are hosted in the sBeacon
• The AskBeacon system is flexible, being able to query genomic and metadata together. If the Community has specific questions or uses cases, they can reach out to Yatish and the CSIRO team to help customize the Beacon interface for their needs

July 8, 2024 - Variation Representation Specification (VRS)

Meeting Recording 

Attendees [Name(Affiliation)]: Wes Goar, Alona Sosinsky, Anne Deslattes Mays, Beatrice Amos, David Torrents, Elif Tuzlali, Heather Creasy, Jaclyn Estrin, Kori Kuzma, Rajat Kapoor, Xuelu (Jeff) Liu, Zina Perova, Larry Babb, Daniel Barrowdale, Benjamin Haibe-Kains

Meeting Minutes

How the Genomics Knowledge Standards workstream (GKS) can help the Cancer Community

• The health data evaluation pipeline experiences multiple bottlenecks at different steps from data generation, preliminary reporting, annotation, human-interpreted resources, knowledgebase integration, and final reporting in clinical reports/studies  
• GKS aims to alleviate this interpretation bottleneck & transmit info between computer systems with accuracy across a federated variant evidence ecosystem
• The use case categories GKS addresses include: Knowledge/evidence APIs, variant API/Repos, Knowledge classification, testing results interpretation, test result annotation
• Optimizing interoperability of clinical workflows is a priority of GKS
• For more information about initiatives that employ GKS standards, please refer to the slidedeck

VRS projects

• ClinVar-GKS Project: Objective is to enhance the utility and accessibility of ClinVar datasets through developing a data transformation pipeline to incorporate GKS standards that would make it more interoperable.
• gnomAD-GKS project: GKS is working to incorporate VRS (a way to create computable identifiers for specific variation) into gnomAD to avoid issues of ambiguity and encourage precision, while providing evidence lines. The entirety of gnomAD has been VRS-ified, so variants can be searched using specific computable identifiers
• MaveDB - Multiplexed Assays of Variant Effect (MAVEs) - refers to a high throughput scale assay that enables the creation of a vast library of genetic variants that have undergone an experimental condition, which produces a variant effect map.
• VRS enables the contextualizing of MaveDB libraries, by mapping the data in the platform to the human reference genome, attaching VRS IDs, which allows for the transfer of that data into different resources that also recognize VRS IDs. Mave also developed a minimum information standard so that data models can speak to each other
• For more examples of platforms that are employing VRS IDs, refer to the slidedeck

Obstacles to knowledge matching and curation

• Categorical Variants (CatVars) link assayed variants to genomic knowledge
• (CatVars) descriptors provide a means to differentiate variations that result in the same outcome under broader categories
• CatVars represent classes of assayed variation and are related to each other in hierarchical patterns
• VICC MetaKB V1: purpose is to harmonize information from different knowledge bases (OncoKB, PMKB, CIViC, etc) and put them under a common data model
• VICC MetaKB V2: some partners privatized their work and didn’t participate in V2, it uses publicly accessible knowledge bases

Variation Categorizer (VarCat) -in progress

• Purpose of this project is variant interpretation and classification to enable efficient, reusable, and interoperable classification information, incorporating public evidence/data, data curation platforms, data normalization, variant interpretation/visualization, standardization, and industry/government/academic resources

April 24, 2024 - GA4GH Connect: Cancer Beacon part 2

Please refer to the document below for the meeting agenda, recording, slides, and minutes

GA4GH Connect Agenda- Cancer Community

March 11, 2024 - Cancer Beacon part 1

Meeting Recording 

Chairs: David Torrents

Regrets:  Romina Royo, Salvador Capella-Gutierrez

Attendees [Name(Affiliation)]: Zina Perova, Angela Page, Heather Creasy, Sharon Plon, Carlos Rodriguez Antolin, Nicolas Gaitan, Jordi Rambla, Daniel Barrowdale, Serena Kher, Lauren Fromont, Beatrize Ruz Caracuel, Jeff Liu, Lauren Hughes, Elif Tuzlali, Rodrigo Martin, bscuser, Robin Navest

Meeting Minutes

Beacon for Cancer

• Beacon is a simple server that allows for discovering data available using genomic variation information + annotations, as well as Phenoclinic information (mostly Phenopackets)
• Beacon V2 has a model that includes individuals (diseases, treatments, pedigrees etc), Biosamples (origin, tumor vs normal etc), Genomic variations (position, type, molecular attributes etc), alongside different levels of access to data
• Oncologists from EOSC4Cancer & TCGI clinics shared what features they look for in a Beacon specifically geared towards cancer. Two categories were mentioned: 1) healthcare queries (eg - how has a variant been classified, case-centric queries which are more sensitive) and 2) research focused queries
• A common issue that arises is mapping queries using natural language (which are difficult to define in a standardized manner) to Beacon
• There are 3 areas to align in terms of tailoring Beacon for cancer implementations: questions, data, tools > following this, group discussion started

Discussion/Q&A

ZP: what needs to happen to make the presence of a patient derived cancer model (PDCM) accessible through Beacon?

JR: To tailor Beacon for cancer, we have been replacing the base Beacon model with a cancer registry model (draft stages) and extending this model with images for cancer. The question is whether to add an extension data model to Beacon V2 or replacing most of the Beacon model with the schema used for PDCMs

SP: Realistically how do hospital systems with highly complex and firewalled systems participate in Beacons?

LF: We’re working right now with several hospitals in Catalunya on deploying beacons in an inter-hospital network, and unraveling new challenges for sure! But it is possible.  Right now, we’re talking to their IT departments in order to see how they can deploy a beacon within their system — once we’ve had enough use cases, we’ll publish some guidelines

JR: Beacon has different use cases. It can be installed inside the firewall, in which case there is no issue. Another use case is having a Beacon outside the firewall with filtered information. Another use case is having a Beacon both inside and outside the firewall, which can communicate (the external Beacon will receive answers from the internal inquiry).

JL:  has Beacon implemented the approved GA4GH VRS model, the Variation Representation Specification?

LF: some VRS examples

JR: yes, in the Beacon response you have VRS, but annotation is not done yet.

SP: An important consideration left out of conversation has been pediatric cancer patients apart in clinical trials. Looking at how Beacon could be incorporated this type of work could be more effective than just working through hospitals (clinicians may not always be involved in the terms/structure of trials)

ZP: what is the limiting factor/what determines which data is available through Beacon?

JR: what data is available depends on the institute/research team/individuals that are deploying a Beacon (what they have access to). There is no limit but the current Beacon model is focused on clinical genomic diagnosis. The Beacon model also makes suggestions of ontologies to use

February 12, 2024 - CancerModels.org

Meeting Recording

Chairs: David Torrents

Regrets:  Romina Royo Garrido, Salva Capella-Gutierrez

Attendees [Name(Affiliation)]: Zina Perova, Daniel Barrowdale, Heather Creasy, Ian Fore, Grace Pendlebury, Amber Johns, Beth Pitel, Alex Wagner, Reger R. Mikael, Ophira Ginsburg, Nathan Boyd

Meeting Minutes

CancerModels.Org: An open global cancer research platform for patient derived cancer models (PDCM)

• Discrepancies b/w preclinical data + clinical outcomes, low approval of cancer drugs, drug resistance etc are some bottlenecks to combating increasing rates of cancer
• Genomics-based precision oncology aren’t always clinically actionable and have other limitations like not representing all aspects of tumor biology, among others  
• Functional precision oncology in comparison uses PDCMs, a field of research that is rapidly increasing
• PDCM fall into 3 categories: 1) cells + 2D models, 2) Patient derived organoids + other 3D models, and 3) patient derived xenografts
• CancerModels.org tries to incorporate the FAIR principles (Finable, Accessible Interoperable, Reusable) by using a Minimal information standard for PDCM, alongside query filters, providing harmonized metadata with a model characterization score (based on amount of information available), integrated sources (like cancer annotation resources), working to expand partnerships to enrich metadata, populations and annotations
• CancerModels.org is a database of different types of cancer models that exist that spans bodily systems, ethnicities and available genomic data
• Zina shared pre recorded demos that demonstrate how to use the resource (view meeting recording)

post presentation discussion

• There is integration of CancerModels.org with NCI Patient-Derived Models Repository (PDMR)
• Funding of the platform comes from NCI, looking to diversify
• Annotations found in the tool come from various sources, they use processed data and link out ot the raw data for those interested
• Submitter indicates where the raw data came from, do not pull all models from providers (e.g. dbGaP)
• Potential GA4GH connections - BAM Files already have GA4GH DRS IDs
• Can deposit metadata from submission process, if it's available from other sources already they don’t need providers to attain the data and can do it independently if possible

January 8, 2024 - CRDC + Velsera

Meeting Recording

Chairs: Salvador Capella-Gutierrez

Regrets: 

Attendees [Name(Affiliation)]: Surya Saha(Velera), Sharon Plon, Wes Goar, Ian Fore (NIH), Heather Creasy (NCI), Zina Perova (EMBL-EBI), Xuelu (Jeff)  Liu, Nathan Boyd, Rajat Kapoor, Daniel Barrowdale, David pot, Egonzalo, Pablo Martinez, Xenia Villalobos, Alberto Labarga, Romina Royo Garrido

Meeting Minutes

Surya’s Presentation: Multi-omics research on Cancer Genomics Cloud (CGC) driven by GA4GH DRS + National Cancer Institute (NCI) Cancer Data Aggregator (CDA)

• The NCI Cancer Research Data Commons (CRDC) comprises different data commons, including the Genomic Data Commons, Proteomic Data Commons, Imaging Data Commons plus others.
• The primary goal of the CDA team (NCI, ISB/GDIT, Velsera, Broad) is bringing all the data in CRDC into a single search interface
• Finding existing data isn’t straightforward and being able to do a data-first search across various  repos/file types etc would be more efficient (this is where GA4GH DRS comes in)
• The CDA now covers several data commons including: imaging, genomic data, proteomic data and cancer data services via a swagger API but a python interface also exists that is less technical (but still needs computational savvy users)
• CDA has open access metadata and users can search, but to access the files, a user requires credentials
• CDA meets GA4GH: once you access that primary dataframe, for each file in the collection, there is a GA4GH DRS URI which can be brought to cloud platforms or downloaded
• The CGC is powered by Velsera and part of the CRDC, it can be used to analyze NCI data
• Typical user flow: create a project, select data sets of interest, selects tools (like R), run analysis
• Within CGC, GA4GH DRS + NIH RAS Passports (one step removed from GA4GH) are being used + HL7 FHIR for clinical metadata
• Surya provides a use case using data from the genomics + proteomics data commons >  selected breast cancer as the tumor of interest > specified file types (VCF + RAW) > used MaxQuant to correlate the 2 data types > cleaned data + multiple test corrections
• Point of the use case was a proof of concept to show you can use CDA to look at subjects across multiple data nodes and bring them together in a meaningful way

Action Items

• Contact leads about planning the Cancer Community April Connect session. Idea to explore could be finding connections between GA4GH annotations tools (like VRS or VA) to be incorporated within the Genomics data commons (contact for this is Wes Goar)

December 11, 2023

Meeting Recording

Chairs: Salvador Capella Gutierrez, David Torrents

Regrets: 

Attendees [Name(Affiliation)]: Zina Perova, Rodrigo Martin, Nicolas Gaitan, Sharon Plon, Ian Fore, Opihra Ginsburg, Deborah Ritter, Xuelu (Jeff) Liu, Marc Arbones Alba, Alona Sosinsky, Wes Goar

Meeting Minutes

Actions

• Share points raised on VCF from this meeting with the VCF group
• Schedule 2024 presentations

• Zina (cancermodels.org)
• Wes (VRS)

• For all if of interest - https://commonsenseoncology.org/ 

Rodrigo Martin’s presentation on Oncoliner

• FAIR data principles guide the way we use genomic data in oncology > Find, Access, Interoperate, Reuse > the presentation focused on the last two principles in the context of somatic variant discovery
• Problem: different tools and analysis methods work differently, making interoperability challenging and subsequently making it hard to reuse data
• A study they conducted among 3 oncology centers found a significant lack of homogeneity along 2 metrics: performance heterogeneity score (measuring how much the performance (recall + precision) of centers differ, & gene discordance ratio (measuring how many genes are not detected by every center vs the total number of genes detected by any center) >> this means the 3 centers would be unable to work on a unified cohort of samples
• Rodrigo explained how the issue of lacking interoperability could be resolved with the Oncoliner tool > a comprehensive platform with 3 models for assessment, improvement & harmonization of variant calling.
• Users of Oncoliner can interact with a GUI (graphical user interface) to access the output of any of the 3 models to address their research needs
• The 3 models work like a chain > assessment feeds into improvement which feeds into harmonization, resulting in a final HTML output report
• The result of the study showed a decrease in discordant variants/genes/driver genes, meaning the 3 oncology centers were calling more similar variants > improving harmonization + average performance
• Oncoliner also offers addition softwares like VariantExtractor (which is readily available python package)
• Question raised in the chat: does the VariantExtractor map to the VRS/ClinGen work on canonical naming of events? Rodrigo said yes and that they’re working on moving towards VRS
• David Torrents mentioned that this is one of the first efforts to harmonize analysis across oncology centers > specifically in post processing of variant notation
• Question: to what extent can Oncoliner be applied to non whole genome datasets? you need a VCF you refer to as “truth” (this includes validated variance) and one referred to as “test” (this will generate your pipeline) so technically it can take any input in this format but Oncoliner wasn’t tested on this type of data, could introduce some problems  
• Oncoliner uses user defined benchmarks which makes it easier to play with for future uses

From Zoom Chat:

• I will be hosting a delegation at NCI. If anyone is interested its Prof Chris Booth and colleagues at Queen’s university (Canada) and Tata Memorial Cancer Center Mumbai India for the Common Sense Oncology movement/initiative https://commonsenseoncology.org/
• I am very happy to present our resource CancerModels.Org at one of the next meetings
• I would be happy to present on a tool we are developing using vrs and other normalization processes. We assess variants oncogenicity and therapeutic response

September 19, 2023: GA4GH Connect

See here for agenda, minutes and recording

August 17th, 2023: ELIXIR Cancer Group & GA4GH Cancer Community

Meeting Recording

Attendees: Salva, Angela Page, Ian Fore, Eivind Hovig, Daniel Barrowdale, Neerjah Skantharajah

Actions

• Salva to act as liaison between the two groups
• Neerjah to reach out to Daniel and Eivind to continue discussion and figure out logistics about potentially synchronizing/combining meetings (e.g. every two months we co-host a meeting)

June 12th, 2023

Meeting Recording

Chairs: David Torrents

Regrets: Ian Fore

Attendees [Name(Affiliation)]: Rodrigo Martin, Anne Deslattes Mays, Heather Creasey

Meeting Minutes

Group: NCI Centre for Global Health

Presenter: Dr. Ophira Ginsburg

Senior Advisor for Clinical Research Medical Oncologist

Cancer Geneticist

• There was always a global dimension to the NCI - studying rare cancers, fostering innovation and efficiency with less encumbrance by policy and practice conventions in any given country, increase contextual understanding in diverse populations and communities in US, validate and extend discoveries from US to other populations, contexts and health systems
• CGH 2021-2025 Strategic Plan

• 2011 - formal start of Centre for Global Health
• 2021 - 10th anniversary of CGH, with an updated 5 year strategic plan
• Read this Viewpoint for more information on the goals of CGH
• Focuses of strategic plan

• Research

• Key scientific issues in global cancer control

• Research Training
• Dissemination
• Partnerships

• Include international cancer control partnership
• International cancer screening

• 5 Thematic Research Areas (Does genomics fit? Kind of yes to all of them! But we haven’t built that out yet)

• Accelerate technology development for global cancer control
• Accelerate global cancer implementation science
• Understand and address global cancer health disparities

• Bridging “disparities” between domestic and global regions

• Increase support for cancer clinical trials in LMICs

• Need to expand genomic technology and understanding of the use of molecular tumor markers

• Increase understanding of cancer etiology and biology through global collaboration

• Aren’t doing a lot here but open to opportunities to convene, and matching people with partners in LMIC

• NCI Extramural global portfolio

• International awards more on the biology side, 7% towards cancer control
• In the portfolio over the last 10 years, from most granted to least → canada, uk, germany, china, australia etc.
• For LMICS → China, south africa, uganda, india, kenya

• NCI International Cancer Genomics Portfolio

• Researchers in NCI intramural divisions led 15 projects focus on cancer genomics/genetics in FY22 including 4 projects with collaborators in LMICs
• Extramural - 465 grants focused on genomics and included 94 grant with collaborators from LMIC
• 3 grants directly to LMICs (nigeria, tanzania, cape town)

• Research Training

• Improve global cancer research skills
• Support global cancer investigators
• Build global cancer research environments

• GlobTREC (Global Training for Research and Equity in Cancer)

• New mechanism for NCI

• Training Research topics in D43

• Cancer epidemiology
• Cancer genetics
• Bioinformatics
• Implementation research
• Oncology nursing
• Clinical research
• Mentored research opportunities

• Dissemination

• Promote the integration of current scientific knowledge into global cancer control policies and practice

•  Partnerships

• Represent NCI in many fora engaging key partners such as ASCO, help to coordinate NCI wide events like AoRTIC - has really upped its game in the last 10 years
• ICCP (International Cancer Control Partnership)

• NCI hosts the secretariat for the ICCP

• Program of Action for Cancer Therapy (PACT) - bolster, implement national cancer plans
• Key Cancer Insitutes

• Important scenes throughout national cancer plans
• Countries will now increasingly ask for more specific opportunities of engagement
• 11th annual symposium on global cancer research (ASGCR) - Closing the Research to Implementation Gap

• Opportunity here for GA4GH sessions?

• 2021 Global Oncology Survey of NCI Designated Cancer Centres

Q&A

Q: With your training, if you think about using the cloud (the great democratizer), have you offered training for computational knowledge to biologists to access international datasets and ask questions of those datasets?

• Have not done that, but we would be able to help connect the dots to such programming!

Q: Pediatric cancers?

• At CGH we don’t have in-house expertise, but we have a WHO collaborating centre. We are in the process of developing the third iteration of this to align better with the WHO three global cancer programs. Right now we’re only working in one space on cervical cancer, second is breast cancer and the third is childhood cancer. We just hosted three colleagues from IARC cancer surveillance. For clinical trials work we want to be as broad as we can be and we are doing a landscape analysis including a global survey.

Q: Other measurement technologies. We did the genome 25 years ago and I’ve moved on to proteomics, RNASeq etc. how are you expanding so we don’t take a purely genetics view.

• Trying to balance the resources we have but engaging with the communities that we have.

Is there focus on specifically, early, clinical, basic etc.

• Trying to align best with priorities as told to u by engagement through members states and WHO
• Affordable technologies

Are there plans or rules behind funding of generation of data beyond protection/

• Not sure we’ve funded anything that’s relevant to that…
• Working with DSI Africa through the research training focus area

Q: If you hear of LMICs or others that are looking to implement standards can they be forwarded to us?

May 8th, 2023

Meeting Recording 

Chairs: Ian Fore, David Torrents

Attendees [Name(Affiliation)]: Alona Sosinsky, Ophira Ginsburg, Sharon Plon

Meeting Minutes

Intro to Structural Variation Special Interest Group

• SIG coming out of Genomic Knowledge Standards (GKS) Work Stream
• Being done in the info model and vocab set up in Variation Representation Specification
• VRS in 2021: Alleles, Haplotypes, CNVs
• VRS 1.3: includes support for relative CNV change
• SV Goal #1:  want to add support for fusions and structural variations
• Rearrangement we want to support that are commonly seen in cancer → Translocation, deletion, duplication, inversion
• Work with CGC focuses on role of SVs in gene fusions
• Want to achieve integrated computable model that uses same components as fusion model
• SV Goal #2: Work across broader community to leverage interact and expertise
• Interest group will begin meeting regularly and will start to identify narrow scope of models around SVs
• Requirements or interests of others on this call?

Q&A

DT: Different communities will use annotation of SV. Cancer community is one (somatic), the rest of the communities (complex disease, rare disease) are still not there.

SP: Not sure I agree about the rare disease community not being there.  It gets discussed fairly frequently.  My question is about how you are working with the Cytogenetics nomenclature groups ICBN (sorry if that is the wrong acronym) currently used for Cytogenetics reports.

AW: How do you name the variant → there are nomenclatures around this. Metadata around how a variant was called is something we haven’t focused on in GKS but it’s a relevant concern. We’d like to get more members of LSG involved because of this.

AS: Is the goal to make this as part of VCF format?

AW: Daniel Cameron is leading the effort. He made the VCF 4.4 release. He recognizes the shortcomings of VCF today because of the historical nature that it was built. There are some complexities that make it hard to represent SV events at large in that format. The idea is to be compatible with VCF and leverage expertise from that team. There will be some differences in terms of what is expressed and how.

DT: to consider - How much do you want to break down the information? Should prepare a standard that can accommodate … for the future. The variant caller gives you the variant, variant interpreters need to determine fusion/gene, pathogenic/non-pathogenic. Challenging to reconcile

AW: In GKS there’s the Variant annotation group. They work to create and standardize these classification statements. Domains of Variation (categorical variation) modeling all the relevant assays and events that would be applicable to knowledge attached to domain - that spec is part of planned work for VR. Our goal is to use components of VRS to help in development of that other spec. E.g. adjacency range - should be permissible by using relevant entities in VRS.

DT: In terms of variant annotation - sometimes we just want to know the gene. Some variant interpretation programs that transform variants into a gene work well with point mutations, small deletions and insertions, but cannot cope with large deletion. Transform that into point mutation. Lose a lot of information…

SP: Relevant meeting with pathology community next week

DT: Beacon will use the annotation and depends on it. Are you starting from single point mutations?

MB: Problem with Beacon and SV is there are few users who provide good use cases. Problem is with what people understand and how they annotate variants internally. In principle we have good documents e.g. is you have CNVs this is how they should be queried. Looking for use cases and how to fulfill them.

JR: GIf Project idea with structural variants and beacon

SP: ClinGen has a very active .. committee that would be helpful on the germline side

DT: CNVs would include deletions to a certain level…snip arrays have been used to define the number of CNVs… on the Beacon side important to define if you want to find the break or if any position is altered. If I want to use Beacon to find any datasets, I’d like to find datasets with certain positions affected.

MB: What Beacon doesn’t do is support variant effects for structure variants. You can query by gene name, or precise bracket style queries where you can define the exact position.

DT: Would like to know whether a specific variation is there. A deletion you can capture by the position or by the extent…

EUCANCan

• Problems with building a benchmarking platform (EUCANCan WP 2)
• VCF file includes a lot of complexity and redundancy
• Table displays possible VCF representations for each variant type

Summary of problems

1. Twos ways of representing the same SVs
2. A lot of researchers do not understand bracket annotation
3. Two breakend notation entries could represent the same
4. …
5. …

VariantExtractor

• More or less fixes the problem but not a solution

Proposed solution: a focus on the breakends

• Easy integration with the representation of larger events (DC +1)
• Same structure for everything (including SNCs and indels) (DC +1)
• Atomic and unambiguous representation of breakends (DC +1)
• lack of differentiation between INS and DUP
• Possible collision with SNVs and indels existing schemas

SP:  Next week IARC/WHO is holding a meeting in Lyon to talk about the 6th edition of the Blue Book series that is used by pathologists world wide to define tumors and with the 5th edition also tumor predisposition syndromes.  They currently use HGVS and ICN nomenclature on all clinical reports and in the books - also available online.  It seems that there is no representation of that group on this committee and I would suggest perhaps asking IARC to name someone to this group.

Discussion

SP: Breakpoints are important but need to understand how to cluster them. Recommend framework and then give groups opportunity to how they will use the framework is an approach we should consider taking

DT: At the end you want to know which genes are fused, could be that the break is in kilobases of window. Perhaps seeing the functional consequences could be a second step.

JR: Questions to consider: What queries do you want to do? How do you want to solve these queries?

February 13th, 2023

Meeting Recording

Chairs: Ian Fore, David Torrents

Attendees [Name(Affiliation)]: Ian Green (SNOMED), Amber Johns (ICGC ARGO), Alona Sosinsky (Genomics England), Alex Wagner (VICC)  , Sharon Plon (ClinGen), Lincoln Stein (ICGC ARGO), Lindsay Smith (GA4GH), Peter Goodhand (GA4GH), Julia Ostmann (GA4GH), Anne Deslattes-Mays, Raffaella Casolino, Christine Wiliams (OICR),

Action:

• Email Lincoln if interested in joining working group
• Raffa to share different working groups
• Connect ideas:

• SV working group → looking for driver project engagement - read up on it in newsletter if interested!

Notes

Background

• Evolution of ICGC : ICGC 25K → PCAWG → ARGO

• Moving from scientific impact to health impact

• The new phase is looking at 100K cancer patients with high quality clinical data
• Looking at addressing outstanding questions
• Heavy focus on comprehensive clinical annotation
• Have 26 member programs involved across 13 countries
• Large donor commitment
• Clinical genomic platforms, regulatory clinical trial, global tumor consortia
• 5 expert working groups form foundation of lancet oncology project

Lancet Oncology Commission (Cancer Genomics - Precision Oncology)

Goal: Highlight and provide solutions for inequalities in two domains

• The patient journey from prevention to end of life
• Global cancer control and regional variation

Potential: to influence and change health policy, improve practice, enhance cancer control and there is no commercial influence

Vision:

• to analyze challenges of precision oncology from research and implementation perspective

• Want to provide recommendation to improve delivery of precision cancer care and harmonize data
• Steps of commission: Landscape, implementation, health systems, payers, and regulatory perspective -> ending with call to action

Background Article: https://www.thelancet.com/journals/lanonc/article/PIIS1470-2045(23)00007-4/fulltext

GA4GH Involvement

• Data process and control
• Recommendations on standards and harmonized methods for collecting, storing, transferring, accessing and analysing molecular and other health related data to improve responsible data sharing

• Monthly meetings
• Up to working group leads to determine meeting cadence
• Main purpose of meeting is not to create something new
• Working groups are still under development

Questions & Discussion

SP: Talked about clinical interactions and working with patients so what are the clinical goals of this effort?

RC: want to invite the main oncology societies. WG leads, half of them are American oncologists who work with patients. We were thinking about ICGC ARGO developing this project but realized we needed bigger involvement of people who work with this data every day.

DT: ICGC ARGO - the way it was designed was to generate an environment so groups who go do research focusing on this clinical aspect. This commission - how does it translate - is it to make recommendations to the groups in ARGO, generate lessons learned from those groups?

RC: It’s a health policy project. Ideally the recommendations will be used and adopted by countries. WHO knew about this project. They wanted to be involved.

AJ: ICGC is coordinating the commission effort. Many members of ICGC countries will be both contributing and implementing the recommendations. The commission project is much larger than ICGC ARGO.

DT: In Europe there is a big movement to generate a European cancer catalogue.

AV: We are looking for ways to make recommendations and guidelines as it pertains to medical oncology. Meeting in Heidelberg will be able to progress work here.

IF: Relation on cancer moonshot program?

RC: Mentioned moonshot because it was developed thanks to work on lancet oncology commission.

IF: Any connection and engagement to moonshot of the initiatives you’re talking about?

RC: One main focus for the new moonshot program is data. Update on two items: 1) data and 2)palliative care. Data - the work we would be doing with this commission could be integrated with the cancer moonshot. We have been invited to join a meeting in the US in April to discuss the issue of data.

AJ: Encouraged by lancet editorial team to work across commissions.

SP: Commission? Lancet is a journal so want to understand

LS: It’s a commissioned article.

AJ: Commission a group to identify and address needs in oncology.

SP: Sounds like there are .. that commission these things that have bought into this effort. Who is setting the goal?

RC: Commission is proposed by journal editors and then they decide which focus to select.

Next Steps

• Leading the discussion and writing of the technical requirements/ best practices for sequencing and sharing

January 9th , 2023

Meeting Recording

Attendees [Name(Affiliation)]:  

DT: Another group to invite is UNCan, a European group. Just presented at B1MG. Discussions in Europe are now focusing on alignment of infrastructure. Good opportunity for GA4GH to connect with these emerging groups.  

EOSC4Cancer Overview

• European-wide foundation to accelerate Data-Driven Cancer research
• Project is in the context of the EOSC (european open science cloud)
• want to prepare EOSC service with data tools and services for the cancer community
• Want to follow standards and recommendations from ICGC, ELIXIR, GA4GH etc.
• Project is organized around patient journey, starting from cancer prevention to diagnostics to different lines of treatment
• At all these stages different types of data will be collected/created and worked with (e.g. imaging, clinical data, genomics data etc.)
• Different analysis systems will also be required for these different data types
• WP4 (use cases) target scientific questions, while WP1, WP2, and WP3 do the technical work to address use cases
• WP1 - access, WP2 - harmonization, WP3 - analysis and utilization portals (considering federated analysis here), PW5 - community engagements
• Have 5 use cases so far focusing on large multi-omics, identification and prevention of cancer, clinical decision support systems (today’s focus)
• Squads are built around use cases
• Squad #3 focuses on connecting omics data from multiple sources to a clinical decision support system for precision treatment of metastatic CRC

ADM: Are pharma and biotech companies considered in this project?

RRG: Not involved.

ADM: Access to existing data like EGA, are you considering how to facilitate that with researchers who want sto connect with other data.

RRG: Trying to develop things with use cases in mind. In WP1 apart from identifying datasets with help of use cases, they also have a task for accessing this data. They will investigate ways to access these different datasets and harmonize ten. DUO is mentioned here to help with harmonization.

ADM: Have you considered moving to a just in time compute?

RRG: Not that I know of. Will bring this to the people in the developments team in WP1.

ADM: Moving towards the use cases to an operations and maintenance models? Long-term sustainability

RRG: In WP6 we will define a roadmap. We are very close to ELIXIR so we continue to follow their guidelines. In terms of infrastructure we want to rely on EOSC.

SP: Never hear about the blue book that defined the names of tumors. Those get updated every 5 years, so does GA4GH have some relationship with that effort.

IF: It’s relevant to a project like this.

SP: Add it to future agendas.

IF: What’s the overlap in terms of approach and technology with ELIXIR?

RRG: ***

Squad #3

XV: VIHO is the sponsor of investigative trial called basket of baskets. Conduct molecular profiling of our patients (7 different hospitals) and use ** decision system to break patients into arms of the trial. We are in charge of task 4.5 (connecting omics data from multiple sources…)

Will be working with data that is usually used in clinical trials

Biomedical question: how can we optimized the use of molecular data in the context of precision cancer medicine.

• How to better match profiles and investigate drug biomarkers
• Efficient allocation of patients to clinical trials
• Management of incidental findings

Objectives

To address the data infrastructure and format specification that a COSS can analyze patients tumor data and inform about biomarker-driven…

Data

• Sequencing files of patients
• Clinical data.electronic case report form (eCRF)
• Compatibility with other resources - link to WP3 (enabling interconnection between clinical trial databases and clinical decision support systems)

For COSS to provide accurate and actionable information access to knowledge bases is required (e.g. ClinVarr)

Will be working on data transfer agreements, server controlled access, restricted IP ranges, data governance and back-up policies

Biomarker driven clinical trial repository

• What field to include?
• Manual curation
• How to keep this up to date
• Ensure interoperability with bioinformatic pipelines

• In summary, looking at automating a common process for the capture, interpretation and reporting of next-gen sequencing to facilitate clinical decisions
• Interesting workshop would be ** to have more knowledge about CDSS portals

Q&A

AS: What is your source for curation?

XV: Now working with * they have a database they have been working with. We could not use that because it was not machine readable.

D: Molecular information doesn't follow standards so it’s impossible to extract. Have a data model that we want to use. Create an interface for clinicians wherein they must follow a specific taxonomy…

AS: Do you expect clinical trial PIs to use this interface? Will you get access to the clinical trial protocols.?

D: **collating information from centers that we are working with.

AS: Like crowdsourcing?

D: Yes.

L:S: Will you be recording the outcomes of the trials? Patient response status? So you’ll have a clinicals trial registry. For the discovery of biomarkers you’ll want to record the clinical trial results when they close. Access to protocols is hard to get. Are you ust working with academic clinical trials?

D: Need to check. Not all the information in the clinical trial database will be shared with partners outside of consortium. We are focus on molecular criteria.

IF: Questions regarding interoperability. Where can GA4GH standards help in terms of data sources you are needing to access. Coming at it from the use case point of view. The need to access data from outside your project v.s. Enabling the internal operation of your project. For example, connecting to EGA was mentioned. Would you want to be able to access data from other data bases? - CRDC, dbGap etc.

RRG: Yes, but not from the beginning. First we are aiming to build it to demonstrate it for specific use cases - they are all based on colorectal cancer. The idea is to incorporate more datasets that can solve the same question but we don’t know how far we will be able to go. Trying to get more national registries on board. I think it would be interesting to expand to include more datasets.

IF: You wish to connect to the cancer registry so you can find information on patients in your study.

RRG: In the context of cancer registries we are trying to link cancer registries and environmental data. Starting on cancer registry in Italy.

IF: EHRs - have you looked at FHIR?

RRG: Yes we do use them.

IF: LS had mentioned NCI match trial. There were common things that were done in that work? Would any connections there be helpful?

SP: they had something called matchbox where they take all of the biomarkers that target a treatment, including a definition of a loss assumption variant and they are able to feed data from * platform and say if there is a mtch. Done by a consulting group for NCI. This might be useful for what you are trying to do.

D: The information or the tool has to be publicly available.

SP: will send an email to Neerjah re: public availability of the software.

IF: There are also people at NCI that it would be good to learn from.

D: Many challenges exist in extracting data from the hospitals.

IF: Do you know if those hospitals are using FHIR on their EHRs.

September 12 , 2022

Cancer Community Leadership Meeting @ Plenary

• September 22nd Lunch (1:30pm -2:30pm CEST)

• We can’t bring food inside so there are two options:

• Try to get a table together on the terrace

• Is there seating?

• Finish lunch in 30 minutes, and have a 30 minute meeting inside

• Meeting Friday night followed by tapas
• Breakfast
• Poster break
• Lunch thing
• September 23rd Tapas on evening of 23rd (Lindsay can’t be there)

Options (send to Ian, David and Alex)

• September 21st breakfast

• We would need to find a venue

• Afternoon of 21st

• Would need choose between this and xSDO
• Alex might not be able to attend
• Lindsay won’t be there

• During any of the food breaks

• Won’t have a place to sit

• Concurrent to the COVID session (or another session that might be less valuable to this community)
• Friday night tapas

• Lindsay won’t be there and possibly many other people because it’s the end of the conference (but Ian, Alex and David will be there)

Cancer Community Form Responses

• What is the cancer problem we want to address collectively and how do we address it?
• Is this group for information collection or a working group?
• Information can feed into WSs or we can get information from GA4GH and implement that, but usually groups have determined how they are going to implement it
• Implementation forum should be an opportunity to have technical discussions but not conduct the work (not a hackathon)
• Through those discussions there maybe opportunities to improve the standards and that goes back into the WSs, and/or we might identify new gaps where new standards are needed and that would go up to the study group step
• AP: This group is the Cancer Implementation forum and the goal should be to:

• 1. Provide improvement suggestions to GA4GH standards
• 2. Identify areas where standards are needed
• 3.Give the implementers a space to learn from one another about how to implement standards

• IF: e.g. how to stand up a DRS Service - this isn’t cancer specific and shouldn’t be discussed in the Cancer Implementation forum
• Cancer Community should stay at high level and make sure each institution can run, for example, Melissa’s workflow
• A different group of people need to be have that “how to stand up DRS” conversation
• DT: Is there a need? If this is a group with theoretical discussions there will not be drive (no project, no money)
• IF: This is the group that identifies if there is a need or not. If a need is determined to be important enough, maybe organizations could pay their implementers…
• DT: European Commission is giving a lot of money towards research. When they generate a call they encourage using GA4GH standards, but traditionally they encourage using ELIXIR. If you want a European project you would say you’d use ELIXIR standards and software. Also a question of follow up and what they do if you don’t use standards. The European Commission has calls on cancer, another path could be to try and influence funding agencies because they define and push for standards. When we were defining EuCanCan GA4GH was not at the table so it was difficult for us to implement GA4GH standards at this stage. Later we became a Driver Project and are now trying to find our place in GA4GH. Project is about to end and I don’t see how we can use it to push GA4GH standards because in June 2023 the project ends.
• AP: Is this an issue of resources?
• DT: Many people need standards to move data and share data, but people don’t think about GA4GH as a place to go and find solutions. The Cancer ELIXIR group has the same problem. We could try to announce that you can go to GA4GH and find solutions for things that you want to propose in your product.
• AP: Meaning you can work with GA4GH to identify those solutions.
• DT: Can someone go to GA4GH and ask how to define standards for their problems? Needs come from money.  Not sure if DPs would be useful.
• IF: The things we build (infrastructures and tools) we need them to be sustainable.
• DT: The call where EuCanCan was generated we have a meeting with the officers, they are going to ask us how the project went, needs etc. to help them generate the next call. I will ask our officer whether someone from GA4GH could come to the meeting. They are the ones who put the money on the table. Sometimes infrastructure is the goal sometimes it is a means to an end
• Discussion for larger group meeting: What are the big cancer questions which will be enabled by interoperability? What are the datasets those groups have available? What is the potential from those datasets?
• What differentiates cancer from other groups is the analysis
• Don’t think we need to isolate the cancer
• Cancer is difficult because it is more fragmented

July 11 , 2022

Meeting recording (to be linked)

Attendees [Name(Affiliation)]:  

Plenary Workshop

Timing:

• September 22nd at 5:00pm CEST (with tapas after!)

Content:

• Presentation from data providers/users

• EOSE ELIXIR? Salva?
• Update from Melissa
• Update from Alex

• More content to be determined through survey results

ACTION: schedule another meeting for leadership once survey results are in

ACTION: Open up Cancer Community to all of GA4GH. Specific call out for data providers and data users.

ACTION: Develop Plenary Cancer Community Invitation List

ACTION: Check in with Justina re: having a meeting on September 22nd at 5pm. Where would it be held?

DT: Is it linked to plenary?

AP: Not directly. Separate registration.

LS: Whether there will be talks or workshops is up to us.

IF: There might be smaller WS meetings. Let’s develop this community and relaunch it. Having an in-person opportunity

DT: My concern is attendance. We should rebrand it and relaunch it. Mentioning it for 2 minutes during the plenary would help.

IF: Cancer was not a big use case for ELIXIR. Has that changed?

DT: Yes, it has. They are about to start a big project around generating tools for cancer. Think we should bring ELIXIR to the group. Put more effort into bringing in data producers (those that organize and ask for the money for analysis)

IF: EOSE Cancer overlaps with ELIXIR?

DT: Was asked from the commission to ELIXIR to generate this group.

IF: Are they using commercial clouds for EOSE?

DT: No, not open in Europe to commercial clouds. But sure it will be considered somewhere in the project. This project should be presented on.

IF: Focus on data providers is good. If we have data, there’s something to work with. Fits with use case: have a pancreatic data set I want to access in the US. Is there a set I can analyze that with in Europe or JApan/ Who would be the data provider? How would we get to the table for discussion?

DT: Data providers and data users should be involved. These people are usually not on these calls.

IF: There’s a difference between data providers and managers. Scientists who do the research are required to deposit their data somewhere. For us it might be in cancer data commons or dpgap. Do you mean the people who run dbgap, EGS?

DT: People that sequence the patients. Their name is on the dpbgap entry. The PIs that decide which data is produced and which data is shared at the end. This community is more on the clinical side. Common for PIs in clinical studies to only use their own data. They should be aware that they can access other datasets. Real research data that is produced every day is escaping those channels because PIs are not in our conversations.

AP: How do we incentivize these people?

DT: They can’t think this is just a technical call. Have to understand what’s happening in the call so if it’s too technical we will lose them. Show that they could do more if they could access other datasets.

IF: We’ve not even invited them yet. The PIs that are producing the data - the incentive would be that there is other data they don’t know about that they can access. They are wearing two hats. They are a data provider and they are also a data user of another dataset. What would be the scientific cancer questions that we can answer if we had this dataset and this dataset?

IF: David is talking about a group that we have not yet engaged.

DT: People we want to target are medical doctors that do research. We should plan carefully how to engage them.

IF: Would be good to identify actual people.

DT: There’s a general impression that things get too technical.

IF: What Melissa is doing ought to be front and center. How do we do Melissa’s use case? At some point we have to get mechanical with it. Analyze the data and use the tools.

AP: One of the things we’ve been thinking about through Strategic Refresh, this group is good for identifying use cases and identifying what’s missing. The engineering part should take place in FASP.

IF: Don’t think we get close to engineering.

DT: Maybe we don’t need those data producers. But trying to generate a more general forum. Some parts of the talks should be more open. Think it’s important to show that data sharing in cancer is beneficial.

IF: The motivation/vision you have is about this scientist who’s just looking at their own data. Is this how we want to do this? Showing them stuff that already exists so that they can use other people’s data. That’s a different question than what do we need to build?

AP :Would it make more sense for us to go to their meetings, rather than inviting them to our community?

IF: Yes we should go to them. Maybe go to their lab. Maybe working with two labs about bringing data together.

AP :There’s a marketing aspect of this. It might make sense to do some targeted specific work with people we know that are on the fence. They’re close to making the transition into this data sharing space. Once you have those successes (e.g. Melissa Cline) people can see that and come. But blanket inviting everybody to our meeting might be (overwhelming?)

DT: We always discuss standards and sharing at EUCanCan assuming that people are already sharing. But people don’t always share because they don’t need to.

AP: Identifying key goals that we can market to the community through the survey.

IF: Feel like we should check in with Alex and Melissa’s use case. Who would be the European PIs she’s working with around VUS? I used WES and DRS to do Melissa’s use case. If there’s a European dataset we can do it with.. There are ELIXIR WES services.

DT: Would try to expand Melissa’s use case to other datasets. Also, we could incorporate use cases from this EOSE Cancer ELIXIR project. We can ask them whether we can discuss those use cases here.

IF: Would call them a piece of science.

AP: Invite the group and have an in-person cancer community meeting. Something more structured could be a brainstorming session around a specific problem. Or look at the survey results and figure out the plan around those.

IF: Could we get the type of people David is talking about to talk about the science they want to do? And make EOSE and ELIXIR the focus of it.

DT: Salva is the perfect person to explain this. There is a scheduling conflict though with ECCB.

AP: potential conflict for cancer members around ECCB and this workshop.

DT: Not sure whether the workshops can be hybrid logistically.

IF: Can we squeeze something on the 22 or 23?

LS: Could be a lunch or dinner meeting.

AP: Could also potentially be a Saturday meeting.

Timing: September 22nd “Dinner” Meeting (after 5:00pm CEST)

June 13, 2022

Meeting recording

Attendees [Name(Affiliation)]:  David Torrents (BSC), Ian Fore (NCI), Amanda Spurdle, Sharon Plon, Montse Puiggros, Alex Wagner, Ben Wilfond

Minutes

GA4GH 10th Plenary

• GA4GH Plenary taking place in Barcelona September 22nd and 23rd. Day 1 will focus on working groups. Day 2 will focus on implementations (Genomics in Health Implementation Forum = GHIF)
• GHIF is the implementation arm of GA4GH. Started from the National Initiatives (NI)  meetings. NI is now a subcommunity of GHIF. Eventually we see the Cancer Community being a subcommunity under GHIF.  We see these communities as sitting along the Driver Projects model since that isn’t as scalable.
• AS: Saw GHIF as a separate since their meetings are separate. Thinking of it as an arm is surprising. GHIF needs to be integrated as GA4GH.
• LS: Yes, we don’t want to hold it as a separate meeting that’s why Day 2 of plenary will essentially be GHIF Day.
• AS: GHIF tells me why we are doing all of this. Allowing for something like registration per day…
• SP :Are these two groups are meeting in the same city on the same day?
• LS: That’s how it use to be. We use to have GA4GH Plenary and then a GHIF breakout or separate meeting. So this is a reason we now want to merge it.
• DT: Are there any highlights on specific standards that we should cover from the Cancer Community (for Day 1)?
• IF: We shouldn’t focus on standards at all, we should focus on the science..
• AS: I need the standards dumbed down. I think that’s why there is place for a three day meeting. You do need more info for people who need to implement them.
• SP: Cancer is full of complex rearrangement. How do you explain that? Is it a standard?
• AS: If you’re trying to engage people who are seeing the results of the implementation and wanting to promote the standards, they need to understand it well enough to sell it to their local funding bodies. Then there are people that need to help with working on it.
• LS: Did you enjoy past GHIF meetings/
• AS: Yes, it was accessible.
• LS: That is what we want to do with the second day of plenary. Just trying to merge the two.
• …
• IF: Funding the standards efforts.
• AS: If you can show that data sharing is critical to your efforts, then we can actually use that as an example for federal funding. If we can show that we improved diagnoses then that improves chances for funding.
• SP: Having examples, not simply you folks using your own standard, it about finding examples of outside people who have used your examples. E.g. ClinGen created an allele registry then Civic went and linked to it.
• AS: We used them for sharing.
• SP: If you’re thinking about inviting people, think about whether there are examples of people who took your standards and applied them in your system who have never heard of GHIF/GA4GH. Try to fin groups independent of GHIF.

CHARM Study

Ben Wilfond: Co-PI of CHARM study. The CHARM study is investigating ways to increase access to genetic testing for those at risk of hereditary cancer in low income, low literacy and minority population. During the CHARM study we also looked into methods of authorship tracking.

Our study has a series of team leads. They lead different aspects of the study. The team leads would identify a paper and create a concept sheet. The concept sheet would be used to identify scope of paper, identify team members, audience, and timeline, identifying resources. Allows PI to prioritize analysis and balance work across team. Realized that there were issues we were facing. How do we promote equity and inclusion in publications? The ICJME guidelines retrospectively tell us who should be an author during submission, but doesn't tell us what the roles should be from the onset. These perspectives are based on discipline, institution, and career stage. Equity = who should be leadin the paper, inclusion = who should be co-authors. The process that CHARm used to determine authorship is the most important thing. Had a team meeting, breakout sessions to discuss authorship (perspectives, challenges, thoughts on solutions), larger report out meeting, drafted guidelines based on outcomes. Shared guidelines with team, finalized and implemented it.

The guidelines had three components

1. Identify author roles and responsibilities

1. Team leads (identified early on) → lead authors → primary writing group → secondary writing group
2. Being led by lead authors, pwg pretty involved, but secondary group is peripherally involved
3. CHARM authorship values = equity, inclusion, efficiency
4. Principles to promote…

1. Equity: team leads should distribute lead authorship to team members are guiding key aspect of the study. Team leads should encourage junior team members to be first authors with mentorship from a senior author.
2. Inclusion: who else other than the lead authors should be included? Who has been involved with related aspects of the study and would be interested in manuscript development?
3. Efficiency: be realistic about the capacity of the team.

2. Develop values and approach

3. Implementation

SP: Is the data freely available? Same people get asked over and over. Do hte leaders reach out to people who haven’t self-nominated.
BW: Yes, that’s the intention with using smart sheet. Team leads tried to identify who was missing. We do not rely on self-nomination
AS: For Breast Cancer Association there are a lot more authors. Not sure we always reach ICJME guidlines. When people have to tick a box, there is a difference in people’s interpretation of how they contributed. Practically speaking, when dealing with a lot fo authors have you come into arrangements for how you can track affiliations, acknowledgements? We’ve done that within ***** we keep a database in endnote of all those different things, circulate it and expect people to circulate it. More time spent on our papers and not who’s going to write it. Top journals are actually ok for you to enter top 10 authors and leave the hundreds others.
IF: Went through this with the GA4GH papers. At some point, we were self-reporting our contribution through dropdowns.
AS: Think our role should be to get journals to follow the same process.
IF: How do we build trust? Talking about who’s going to be responsible for a certain area of a paper.
BW: Anybody can meeting the journal guidelines simply by reading the paper in a thoughtful way. It’s more who do you ask to read it and send it out to. Within a consortium, who are authored papers vs. who are non-authored contributors? Can make that decision in a variety of ways. We do use smart sheets to track actual authors themselves. It never occurred to us that we would do our process based on institution because people contributed in different ways. Didn’t make sense to break it up based on institution due to the differences in contribution levels regardless of institution. Having an institutional data doesn’t seem a fair way…
AS: Missed out a few details. They would be core people who were involved in running the database.

April 11, 2022 [CANCELED]

• Meeting Recording (to be linked)

Chair(s): David Torrents (BSC), Ian Fore (NCI)

Attendees [Name(Affiliation)]:

Minutes:

March 14, 2022

• Meeting Recording

Chair(s): David Torrents (BSC), Ian Fore (NCI)

Attendees [Name(Affiliation)]: Michela Riba, Marco Morelli, Giovanni Tonon, Daryl Waggot, Montse Puiggros, Gabriele Bucci, Teri Manolio (NHGRI), Ariana Gonzalez, Alex Wagner, Ivana Cecic, Alfonso Valencia, Sharon Plon

Minutes

An Overview of B1MG and Cancer WP9

GT: 1+MG Project, goal to collect 1 million genomes by the end of 2022. This is quite ambitious. Quite a few difficulties. Most countries are proceeding with panels and have decided not to include this kind of tools and platforms inside 1+MG. Panels are very heterogeneous so considered to be difficult to include them into comprehensive assessment. EU countries agreed to cooperate in linking genomic data across borders. 22 countries have now signed and 6 are observers. Working group 2 focuses on ethical, legal and social issues (Marco Morelli) difficult because of the heterogeneity across Europe. Working group 3 Standards and quality guidelines, probably the most relevant to this group.

Cancer Use Cases

GB: Case 1: uncommon mutation in target cancer gene: rare somatic mutation in EGFR gene in NSCLC cancer patient

• Questions that might be asked include:

• Has the mutation previously been observed?
• How many cases have been treated with the SOC…?

• Outputs will typically include the number of patients fulfilling the query, survival rates…
• Would be useful to have a quick list of mutation and approved drugs

Case 2: acquired resistance mutation occurring in tumor sample or liquid biopsy during a targeted therapy

• Patient with a peculiar mutation, not a single nucleotide. This patient has undergone SOC but doesn’t respond. Want to see whether a new mutation arises.

Case 3: Consistently compare TMB across studies

• TMB is often calculated in the context of a specific study cohort. Each patient tumor differs in terms of mutation burden*
• Can use aggregated data from cBioportal to put together a cohort of patients.

Cancer cases you have a range of possible choices of what kind of aspect of a cancer you’d like to understand from single point mutation to timeline of events which happen during the lifetime of the patient. This is why cancer is also difficult to include across standards.

GT: One of the challenges that 1+MG is facing is modeled against the rare disease.  There are some ocuntries in Europe which do WGS for diagnostic purposes.

SP: There’s much more interest in RNA DNA panels and some interest in clinical genomes. Many fewer large scale projects doing genomes. Does the project include transcriptome data?

GT: At the present time, no.

Cancer POC

MM: Interplay between infrastructure, ELSI and need from use cases

POC is focused on use cases from rare disease. Have only two countries which exchange synthetic data. The standards in Rare Disease PoC was actually from GA4GH (Beacon, DUO, SAM/BAM/CRAM etc.). Here is a diagram of what a researcher might want to do. First, search for data e.g. what patients have a particualr genetic trait (done with Beacon). You might also want to query ..Then you have a couple of infrstrctures to store the data and manage identity of people who access the data. Then you have a part where you want to do quick analysis on the data you have carried out. Done through GPAP platform.

3 questions:

Re-analytics of raw data is not always a requirement. Need something that can do analysis on the fly. We broke down the diagram to adapt it for cancer. The genetic query is not doable with Beacon v1, but there is  beacon v2 hat is about to come out and it may be able to address questions that come out. The search on clinical data must be changed because it cannot handle the complexity of cancer data. The analysis should be updated to fit cancer.

Minimal Data Set for Cancer

MR: Would like to propose a data model fitted for cancer genomics in Europe. Minimal for easiness of use but comprehensive for both clinicians and researchers. Capturing longitudinal and complex aspects of cancer. Considering data on treatments and their outcomes…

Took into consideration existing data models mCODE, OSIRIS, ICGC-ARGO etc.)

We compared the three of them and found out they share certain aspects.

The process towards a minimal set for cancer started with dialogue with clinicians and researchers. Afterwards we created a definition of a set of tables, sub tables and items. It underwent revisions by the WG9 NMG community. Minimal set consists of tables focused on both observations and longitudinal aspects. Consists of 13 tables where a definition is provided, examples, sources and whether it is mandatory, recommended or optional.

IF: The observation about not needing to access raw data is common. However, the calls are often against different reference genomes and from different calling pipelines. How does that affect whether access to raw data is needed? What might it say about harmonizing variant calling?

That question intended mainly for 1MG - but is obviously applicable more generally.

MM: Just to be clear: the access to raw data, if privacy allows, MUST be always an option.

MM: There are couple weaknesses with reanalysis of the data. Firstly, it is double the time, and it’s a bit static in the sense that pipeline analysis often needs the usage of * tools. As a proposal could see requiring a set of data standards and then receiving the raw data and call from submitters, and give the user the opportunity to redo analysis if they want. But the query would be done on the call made by the submitter.

DT: One way of population the 1+MG project, Spain has parallel activities. The countries are starting open calls to generate data. We are in a project where we probably will start at the end of the year sequencing and analyzing genomes. Would be good to have a small table.  Want to filter for the clinical, technical sample issues. Not sure whether there is time for this.

MM: Don’t have the right answer. I’ve had experience trying to harmonize pipelines across Italian centres. Tools were the same but results were very different. At the level of raw VCF, as it comes out from the caller, if the caller is ** they start diverging a lot. You can ask people to document how they go about the process. If you want to reanalyze you should do it on the platform where the data is stored.

DT: We should have a very basic recommendation for those genomes that will end up in 1+MG project.

MM: There was a paper that suggested 5 scores on the data. If your scores are nto good, we won’t accept your submission, but if they are good we don’t ask you how you obtained your data.

DT: Quite an old paper and may need a bit of an update.

SF: If you’re ding exomes vs gnomes, many of the drivers in cancers are at a minimal levels because of tumor heterogeneity. Are those tables available?

Caution that these are not validated by !+MG. It’s an ongoing effort that is soon to be validated so you can use it with a grain of salt.

SF: Just had cancer disease in table 5. Which ontology are you using?

We stay away from ontology and only specify which fields should be filled. WG2 focuses on ontology.

IF: Going back to use of GA4HG standards. Ad something on their called ** storage. Seemed to be touching upon questions asked by the DRS standard. Oriented around cloud storage. Doesn’t mean it can’t be relevant to storage outside of the cloud. What is FEGA storage and is standardized access in the way DRS would it useful to this?

MM: The idea is that you have EGA as a database for sensitive data. FEGA = federated EGA. data generated cannot leave the country. The idea is to build a node of EGA in your country so the data created in your country does not leave. Then we federate these nodes in a type of EGA network. This is just for storing. Easier said than done.

IF: I think this is an important area to see how we might look at it differently. Ties in with the use case that I’ve brought to this group. The use case was about finding pancreatic cancer data set in EGA and dbGAP and wanting to federate analysis across those two. Dbgap is looking at drs storage. EGA data was generated in US. Highlights an area where these standards can be useful to us.

MM: The ELIXIR consortium provided an identitymanagement layer.

IF: Would having them begin to merge be a useful thing (dbgap and EGA)?

SCG: EGA is working on passports and visa as long as dbgap is doing the same.

IF

February 14, 2022

• Meeting Recording

Chair(s): David Torrents (BSC), Ian Fore (NCI)

Attendees [Name(Affiliation)]: Lincoln Stein, Alona Sosinksky, Angela Page, Amber Johns, Anne Deslattes Mays (NIH/NICHD), Montse Puiggros, Sharon Plon, Alex Wagner, Malachi Griffith, Ivana Cecic, Melissa Cline. Daryl Waggot

Minutes

IF: For future review, the group’s charter: https://docs.google.com/document/d/1tqqnXri2QjaHiOK91CxUWuCntCfDyrtlVF8xmzsEDek/edit?usp=sharing

AP: GHIF was created as a way to share resources and to have GA4GH standards drive data sharing. Starting to see the GHIF community as a use case generator for GA4GH. In that process, we are starting to have domain-specific communities like this one for cancer. These groups are for generating real word use cases where we can use or create GA4GH standards to address those needs.

IF: Lincoln gave a presentation before at the Human tumor analysis network. The topic of integration across initiatives seemed relevant to this group.

LS: Going to focus on technical integration challenges today.

PCAWG Project

Completed and published in a series of 6 nature and companion papers. Full-scale analysis of 2658 whole cancer genomes. First time that the entire cancer genome had been analyzed at such a scale. It provided 38 different tumor types. Based on the international cancer genome consortium's collection of 23K cancer genomes. TCGA project from the US is essentially half of ICGC. ICGC was founded in 2011 and grew rapidly. Reason we did PCAWG, before this only the cancer exome had been explored very well. It is a small slice of the gneome (1%). A large amount of the genome had note been explored (introns, regulatory regions, non-coding RNA genes etc.) I was involved in the data management and processing of this project. Marshalled whole genome tumor and normal pairs from the ICGC project. Uniformly reanalyzed each read set from alignment to variant calling. Organized ~700 researchers into a set of working groups focusing on different aspects of tumor biology.

**

Ended up with 2834 cancer genomes from 48 projects. At the beginning it was important to harmonize the variant calls. Challenges included repetitive regions, low coverage, regions of high GC content and multiple published variant calling methods each claiming to be the most accurate but with high discordance among them. Each center did their own variant calling and among them there were 8-9 different methods used. We identified a set of 50 core benchmark tumors and set the reads out to each of the groups and asked them to run their variant caller on this same set. They sent their calls back to the central coordinating committee and there was poor consensus among variant callers. Some callers are promiscuous and others are very conservative. We took each of the ten callers, took all the variants that they had called, did a sampling of them and had them experimentally validated by resequencing at two different laboratories in order to see the true baseline. 3 callers came up as having the highest sensitivity and precision (Sanger, DKFZ, MuTect2). We created an ensemble of callers based on those three. We came up with a similar process for structural variations, CNVs and the whole RNA sequencing pipeline. The workflows do some pre-processing, do the variant callers, ** and we end up with the final set of variant calls. The workflows use the dockstore system for management. We set off to run this pipeline on each of the whole genomes. We had a series of groups volunteer their informatics capacity and we ran the workflows together on a whole series of compute centers. Some of them were cloud-based, others were bare metal, and others were private clouds running on open stacks. We ran the pipelines as docker images. To manage exchange, **

We developed a final set of 2583 whitelisted tumors because some didn’t pass QC steps. Ended up with 45M SNVs, 2M Indels and 288K SVs.

**

Found 705 SNV driver hits spanning 179 genomics elements. Of these, 602 were in protein coding genes. We did driver calling for SVs. Found 53 significantly-recurrent chromosomal rearrangements, all involving known cancer genes. Interesting finding was that there is a order of magnitude * coding to non-coding drivers. 5-10x more coding than non-coding drivers. We found alterations in the expression patterns of multiple known cancer-coding genes. When we combine coding and non-coding drivers we can find at least one driver event in 95% of the cases. Significant increase than if we used exome coding alone. With whole gnomes we were also able to date the mutation in the cancers by looking at the clonal representation of the driver and passenger point mutations. Identify late clonal mutations, and distinguish between sub-clonal and clonal mutations. By using the clock like mutational signature 1  we were able to date the origin of the tumors. One lesson we learned was that each cancer type has a unique trajectory of early and late driver mutations.

The group was also able to show the first genetic alterations in a tumor can occur decades before diagnosis. In some cases, such as in Melanoma, it can happen 40-50 years before the patient presents.

Passenger Mutation Patterns Predict Cell of Origin

We asked whether you can predict the cell of origin of a tumor based on the pattern of its passenger mutation. The pattern of passenger mutations is very different from one tumor to another and is based on mutational rates.* Put together results into a website (deeptumor.oicr.on.ca) upload VCF file, click analyze and then it makes a prediction amongst 28 different tumors of origin.

Wish List that GA4GH can help with:

• Pre-harmonized descriptions of tumor types and histopathology
• A set of turnkey pipelines incorporating best practices for cancer variant calling

• In some extent we’ve achieved this already through dockstore
• We’ve been rebenchmakring them because new versions of the software have come out

• Integration of workflow system into provenance tracking

• A future version of WES

• A database management system for all authors and affiliations

• Maybe integrated into provenance tracking system
• AW: #4 is such a big deal for consortia efforts; closest we have come IMO is ORCID.
• AJ: Agree Alex- currently dealing with this in ICGC ARGO. Admin nightmare..
• SP: Katrina Goddard when at Kaiser worked out such a system for number 4 for their clinical sequencing consortium
• ADM: +1 Re: ORCID and there provenance could be automatically calculated - captured with a combination of GitHub Actions I think

SF: Which groups represented in this meeting were involved in PCAWG? Trying to understand how much of this experience is shared between us. Please indicate here in chat if you or your project was involved. ClinGen wasn’t involved.

LS: We are actively incorporating ORCID into dockstore and overture for ICGC ARGO we’re using via passport ORCID identifiers.

SP: Your comment about authorship tracking, Katrina Godford actually developed a very sophisticated way of tracking contributors. Thought they were going to write it up but it may be something portable.

LS: would love to have that

SP: Can email her about that. I’m less optimistic about your first point re: tumor nomenclature. They still don’t map very well together and whether a more computable framework is needed. Has come up in clinGen efforts. Your project worked on the most common tumors but as we start to dive into rarer ones there is a missing tumor ontology that’s mappable across this.

IF: Is there already a group formed around this?

SP: The group got started from a series of conversation clinical sequencing consortia and clingen. There’s a subgroup that’s been on an email thread. Wil;l think about whether this is something that GA4GH can help out with.

AW: Mondo is working on this exact problem. nicole* wanted to submit re: Mondo cancer classification. Seems like what we’re talking about here.

IF: In your project did you do that centrally with the same group of people or did you do it in a distributed fashion?

LS: the effort was distributed but it was the same core group of people. Two pathologists doing brain bases, a urinary pathologists, GI pathologists all located in different parts of the world.

IF: By the magic of zoom they could show each other slides

LS: Would’ve been easies with zoom but we weren’t using Zoom then (6-7 years ago). We were going off text-descriptions of the cases by the source pathologist. Not as rigorous as it should’ve been

Participating in the NC3 Barta challenge. They used OMAP terms. The teams that are performing the best are those that map to hierarchical ontologies. That’s the benefit of not doing the ICD10.

LS: Yes, like Oncotree which is hierarchical.

This is a shared experience. Lincoln had a wishlist. Does it seem like this issue is something for this group to put on it’s agenda for future consideration?

ADM: Perhaps adjusting the list for post-covid times.

January 10, 2022

• Meeting Recording

Chair(s): David Torrents (BSC), Ian Fore (NCI)

Attendees [Name(Affiliation)]: Susan Fairley (GA4GH), Sharon Plon, Alex Wagner (Nationwide), Kyster Nanan (OMPRN), Jordi Rambla (EUCANcan), Michael Baudis, Alona Sosinsky, Babita Singh, Lincoln Stein, Laureen Fromont

Minutes:

DT: Invite Jovani (?) from the cancer group in B1MG. I will check with them.

IF: Jordi is leading a work package in EUCANcan.

EUCANCan

JR: Focus on the usage of GA4GH standards in EUCANcan. Create a network of federated data, both technically and legally. The idea is to demonstrate or show that by analyzing data in a uniform way and sharing it with legal aspects addressed, you can use it for solutions. Benefit end patients.

4 work packages. Clinical and genomic data collected and shared via nodes. Data should be shared and maintained in the long term and be able to be reanalyzed. Can generate a network of data to share in a uniform way and keeping it available for long term usage. Focused in making this happen.GA4GH standards in WP5 (standards for clinical data, phenopackets) and WP3 (infrastructure, cloud standards)

SP: Are these research genomics, or do they include clinical panel type projects?

DT: In principle we are considering whole genomes, but the concepts could be applied to exomes and gene panels.

SF: For WP2, sounds like an area that variant annotation work might be applicable?

DT: Idea was not to reach interpretation, but more identification. Project includes a report for the clinic, could find space for interpretation, not a priority.

SF: VA is having discussion about unambiguous representation of variants. Very detailed discussions, but don’t often hear European voices on there. [Action: connect European projects with VA/VR work - variant standards are a shared topic of interest]

JR: WP3 - to define a circuit for long term raw data and metadata storage and flow, as well as access protocols. Metadata, phenotypic and raw or genomic data. With the metadata, we have no standards in GA4GH that applies to what we need, using INSDC model.  Using Data Use Ontology (DUO) and OIDC (open id connect) for authentication of users. Passports doesn’t make sense for us yet, but base for passports. Does not include phenoclinic data. Raw data is Crypt4GH format.

Looking at phenopackets for phenotypic data. Piloting with submission, encrypt and validate with a tool provided by Peter Robinson, made accessible for data findability.

Using passports for data access.

What happens when data couldn’t be send to EGA and needs to stay locally? One use using Federated EGA, different countries aggregate data locally. Hospital doesn’t have the capability to store data locally, so they send to their national centers. In other cases, we expect data goes out from the institution itself, “EGA Community Platform”. Fostering a modular approach: manage the data, mechanism for allowing access, and then discovery and how to expose data. In some cases the institutions have something already, maybe a solution for browsing or storing data. Only thing we ask is to reuse what they have. If they have something but it’s not easy to integrate, then we ask them to extend their solution, like putting a Beacon on top of the solution.

Let’s assume you use cBioPortal, can analyze and visualize but cannot integrate. In this case we would say Beacon on top. You also don’t have management of the files, so need GA4GH standards applied.

You have a patient visiting clinician and that clinician is sending information to the lab. Lab comes back with a report and maybe a vcf. Sometimes the repository is internal or external. Data is in this cycle. Then, researchers dig into the EHR and generate one database with the data they need for a study.

Problem of moving from distributed process to something much more organized. Have to get the data in using security, need to store is somewhere, need to analyze the data, and need to index that to make is discoverable. Can pick the pieces you need.

WP4 - cloud infrastructure

EUCANcan contributed requirements to Beacon v2. Strongly based on the scenarios shared here - eg. being a clinician needing to do analysis and share information to third parties.

We are EUCAN image using cancer images associated with phenotypic and genomic data. Seeing if we can reuse the model as a phenopacket. Should be able to get information from oncology domain and share in a portal. Want people to use the same model as much as possible, using the same ontology, dictionary terms, etc.

DT: Can you share the slides with me?

JR: Yes

SF: You mentioned nextflow - Jeremy has been doing work to making WES compatible with nextflow. Interested in your point about defining controls, sounds like something to feed to cohort representation. Also interested in the idea of the community platform, if someone was interested in finding about more about this, is there somewhere I can point them to?

JR: Not yet, more of a brand. Newbie friendly “how to” that shows how components can fit together. Working on a document about this.

IF: Is the community platform for filling needs shown on the first slide? Could you comment on how this works from a EUCANcan wide perspective, is this platform used in Canada too?

LC: The canadian component of EUCANcan is the Overture platform for data sharing and management, which is also used by ICGC. Happy to see it fit into this larger vision. Canadian researchers generally submit genomic data to EGA rather than dpGap in the US. Given the GDPR restrictions which are inhibiting access to the US, Canada will become more closely associated with European efforts. Discussion about creating a sequencing archive, need to discuss if time is right for Genome Canada to fund an EGA node.

IF: A lot of US data goes to EGA as well. What’s your sense of the the US content in EGA?

JR: 15-20% overall I think, including both UK and US.

SP: How much clinical data do you actually have? Treatment, outcomes?

LS: EUCANcan has been a great success at the technical level, but when it comes to our clinical partners committing to share real patient data, there is a lot of caution on the part of hospital admin in making data sharing happen, there is neither the incentives or the risk reduction needed to allow hospitals and healthcare providers to actually make it happen. We have servers running, but doing demos with synthetic data right now. One of the issues WP5 is addressed but will require more than just policy papers to break through this final barrier.

SP: In the US, the Childhood Cancer Data Initiative (CCDI) has patients already in a trial, funded by the government but designed to get all of the rich data in. Have the advantage that patients consented to be in a cooperative trial. One way to overcome this is to start with patients who were in a treatment trial who we already have genomes for.

IF: Jumping point for something to follow up on in a future call. CCDI is closely allied with the CRDC, we could probably bring in some people from that to talk about it.

LF: In EUCAn image, we have the hope that some issues with sharing reluctance will be resolved. EUCANimage has some use cases to use AI tools in order to predict outcomes depending on diagnosis and treatment. Some clinical partners are willing to share as much data as they can. The number of variables that are available depend on the use cases, but there is quite extensive information about the different kinds of treatments. I think the clinical partners are still talking about the clinical parameters that are to be shared.

Here are the fields we are considering in EUCANCan that derive from ARGO: https://docs.google.com/spreadsheets/d/1Ph5tBD3tG0h1bOpv5sj0KLalx75_ZA-V/edit?usp=sharing&ouid=112786736772019968234&rtpof=true&sd=true

December 13, 2021

• Meeting Recording

Chair(s): David Torrents (BSC), Ian Fore (NCI)

Attendees [Name(Affiliation)]: Lindsay Smith (GA4GH), Neerjah Skantharajah (GA4GH), Sharon Plon, Amanda Spurdle, Alona Sosinsky, Anne Deslattes Mays, Melissa Cline, Peter Goodhand, Obi Griffith

Minutes:

IF: One of our key interests is the context of sharing childhood cancer data. That's the driving use case as to why different groups might want to interoperate with Genomics england.

AS: I am the Scientific director for Cancer Genomics England. I will first introduce what we do and what kind of data we hold/is available. We work closely with NHS England. Consent patients, collect samples, DNA gets extracted at genomic labs and get sent to illumina labs for sequencing. We outsource sequencing. In house we do variant calling and summary of results get sent back to the genomic labs to be viewed by clinical scientists. They are then presented at the Genomic Tumor Advisory Board. They make decisions about patient management. OpenCGA is the database where we store files, variants and metadata for the patients. PanelApp is a crowd sourced knowledgebase where we store our panels for interpretation. Everyone can make a submission of the new genes which are associated with disease. Our in house curator reviews evidence and makes decision about including genes in this panel. Clinical Variant Arc only used for rare disease so far but we hope to use ist for cancer soon. It is a knowledgebase of clinically relevant varant-phentoype relationships captured through the interpretation process.

SP: Are you talking about all germline or gremlin and tumor?

AS :At the moment we are using panel app for germline variant for our cancer patients.

ASp: from What you’ve said it sound like only pathogenic and * will be curated. **

AS: I’d have to check. I think it stores all variants.

IF: How long does it take to go round that loop for a given patient?

AS: Those circles at the bottom is about two days. Sequencing in average takes two weeks. The rest depends on the genomic parts. We’ve seen cases where the DNA was extracted in a couple days, results from the clinic were presented in a few weeks, so the entire circle was 2.5 weeks. But we’ve also seen cases that take a couple of months.

IF: We see many cycles like this that take months or years. This is something that is providing immediate feedback for someone’s care.

AS: I’d say it takes 2 weeks - 3 months. Depends on the capacity and enthusiasm of the genomic lab. The current strategy to engage them more** We plot tumor mutation on the background of the patients we’ve seen before. This red line is a mutation burden for this patient. As yo ucan see it’s pretty high. We also return results for the mutaiton of signatures. In this specific case we see evidence of smoker signatures. Pretty common for lung patients. If you put all the pieces of this puzzle together it’s likely that the primary for this tumor was in lung.

I’ve explained the cycle in the clinical space but it also goes into research environments. We call external research, genomics england clinical partners. All data from the clinical cycle geos ino the research environment for public research needs with the expectation that new types of analysis can be translated back into the clinical space.

100,000 Genomes Project

AS: We were able to put this genomic data and match it to clinical data which is stored centrally in public health england databases in NHS digital and also in the national census* We have data on tumor staging, typography, histological subtype etc.

SP: Do they have tumor or normal genomes?

AS: for 17000 patients we have tumor and non-mal.

We partenred with academic research and so far it’s 84 institutions which are registered with us and half a thousand of the research projects. We also partner with pharmaceuticals. We witnessed about half a thousand covid genomes.

ADM: For this group it would be useful to point out that one of those is the lifebit cloud OS platform. That is next flow work flows. What is the second research environment?

AS: The standard HBS.

ADM: On premise.

AS: Most of the data is based on HBS. Covid data and our pilot on the oxford nano board goes into lifebit.

So when data is being used in life bit. We should distinguish the compute and * platform. Are yo prvodign any of the compute resources?

AS: Depends on the project. Majority of the 100K data sits on our local storage.

ADM: So researchers including pharmaceutical companies do you have them go directly into on premise.

AS: Majority sits on premise. Only covid data and data from long read sequencing sits on Lifebit.

IF: this may be going back a bit, but we heard that all of the data was on a former military facility and the only way to compute it is to go there. Could you put this into contexT?

AS: Majority of the 100L data sits in HBS environment. We are developing new research environments on AWS using life bit, but not all data is there. If you want to access paediatric cancers you’d have to go into our legacy research environment. IT’s on * Cloud. We are migrating to AWS so in the future everything will be moved to AWS but can’t predict how long that would take. We finished the 100K Genome project. As an outcome, NHS established genomic medicine services. Clinically accredited WGS service for cancer and rare diseases. NHS commissioned pediatric, hematological, cancers and sarcomas.

SP: Genome data is done as a primary test in a few centers, but it’s not the going test. In the UK, are these patients who would’ve had their tumor analysed in a RNA/DNA panel, or is this the only genetic testing that would have been done?

AS: For lung cancers we are still at the panel stage. For paediatric cancers, every patient has access to WGS. We’ve had quite a few cases where we’ve seen variant fusions in WGS which were not spotted by the RNA panels. Same for hematological cancers. Depends on the hospital and genomic labs. I would say that since pediatric has become eligible for WGS i can see a lot of enthusiasm among pediatric clinicians. So far 92% of patients sign consent to become a part of the national genomics research library.

New at Genomics England

AS: Cloud native trusted research environment on AWS using Lifebit’s CloudOS platform is still in the process of adoption. We started a large pilot on screening newborns using WGS. Also started a diversity program so we can sequence genomes from different ethnic groups in the UK. Our genomic data is supplemented by clinical data. In the future we hope that this will show up as live data in our environment.

IF: Any GA4GH standards tha you’ve looked at which might be relevant to this new AWS environment? E.g. cloud standards to help enable compute on AWS. it would certainly help. The GA4GH standards themselves can help with this gap of knowledge. Was able to make use of the DRS standard and enable use of it from the cloudOS platform. WES and DRS could help with things like containerization and access to data on cloud storage.

ADM: And there are other approaches. Like instead of using WES*

AS: I think next flow is what is in store.

ADM: Are people using next flow also on prem?

AS: Yes, I think people still have a preference for on prem*but I think in Lifebit we are probably forced to use nextflow because it’s the only thing that is enabled.

IF: DRS you’re not suppose to get the data out. You would compute with it on place. Where there are use cases like this it helps drive what that solution needs to do.

IF: For data on that research platform,w hat would you need to do to be able to authorized to access it?

The institution needs to sign a * agreement with genomics england, submit proposal to GC* domain and once that’s approved you can start submitting your workflow and run analysis.

IF: that’s a fairly typical pattern.

IF: What you’ve described is similar to what you’d see in EGA and dbGap. DACReS which standardizes approaches across data access committees.

SP: Maybe we should consider changing this name.

ADM: One thing genomics england has is the consent. It’s very transparent. Don’t know of anyone with such far reaching consent.

SP: Nothing has this kind of scale.

AS: Cannot re-consent retrospective patients so have to deal with the consent they signed for research studies. It’s usually difficult and limited. Given that this consensus is given so much flexibility more than 90% of our patients are happy to sign.

ADM: Don’t think we think this way because in the US we don’t have universal healthcare. Not sure Sharon if you’re seeing a change in that. In consent going forward is there an effort to nationalize that?

AS: The AGHA has developed a proecm called control which is a dynamic consent program where patients can go back and review it, and also see if it’s being used for commercial purpose or not. How that might go in the future in terms of national agreement is still ongoing.

SP: the closest is the children’s oncology project Every Child. 90% of children are treated at the child oncoloyg centre. They are attempting to enroll every child at diagnosis, there’s an epidemiology questionnaire.. In the pediatric cancer realm there is an attempt to do that. In the adult realm there’s more pieces.

AS: On the question is consent, specifically pediatric consent, childrens grow and once they past 18 they have to be reconsented. We are facing that issue now with patients from the 100K genome project.

IF: The other new standard announced at the GA4GH SC was a pediatric consent standard. There’s some mileage there for exchange of information.

SP: I don’t know how much it applied in a cancer setting. Typically in a cancer setting you have alot of language about additional tumors, relapses etc. so I didn’t know how much more of how…

AS: Standards I had in mind are a bit different. Standards for DNA Sample Collection, sequence, mapping/variant calling, and interpretation. This bit about collecting gremlin samples, I’ve gone through so many things over the few years. We put together sample collection guides and thought it was thorough but a year later we reviewed this germline that we collected* Now we try to develop pathways for collecting skin biopsies.

IF: Biospecimen Research Database: Database that NCI setup and continues to maintain specifically on this question. Nobody publishes a research paper specifically on methodology. This was intended to do exactly the thing you are describing. It looks at different analytical platforms of how handling and collection methods effect the outcomes.

AS: At the beginning of the program we tried to convince the pathologists to do WGS we need to collect fresh frozen samples. It was not easily transferable to the new platform. What is the DNA of sufficient quality? What is the genome of sufficient quality?  Started using Illumina platform and faced some resistance from research community. What are we using as reference. What validation can we use across platforms. Finally, what are the standards for interpretation. We have seven genomic hubs and they all run their in house classifications. We did work 5 years ago on an overview of somatic changes in FF and FFPE samples. Quality of genomes from samples in FF vs FFPE. seen lots of false positives in FFPE.

LS: Re: quality metrics - for anyone interested, a few national initiatives (led by Singapore and AGHA) have been working QC metric definitions - https://docs.google.com/document/d/17bVufpacyoUM4UDKlwkr-0KOG-yZrcVSvM6yC9gbrk4/

AS: This is our current pipeline for Genomic medicine services. We adopted Dragen alignment. Until three weeks ago you had to have Dragen hardware to run the algorithms but at the end of November GSDK announced that you could run it* We are doing benchmarking and validation. What we do for clinical accreditation is a bit different. So there are a lot of comparisons with DNA RNA panels. Research users want to see comparisons with the ICGC pipelines. Different users are looking for different validation. Also looking to adopt this graph maker from Dragen. Saw big improvement in accuracy especially decreasing rate of false negatives.

MC: This is very interesting. What are you anticipating with the future human pangenome reference?

AS: My thoughts now are with this telomere to telomere reference that were generated from the long- read sequences. Would like to map the long reads to the telomere to telomere reference. But it still doesn’t have much annotation so can’t use it in clinical space.

IF: What’s the right way for us as a group to continue to bring these kinds of questions to this forum within GA4GH. Which of these need to go off into one of the GA4HG Work Streams. Perhaps we can discuss this more in future calls. One of the questions would be whether we serve some of these things up in the group or if they’re better picked up in a work stream.

LS: Part of how I see this group is as a clearing house. When we see things of interest to WS we feed them in. For example, Melissa’ container work being moved to FASP. Part of this group should be directing the efforts. If there isn’t a place yet then we can do work here or finding the appropriate spot. We need to do some exploring about how to organize this.

DT: Do you have a catalogue to search for data?

You can use the open CGA catalogue. We also provide a lot of tables which can be interconnected. Clinical data exists in the form of tables.

DT: Do you have in your plan ways of connecting to the outside (Europe, US, Canada) in the future?

AS: In the end, if we’re all on the cloud.

IF: Want to access Genomics england data. Does access extend internationally?

AS: Yes.  There are some clauses about IP in our agreement that not all American Universities are happy to sign so it also depends on Universities.

IF: What we can provide is the motivation and use case to specific working groups. If DURI is the right group to hand you off to, we want to stick with it. Taking your motivation Melissa partnered with Alona we can carry that  over to DURI. To set the target for that group of solving the problem.

MC: I’M GAME

November 8, 2021

• Meeting Recording 

Chair(s): David Torrents (BSC), Ian Fore (NCI)

Attendees [Name(Affiliation)]: Melissa Cline, Angela Page, Sharon Plon, Lindsay Smith, Alex Wagner, Alona Sosinsky, Lincoln Stein, Michael Baudis

Minutes:

DT: Revamped the document. The discussion points: how to translate the commitment, clear that what we do is combing, bridging, integrating the technical and practical worlds. Need to discuss whether and how to expand the group.

IF: This is the first time everyone has been seeing the revised version. Changes were made over the weekend and this morning. Might be worth not going through every line but talking about the structure.

AP: There was a lot of repetition in the doc, tried to organize the thoughts. The introduction introduces what the group is, who is currently a member, what the goals are. This is unique group because it brings together a unique community that can bridge tech communities. Put the mission into one statement that doesn’t repeat itself - collaboratively define scientific use cases, commit to advancing use cases, and provide practical examples.

SP: Seems weird for the mission to be defining the use cases, as opposed to the mission being more about what we hope to achieve. This is more of a strategy.

AP: Is the intro more of the mission?

SP: Yes, that’s why we came together. Using the use cases as our strategy for achieving this goal.

IF: Good change, more tactical. What’s particularly unique about this group is that we’re tying to GA4GH standards.

SP: Don’t want the mission to be curing cancer. Applying the GA4GH standards to cancer to improve data sharing is a mission. Mission should be the wording around applying standards.

LS: I think the structure is fine, should say it aims to promote global cooperation, data sharing and implementation of GA4GH standards.

SP: Have a sentence that says we aim to accomplish this mission by….xyz

AP: The rest of the meeting is about the goals, and the structure and meeting format. One of the things that has been making things difficult, is that we (secretariat) need to be more proactive with the minutes. Big thing that David mentioned is expanding beyond the Driver Projects. How do we want to open up. Is it targeted invite, or do targeted invite as well as an open call. There are some important groups that should be here.

IF: KidsFirst would be a group very relevant to Melissa’s use case. How to bring KF in. There’s a certain level of formalities. Hope we can be organic about it. Does KF become part of the group or specific collaboration? Don’t want there to be barriers to whoever needs to come in for a specific use case.

DT: Would make it completely open. Is there any reason not to?

IF: Nothing that says the group is closed. What does being open mean. Anyone can come along? At the same time we want to make sure we’re effective at getting things done.

SP: I wonder that by starting out by saying we’re across DPs that means that you have to be one. Having that as the first statement will imply.

DT: Not sure if we need the structure

IF: Then do we put it out there as an announcement.

AP: This is something that the GA4GH comms team can take and work on.

DT: Check on your side that if we open it, it doesn’t overload your day.

SP: Work on something - question of whether you need to become a driver project. What’s really involved in becoming one.

AP: On tech team side, presumably the groups themselves would put their own devs

IF: If the expectation is that you come along with an use case FASP will solve it for you is not the model we’re looking for here. We’re putting in place the standards. In a situations where there are many things that need to come online, try to enable this. That doesn’t exclude anyone.

AW: What do we envision outputs to look like?

IF: When we say guide, hopefully not just a pdf, example code that someone could use. Essence is giving people something that would help them. Concern that this isn’t being prescriptive or requiring anything from anyone.

DT: Data users (and re-users) are not so common. What to do with shared data, it’s not obvious for many people.

Federated Analysis of Data Use Case

MC: We presented this to the FASP meeting a week or so ago. The FASP group met with the idea favourably. Targeting some hackathons in January where we can do further work on integrating with the starter kit. International biobanks that might have data that hasn’t been mined thoroughly, data from underrepresented populations, but in the shorter term, data that are more accessible. The research use case is Fanconi-anemia where a person may have 1 or 2 BRCA variants in trans from each other. Shorter life span in patients. It;s a disease with a very diverse phenotype, it’s a rare disease so there’s a lot of value in identifying and recording cases. Hypermorphic use cases are part of where the rubber meets the road, a lot of research value in identifying the hypermorphic variants and characterizing them further. Might or might not be value in terms of variant interpretation. Suggesting that a VUS in trans is some evidence of pathogenicity. With all of that said, there are a number of reasons for identifying and characterizing more fanconi-anemia cases. The KF cohort is a compendium we haven’t got around to mining yet. I understand Ian has made some progress in looking into KF. Had reached out to Momo to ask if we might work further, perhaps under FASP or under the starter kit, he responded that they have a number of disease cohorts we could potentially analyse. There is some sensitivity to how much work we ask of them without securing additional lines of funding.

IF: Broad update is that a number of pieces are coming together. With the discussion with Jeremy, the question has come up around getting access to the KF dataset, or whether to work with synthetic data.

SP: Confused, the work with Japan is not around fancomi-anemia right? [yes]. Don’t want to combine these two ideas together because they’re very different projects. All the KF data is all available via dbgap.

IF: They are available, but they are controlled access.

SP: Can go through the request for access. Want to be clear we’re not converting KF to open access. I think we need to crystalize the project. The goal of mining the KF data, the goal is to look for BRCA1/2 genotypes that might be consistent with fanconi?

MC: Yes, the goal of mining KF is to look for patients that have genotypes consistent with fanconi. The goal of doing so within this discussion, it’s one of a number of ways to make these work flows more accessible. In the case of biobank japan, had the benefit of IT savvy collaborator who was able to download and run our docker container and that’s not going to be characteristic of all institutions. One direction for expansion is integration with the starter kit that makes the analysis more easy to run on collaborators institutions. Integrate more easily with Cloud portals like Cavatica. Mining KF is a direction that exercises another mode of container delivery.

IF: getting the container to run wasn’t very hard. This is the container [shared screen]. It takes a list of pathogenic variants and a vcf file from some cohort and produce variant summaries. Was able to take that python application and put it in dockerhub and run on three different WES platforms. Orchestrate these three services through validator platforms. Was on test data. Find pan-can atlas germlines variants. Point was to demonstrate it could be done on the GA4GH platform. Basically, we could get to the situation where we could run the same workflow via WES through three data sources.

In Jupyter, the data doesn’t leave by data catalyst. The piece of analysis that happens centrally is the merge. One of the things not entirely apparent to me is that we’re not going to find vcfs that are mergeable. Had to do prelim step to merge into single vcf using vcf tools. The message is that we can’t be overly simplistic on what it takes to do these analyses.

Big part of what we need to do is not solve other people’s problems but enable them.

MC: Over the past year, had a fellowship under the biodata catalyst platform. One of my feedback elements is that the platform takes analyses that are in principle already possible and makes them a lot easier.

October 18, 2021

• Meeting Recording

Chair(s): David Torrents (BSC), Ian Fore (NCI)

Attendees [Name(Affiliation)]:

Minutes:

Planning Document

• Mission is to bring together standards development with application work
• Q/ We need to agree on what we do, what the group does. This document is for…?
• ADM: Felt like we were diving into the weeds, but really we’re really not quite there yet.
• IF: The science questions is a different set of weeds but the right set to be in. Technical questions lose sight of the science.
• Who is it for? It’s for us to be clear and have a common purpose
• IF: Only when we start doing things do we encounter those barriers
• DT: Are we in the early stages of connecting one standard to one need. This implementation might come much later, and come from the Driver group. Agree that not only discussing should be the point.
• ADM: Action could be to devise a priority. May not be a technical outcome.
• IF: To the action side of things. We are a collection of Driver Projects, and what’s common is that we have a focus on cancer. We have existing resources, good comes out of bringing those together. Identify different paths, collaborations that can be put together with GA4GH standards. Built on the standards we already have.
• DT: Maybe this is as far as we can get with what we have. Would invite anyone that wants to join. Happy to have this document available to show to people. Would not make it very strict.
• ADM: I like the stage you’re setting. It’s a forum where anyone can come and present their ideas. We bring each of us the ways of solving our problems. The best ideas are ones that make sense giving the environment, and maybe we miss something by not opening it up.
• IF: Initially started with the GA4GH Driver Projects. I don’t think we have engagement from all of those projects. We should be asking why, maybe we’re missing something.

Propose Stage: first a consultation and design stage, and then look to opening.

MC: Trying to decide if this fits in the second stage or not, but potentially engaging with parallel disease communities.

ADM: Differences and similarities between the different groups. Need to be interoperable.

IF: Need to ask Alex, this is a gap we have. How can this community help?

DT: Important that when we choose an use case, try to identify other projects that need that help as well.

https://elixir-europe.org/focus-groups/cancer-data

^ ELIXIR Cancer group. Finding new projects and clear ideas.

IF: Maybe get a presentation from this ELIXIR group, see what this group needs from a cancer community in GA4GH

AF: Cancer imaging group - a bit out of scope for GA4GH but worth exploring. Gary has left ELIXIR.

[Lindsay to reach out to Serena & Melissa re: new chair of cancer data group]

Have regular monthly meetings, with working groups on specific projects.

OH: Figure out how to connect with the AGHA Cancer work. Where the cancer community have different needs than a rare disease group. What is the argument to get involved.

IF: Act of finding a problem, and applying the standards to them.

OH: From the infrastructure work, there hasn’t been a lot where we find something doesn’t work for cancer. We’re not too much involved in groups like GKS where there could be better representation from other groups.

NS: Instead of having a separate group for Passports, proposed invited people to attend the Passports group and present what their needs are. Put out a call to see who wants to go to those passports meetings.

DT: Imagine that I did learn more about Passports for that meeting. I checked later whether we could use, and perhaps EUCANcan is not at a place where we can implement. From this point one, would have to force myself to go

IF: The FASP scripts I have written do not use passports because it’s not ready. Passports is being driven by NIH. From a cancer POV, I would like to be working with AGHA and ELIXIR. We have this international objective, does the standard enable that.

DT: Good for me to know that there is a group.

IF: The action item we have is to have Melissa come talk at the next FASP meeting. In the meantime I asked Melissa about what is the virtual machine we’re trying to run. Gene fusions was a big part of the disucssion

September 13, 2021

• Meeting Recording

Chair(s): David Torrents (BSC), Ian Fore (NCI)

Attendees [Name(Affiliation)]: Anne Deslattes Mays, Alex Wagner, Lindsay Smith, Angela Page, Susan Fairley, Sharon Plon, Digant

Minutes

DT: Will take Passports back to Eucancan.

ADM: What worries me is if you don’t keep the use case in mind, the technical becomes divorced from the pragmatic way of addressing the problem. Want to keep it simple and pragmatic.

Gene Fusion Discussion (review June presentation here)

Slides

AW: High-level discussion and talking about how the cancer community can get involved.

AW: There are a lot of different representations of fusions. Broader representations such as ALK fusions can be difficult to match programmatically. Trying to characterize and represent this stuff through GA4GH VRS. Reduces the amount of human analyst effort in doing that matching. Encourage GA4GH driver projects that are not already engaged to come to the upcoming VR working group meetings. Fill out this form if you want to participate.

AW: The idea here is to make it easy so that when you're when you've got preliminary variation data sets that you can, you know, basically represent these in a format that allows computational access and search of various, you know, knowledge providers, like civic or ClinVar, etc. and find that information very readily without needing a human to go in

So that's that's the objective. We've been doing this in the context of a multi consortia initiative. So this started out with the Vic and Clinton driver projects, particularly the quinjet, somatic clinical domain working group working together to drive this since then we've brought on the Cancer Genomics consortium and the ACM G and caap, cytogenetics. Committee, to basically help us with this characterization efforts. And its role in clinical relevance, which is, which is really what are, our primary objective is here, I'm bringing this all the way back. And there's a bunch of technical detail here, which we're not going to go into today. Again, I asked that people go back and see the recording for a little bit more info on this. But you know, the high level view here is that there's there is a, you know, genomic observation model that we're trying to accomplish reverse and there's a broader model that we're trying to do around genomic knowledge, and how you represent these variation concepts, as the complex genetic, like variation, knowledge entities that they are.

And, you know, we'll there's, like I said, a very active Working G6roup around this whole thing, our next meeting is actually on Wednesday. So you know, anyone that has an interest in this, I'd recommend that they participate in that initiative. (form to get involved) And I think it would be great to get more g4 gh driver projects involved. Now, obviously, as I mentioned, Vic and Clinton already heavily involved in this, but but you know, we're really on this genomic knowledge provider side of things, like we're talking a lot about the genomic knowledge that we've been curating and annotating and capturing for use in the clinical setting.

Sign up for ongoing discussions: https://forms.gle/LABme2LMfNBAP2QC8 

LS:  Who if anybody is doing whole transcriptome RNA seek in a diagnostic context? Because in theory, you pick up the known and new fusion fusion transcripts, that way, if you can get sufficient depth of coverage, it might be as accurate as the PCR based methods.

SP: Large funded projects. So for example, we're doing it on an NIH project, but I can tell you, we report to the oncologist based on a panel a long time before they get their transcription report.

LS: But it's all in research all in research context.

SP: It's not a research context. It's a clinical transcriptome and St. Jude does, I think transcriptomes on everyone. I believe the Canadian profile, I think is the name of their large national gastric cancer project. I think it's transcript them. The problem is it's it. It needs external funding. Yeah, um, and so I don't know, these are all trials that you're talking about. They're all trials, but they're clinical. I mean, at least in many of them, at least I know ours in St. Jude, there's a clinical report that comes out of that transcriptome. But I'm just saying they're funded, not through payer regularly regular clinical meat, they're funded through research or philanthropy or stuff like that. Oh, Alex, do you know others that do clinical transcripts?

AW: Yeah, our group here at IBM does clinical transcriptomes. And it's it, but as Sharon mentioned, you know, you need funding for it. It's not reimbursable.

LS: And if cost if if it were reimbursable, would it would whole transcriptome be a viable alternative to PCM PCR based tests and other typically specific assets? Which is it? Or is throughput? Also, the the or turnaround time also an issue

AW: Turnaround time is not really the issue, I really do think it comes down to cost.

LS: seems to me that whole whole transcriptome and least whole exome should have been, should have been, the uptake should have been much faster. I was expecting by 2021, that we'd see widespread adoption of next generation sequencing technologies throughout the healthcare system. And it's surprisingly slow

SP:  I actually would correct one thing you said there is wide scale adaptation, adoption of next generation sequencing technologies

LS: In US, not in Canada.

SP: Well, I don't know how many panels are being done. But what I was gonna say it's not exome. I can tell you so far, the panels doing really pretty well, they're, you know, among we've now enrolled 600 patients, and the data is not complete, but the number of patients were something important was picked up by an exome and transcriptome and array comparative. current state RNA DNA panels, is going to be very small. So, and clinicians are very pragmatic. You know, if you can get your answer in two weeks with a simpler test, then that's what people do. And it doesn't mean that there's not a role for exomes, or genomes and research. But its role as a clinical test has been much more rapidly adopted in neurodevelopmental disorders,

DT: How about variant calling from RNA seq data, somatic variant calling, is it sort of is it okay. I have not followed the developments on that front. But is it people working on trying to substitute taxon?

SP: I know groups have worked on it but I don’t know if there is a good agreed upon program at this point. There are a lot of RNA errors.

DT: How about promoter gene fusions? That happens often in blood tumours and IG regions.

AW: I'm really glad you brought that up. So I think one of the one of the big conferences in the field of gene fusions is or not conferences, conflation is in the in the field of gene fusions are the differences between what we call D regulatory gene fusions, and chimeric transcripts, Gene fusions, both of these are considered gene fusion, two, they're very distinct mechanisms. And, you know, a promoter fusions can kind of span that divide a little, especially if you get the promoter plus a little bit of, you know, CDs or something. But the point here is that, you know, um, you know, there's there is this large class of, you know, promoter driven or enhancer driven gene fusions that are very clinically relevant, you know, the HGH MC fusions, for example, where you have the, you know, the the HGH, super enhancers that, you know, basically way up regulate MC, and that causes, you know, all sorts of problems, and it can drive cancer, you know, this is a very clinically relevant rearrangement that doesn't involve any chimeric transcript at all. But it's still called the gene fusion in the same way that a BCR Abel one gene fusion is. And one of the challenges that we're trying to address is basically being more precise about the class of gene fusions being represented. You know, there are proposals out there right now, that for gene fusion, nomenclatures, for example, that kind of overlook this challenge, and so one of our one of our goals, is to try and help educate and promote and revise these types of nomenclatures. So that they can, they can, they can help us, you know, kind of disambiguate between these different types of events. But it is absolutely a critical part of the salient elements of gene fusions that were characterizing this project.

AW: It might be worthwhile spending the remainder of the call, talking a bit about the call. So I think that I think that I have talked with a number of people that are associated with this initiative, that had comments similar to what we saw on our last call, which is like, this is too technical, I'm not sure like what I'm doing here, like, I'm very engaged in this thing, but not that thing. And I don't know what we're talking about this week. And like, there's a lot of confusion about, you know, the purpose and intent and drive of this cancer community effort. What we're doing in these meetings is a unclear. As such a targeted call, it's not even clear, like to what level we all benefit from us going over this and at what level. And so, you know, I think that there should be maybe some discussion as a group as to you know, how we better organize these calls, decide when to have these calls, and how we can utilize the community in a way that's, that's very effective for everyone's time and energy.

SF: I agree with you. Absolutely. I think these calls were incredibly useful and incredibly valuable. The other thing that strikes me, Alex is, as you're speaking, you're describing these issues, because we are very fortunate to have on this call people who have real uses for infrastructure to solve real biology and clinical issues. And I think, for GA4GH we need that. I think one of the things that we need to cannot overcome as the wrong word, but I think something which is perhaps a feature of the way in which we are working is that inevitably, when we come to develop some specific standard, we can end up having a very focused, very technical discussion, that of necessity, ends up sometimes generating its own jargon and vocabulary and language to talk about those terms. And then it can be quite difficult to kind of bring these two back together again, and to figure out how you go from this.

DT: Now there are there are many of these application environments that are being generated. So it's also very good to have this feedback from this environment to the group. Because you were originally starting with quite broad idea and broad goals, but many of them have those goals were only theoretical, for example, many of them, they cannot be set up practical level because I don't know, legal issues. So now the real environments that should adopt those those those standards, in many cases are being generated.

ADM: Does this group have a mission statement? That's my first question. If we had a mission statement it could act as a tester of whether this is going in the right direction.

AP: Really like that idea. I think that we can probably do that offline, and maybe just iterate a little bit together. Goal of this group might be different depending on who is answering the question. Coming from ga4gh, I think my goal is to get broader engagement with this entire community to first it was to bring together the cancer driver projects to figure out what are the use cases that we can immediately start working on? But then also, how do we use that existing driver project community, the existing cancer projects that are connected to ga for gh, to reach out beyond those driver projects into the wider cancer world?

ADM: Along with the mission statement, maybe making it clear if somebody from the larger group comes involved just to like, give them if they take on this particular persona, you know, I'm a clinician, I'm going to ask this or I'm technical, I'm going to ask that. And, you know, where does the persona fit in the mission statement might be useful to take a stab at trying to understand so that when we invite the larger community, they will know, okay, well, I fit here, and these are the things I should be looking for.

NS: As a next step we can create a mission document and work on t offlin. We can come back in our next meeting and use that document to discuss what the best format for the meetings would be.

August 16, 2021

• Meeting Recording

Chair(s): David Torrents (BSC), Ian Fore (NCI)

Attendees [Name(Affiliation)]: Alex Wagner (Nationwide Children’s), Kurt W. Rodarmer (NIH), Susan Fairley (GA4GH), Angela Page (GA4GH), Melissa Cline (UCSC Genomics Institute), Salvador Capella (BSC/ELIXIR-ES), Mandy Spurdle

Minutes

IF: Interested to pick up on passports discussion. Passports is more than cancer, but it can solve problems within cancer. Questions for the group: What are the problems we want to solve? And what are the expectations from passports? Seems in some sense that there is an alternate proposal cased around ERACommons that we should come back to. One of our goals as a group is if we can provide passports a scientific use case. What can we as a group add [e.g. a use case] where we want to use passports to validate it. Do we have a set of providers that would be willing to support passports.

ADM: Been offered and accepted a position within NIH. Going to be NIH CAG data scholar. Focused on getting data, making it interoperable. Have a specific problem and want to make access to data very clear. Being pragmatic.

IF: Gave Kurt some background earlier today.

GA4GH Passports Presentations (to be linked)

KR: Passports are digital tokens of authority. Just as binding in many use cases as legal signatures on a contract. Introduced with idea of increasing accessibility but not at expense of security.

Identity-based Security (current state)

Identity based security is what exists in every corner of the world right now. Considered the general standard, but doesn’t work. Can’t easily federate. Every time we connect to another system, that systems needs to know who we are. Not just recognizing face, but actually having an account. Very cumbersome and puts up barriers to federation. Uses late-bound authorization. At the moment you request access, the system decides what you can access.

Token-based Security

Very well studied. It is what is used in secure operating systems.  Tokens can be issued by anyone but only recognized if issued by an acknowledged authority. Can not be forged. Work well because of the digital signatures on them. Good at early-bound authorization. User gets authorizations up front in the form of a token. Passports are an example of this type of token. The kinds that we use at NIH are more similar to the capability model. Express specific permission in regards to specific objects. One of the common ideas in the GA4GH model is some assertion of good standing at an organization. That can be carried in a passport. Whether it is accepted is based on several conditions. Really only good if the recipient decides to accept it. In the case of data owners/stewards that have distributed systems, will establish a policy with the system that they need to accept tokens that come from the owners, and to not accept those coming from non-owners. If the recipient owns data they are handing out, they may be able to assert/assume liability of others.

Utility of tokens depends on issuer. Non-owner token may be acceptable depending on policies. Regarding Anne’s assertion of...trying to get away from identity-based security and ERA is all about that. ERA is used to attest to users identity. That does happen in the token but the authorizations that are carried are from dbGap. Receiving host not in a position to look for specific identity token*

ADM: Having had to go through the process many times of getting access and authorization, want to see the mechanism replaced but not in a different location. Making lives easier for researchers to find and access data. That is the common way of many people accessing it, so one way to solve it might be to take your replacement and put it in there.

KR: That's what we’re doing.

ADM: Never know who the “we” are.

KR: Work on the DURI passports project. Within NIH I work on the RAS project. Which is a manifestation of this entire thing. Dbgap is converting 1000% to GA4GH tokens issued by RAS. Researcher Authorized Services

IF: Completely different model than this one where we receive an ID and we have a local database that says what that person is allowed to access. In passports, it’s in the passport where it will say that the bearer of this passport is authorized to access dataset e.g. PHS00154 or some other dbgap dataset. That will not change that to get access to that dataset you’re still going to have to go through the back process. GA4GH cannot change that. Got other situations where the DAC is actually the investigator who created the dataset. What problem are we as a group trying to fix? Let’s look to the right people to try and solve them. One of the other policy things is the thing in your passport will say you are a bonified researcher and the model we want to get ot is not that we have to go to a DAC and prove that you are a researcher and will get access, but your institution will say they are a (e.g. pancreatic) researcher and you can access relevant (e.g. pancreatic) datasets. What are we looking for passports to fix?

DT: How is it used in NIH now? Is it down to the person? Still don’t understand, is the token bound to the requisition, data, person?

IF: Log on to NIH, it would give you the passport to give to another person who has access to data set

KR: log in, NIH is able to retrieve early-bound authorizations, those are packaged into the token, and you are given a token with NIH signature. It is bound to you and explains your authorizations. When you go to  another system that doesn’t know you but knows NIH’s signature and you are willing to honour that then it works.

IF: If I got NIH and went to ega what happens?

KR: Given that EGA is not in any agreement that I am aware of with passports, it wouldn’t get you very far. It represents to the researcher instant access. Institutionally it represents liability for any of the actions that the researcher takes.

NIH can’t give me a passport that knows what EGA datasets I have access to.

KR: NIH not in a position to say you get access to that data.

DT: Would be for data club* to see who can login to the system? Then within the system you can access different systems.

ADM: If you login to NIH isn’t that an identity?

IF: The way that access is being granted at the moment - I’ll grant access to people who are working on my disease.

ADM: Genomics England - the success was based on very limited consent. Think we are over complicating it. We have lots of technical solution but we want machine and man to use the data to solve the problems. Don’t want to get lost in the weeds. Would hope the solutions lower the barrier to entry.

SC: Following the passports implementation. Working with people who want to ensure access to data sets. Who is issuing visas? How to recognize*

DT: Saw you mention EGA. If you have to ask permission from investigator, how can you skip this step?

SC: There is a combination of passports and dUO. Related to bonafide researcher. If you as a data owner say this can be granted access to* when someone comes in with passport saying they are a bonafide researcher, access is automatically granted.

KR: Does not happen automatically. Data owner would have to have a policy that permits use of DUO codes and automatic grants.

DT: Justification of why we want the data for is then gone?

IF: Yes, there’s a recording from plenary where this is described. Tried to walk it through to demonstrate. The proposition to a researcher who is sharing a data set - we’re asking them to tag their data set. If you tag that set with the DURI code, you’re then off the code for having to grant access to people who want to individually see your dataset. The system then deals with it, because it validates that this person matches your tag.

KW: Just want to note that this is not what is currently happening.

SC: One of the plans we have been discussing is having associated user with visa. Not necessarily being used in DUO DURI functionality*

KR: Want to mention that one of the things you do with tokens and will be happening with passports, is a passport may not carry all of your authorizations. Multiple levels of authorizations and depends on sensitivity of data.

IF: Salvador, are you part of Eucancan. Do we have from the cancer communtiy a set of organizations that are interested in implementing passports? Do we have a set of repositories that we oculd work this through with, with a specific cancer application. Chop, eucancan, NCI. How does this work out in practice?

SC: Happy to be part of it.

DT: Eucancan is example that Kurt was describing. Different countries, different partners. Probably will be close community because there is sensitive data that will be placed in the catalogue. There has to be an authority in the end. There is not single one authority in Eucancan. Will have to agree on some common ground.

IF: Yes, must be a need for authority among the consortium. The important authorit, is the authority over the data. And whether the people in charge can do what we are asking of them.

DT: Will start those conversations in September

IF: How could we write this up as a test? What would be the test that if you passed that it would work?

ADM: Starting to make me feel like when you wanted**

IF: If this works, I will be able to do this. E.g. if this works I will be able to take a passport and send it to OICR, Barcelona, EGA, and whatever set I can get access to in an easy way.

DT: Guess it’s how many steps you skip from the old protocol to the new one. Should have the same access but the time and effort should be less for everybody. That should be the criteria. Need to understand better everything about tokens.

IF: Just trying to get to a point where we can write down what we are doing together

Commit by the end of September to have some plan/conversation at Eucancan. Bartha is one of the partners so she knows quite a lot about passports as well. Will be happy to present it as a text or presentation.

Gene Fusion

Is htis something for the cancer group to do or is it something that variant rep group needs to do and interested parties from this group should get stuck in in this activity within tht gorup? Alex is on a roll with what is described here (disambiguated fusions, what are the salient things that need to be collected), but it may be better handled by experts in VR.

AP: Could it just not be a use case...are you saying it’s technical enough that it’s not for this group to discuss? Think it’s find for this group to generate stuff for multiple WSs.

MC: It is something that the VR group has been working through so Alex is looking for an intersection between the WSs.

IF: Question I’m asking of people in this group is, should this be an active discussion topic within this group?

DT: Think it’s more a technical thing. Structural variants are not well resolved in cancer. More common than gene fusions. Not saying we should not discuss gene fusions and other complex annotations, but it’s more a technical topic.

MS: Not in the somatic space yet, but this would be a key problem to solve. When is the gene fusion clinically actionable? What makes up this group? How many people say they are a prt of this group and didn’t come today? If it’s something we think is important to move the community forward, then why wouldn’t it be part of this gorup? How did these get on here? Was it a poll?

Got on here by talking in this gorup about use cases that were important to a number of the cancer driver projects.

Then isn’t that a reason to keep it up? If you want to say we only concentrate on a certain number of things then we need to rank them based on how big the problem is.

IF: There’s also the divide and conquer approach. The service that this group is doing is connecting with all of the cancer driver projects. Icnreasing awareness that this activity is going on but not necessarily being the forum through which it is resolved. We probably got into too much technical details about passports.

I don’t think it’s only technical I think there’s so much more to it. Given the limitations of where things can be represented because of the way gene fusions was identified, how can it be used clinically.

Is there an argument for arranging a conjoined call. We invite Alex/Larry to take the time to talk it through. So we can make sure concerns get passed on to WSs.

We should do some more promotion to the cancer community that will be the topic.

Let’s say that for September, we will pick passports up again in October

AP: Other thing is that passports doesn’t need it stop. Perhaps we move folks interested in passports to another venue.

MC: Quic updates. We have the paper out for review. In terms of looking at next steps. Have an email to to ** who said that he would love to be involved in further work moving forward. Think more generally, we might frame this use case not strictly on allele frequencies and co-occurrences but federated approaches to collecting variance evidence. For example, Mandy and I have had a number of discussions on gathering federated evidence on family pedigrees which are highly identifiable, usually cannot leave their institution highly informative for variant interpretations.

IF: This one seems like a perfect use case for FASP. Maybe we can look at taking the docker container and saying how would that run under WES in a FASP context. This is a reason for why we would want to do a federated analysis. Need sites that can do a WES service but can we set up a pilot where we can do this through WES in this use case.

MC: Let’s set it up.

IF: I think the idea was to have you come to a FASP call and to get a FASP expert looking at a use case.

MC: Angela, can you connect me with the FASP people?

SF: Like the idea of taking this use case forward. Been some discussion about looking at data connect and connecting it to passports. Starter kit development has been looking to bring WES, DRS and passports together. Working to get this installed in a few places. If this is running off of WES...Like that there’s overlap between use cases and standards being used. If there’s any way I can help, feel free to get in touch.

July 12, 2021

• Meeting Recording

Chair(s): David Torrents

Attendees [Name(Affiliation)]: Alex Wagner, Anne Deslattes Mays, Peter Goodhand, Hidewaki Nakagawa, Digant Shah, Lindsay Smith, Neerjah Skantharajah

Notes

LS: How do we connect these discussions to similar ones in passports/AAI?

AM: Pragmatic outcome that would affect a large number of cancer researchers. It is a barrier. Ian would be key to understanding who is it a NHGRI that we would need to reach out to.

Separate topic: Need to make sure that regardless of how people access data, all sequencing data should be sessioned in..

PG: Question about boundaries. Passports/AAI makes sense from a genomics point of view. There is a lot of cancer research that does not involve genomics. Will we have a boundary issue? Is that why you are saying to link it to ERACommons?

AM: What other data?

PG: Imaging data, clinical/pharmaceutical type data.

AM: ERACommons is a unifying force.

SP: People who don’t have grants generally don’t have an ERACommons ID. Shouldn’t assume that ERACommons is a common metric across oncology.^[a]

PG: Came up in discussion with NIH. Against it because they work with individuals from all over the world. ERACommons needs to be in there but also needs to extend beyond that.

AM: Zenado uses data site to..ERACommons is a pragmatic way to apply things. Point is to get a unique identifier for the individuals and link that to access permissions^[b]. What would be the unifying identifier?

SP: Not sure how widespread ORCID IDs are. Haven’t seen anything that reported the proportion of academics that have one.

PG: Vague and crosses across all fields. Anyone can have an ORCID ID so ORCID doesn’t mean anything but ERACommons does. More of a branding thing. Have something good in Passports. Need to think about a name for it that is less branded to GA4GH.

AM: ORCHID is a unique identifier for individuals and institutions. Does the passport also require a unique identifier^[c]. If so, we need someone to maintain that.

PG: Take it back to the passports group. More positioning and branding than functionality.

AM: Who is the unique identifier generator for the passport? Suggestion is that the ORCHID ID could be the unique identifier. Specific VISAs will also need a unique identifier - where is that generated? What will the pharma companies be open to? Etc.

SP: Why don’t we want a branded GA4GH passport?

PG: When taking it to non genomic communities it will help it get broader use.

SP: NCI is putting a lot of effort into the data Commons vs whole child cancer initiative. If GA4GH decides to take this goal, which is a pretty big one, there needs to be some distinctive messaging in regards to why this is needed. The new NIH policy is that all data that is funded by NIH needs to be submitted. In the midst of this, it might be nice to think of ways that this could support that effort, and not be a totally parallel effort.

AM: Might be able to sell this whole passport idea in anticipation of global need of sharing. Would be good to have use cases that span multiple countries.

PG: Best thing to do is feed these conversations into the passports group.

LS: Can aim to have the passports group join next month.

AM: Ahead of our next meeting, maybe we can get people to sign up for use cases.

LS: I will circulate something in advance of the meeting.

AM: Do you have an implementation of a passport at this time?

LS: Since passports is approved there should be some out there. Can find it and send a link to it.

SP: Is it clear how broad this can go? Are we set up to submit transcripts of interviews 2 years from now example? What does the consent say? Difficulty in truly anonymizing transcripts? Another type of use case is to think about how broad the data structure goes? Or is passports just about recognizing researcher identities?

AM: Sounds like Phenopackets could be related to this...

LS: Not aware of them looking at phenopackets...

AM: Genomics England’s consent process: 1) consent to get commercial return on data 2) access to reach out to the patient in the future and in the past to clean up items if they need to

SP: Have always asked for permission recontact. Never actually specified that it’s to go back in time. Assumed that permission recontact is fairly broad. There’s a difference in saying that by signing this consent we might have to go back into your medical records vs going back to ask them something.

PG: In Cancer Care Ontario it was about patient records. Let us [the patents] make sure that you have the right clinical data on us.

SP: Critical for any study looking at side effects. Sometimes people really emphasize the importance of recontact forgetting how much work it is to recontact people. Want to make sure you are collecting as much data upfront as you can.

June 4, 2021

• Meeting Recording

Chairs: Ian Fore, David Torrents

Attendees [Name(Affiliation)]: Melissa Konopko, David Torrents, Ian Fore, Melissa Cline, Alex Wagner, Angela Page, Anne Deslattes Mays, Xi Luo, Susan Fairley, Christina Yung, Neerjah Skantharajah, Lindsay Smith, Salvador Capella, Michael Baudis, Alfonso Valencia, Malachi Griffith

Notes

VICC: Alex Wagner

Slides

Challenges in variant representation, impact on ability to rapidly classify variants, esp gene fusion events

Overview of efforts in ClinGen and VICC and others

Working on protocol for characterizing gene fusions

Background: VICC established to build expert curated interpretations and integrate into reports, search consistently across knowledgebase (KBs) and use in aggregate, standardize cancer variant knowledge

 CIViC used by ClinGen to curate info about gene fusions and other variants causative / applicable to clinical interpretation in cancer

Eg. BCR-ABL, ALK Fusions, single genomic coordinates provided on CIViC, but otherwise all plain text

Contents of KBs can vary quite a bit, eg. fusion notation and specificity

Standards to computationally semantically describe don’t yet exist, working toward that in VRS

Variability of data depending on assay used in clinical setting

Scientist has to take preliminary datasets, from different assays, annotated in human readable text, based on observations etc., how likely does variant apply to this patient/tumour

Ambiguity takes up lots of time when trying to interpret gene fusions

VRS - well defined semantically define computational model could enable automation of the process

Quickly ID when observed variation actually matches literature curated in VRS supporting KBs

Export into report to variant scientist, who validates it, stores it and reuses in downstream studies

Scale variant interpretation

Cross-consortium initiative, ClinGen, CGC, VICC, CAP/ACMG cytogenetics consortium

Goals:

• Define and disambiguate gene fusions
• Reccs to collect/curate salient elements
• Reccs for fusion representation

Problem scope: what defines a gene fusion in the first place?

• Deregulated or novel transcripted formed by the the interaction of functional element of swo or more genes

• Those that lead to loss of product are not gene fusions, out of scope

Chimeric Transcripts and gene fusions that drive them, Marilyn Li HGVS

MVLD Structure, Gordana Raca and Angshumoy Roy

Curation, …

Gene fusion curation workflow captures elements of the model

Open community project, pulling together into an SOP, strong recommendations ready for initial draft, looking for feedback interest from the community

Q&A (in chat)

Ian Fore to Everyone (10:26 AM)

Alex - are the knowledge bases you would like to use this representation of fusions respresented here? Is this group a way to get more Knowledge Bases involved.?

alfonso valencia to Everyone (10:30 AM)

Alex: Milana Morgenstern first in my lab and now in her own lab has been working in fusion genes product of the combination of a direct and reverse copy of a gene. I wonder If you have deal with this type of cases.

Alex Wagner to Everyone (10:35 AM)

Ian: The VICC has several KB partners (CIViC, OncoKB, JaxCKB, PMKB, MOAlmanac, CGI, others…) that we work with to develop these guidelines and tools. We have been discussing specifically with CIViC on how we will capture these recommendations in the curation interface. The other half of this challenge is how to precisely characterize fusion assay data from clinical & research laboratories and the tools to match to these data. I think this group may have more representation / interest in the latter half of this equation.

Michael Baudis to Everyone (10:36 AM)

We run Progenetix  (somatic/cancer CNVs) on a Beacon v2 stack, with e.g. relative variant frequencies per diagnostic code. So var frequencies per phenotype are +1 in Beacon v2.

Alex Wagner to Everyone (10:36 AM)

Alfonso: sounds like a very interesting case! Would be great to get input on challenges in representing or searching this.

alfonso valencia to Everyone (10:38 AM)

Alex: I will follow this examples/paper by email to see what do you think. Beside being biologically interesting, the issues of detecting and representing are also potentially interesting.

Mining germline variant co-occurrences: Melissa Cline

Slides

Largest pop of variants are of uncertain significance, data out there could address this problem to interpret them but accessing data is problem

Successful proof of concept analysis with Riken on Biobank Japan (BBJ) data, shared with them containers for analyzing BBJ cohort, now interpreting some VUS, classic case of data we wouldn’t be able to access directly

Data on japanese population also helps address the fact that most research data is on caucasians, need more ethnic diversity

Best traction not from co-occurrences but from allele frequencies, looked for co-occurrences of VUS with pathogenic variants, measured allele freq as analytical control, that was where we interpreted most data

Even compared to what we’d get analysing same variants with data in gnomAD, able to get more information simply by having the 24K controls in the BBJ (vs. 10K east asian controls in gnomad, small handful of Japanese per se)

Allele frequencies is a nice and simple place to start in terms of data collection, also requires very little information

Useful to see phenotype to know if working with affected set, but “one person’s disease group is another person’s control”

Allele frequency info is also a great way to preserve patient privacy, aggregated, privacy nuances in aggregating data on variant level, come in when you have highly rare variant, unique to a single family or individual; for collecting variant evidence for allele freq’s you can filter out that scenario

See a variant at some threshold frequency, filtering out highly rare observations

Would like to analyse highly rare, but starting with ID’ing common observations still gives us traction

How can we do this with a greater number of datasets? Can send container to another health initiative

This is a great use case for a Beacon or Data Connect API

Next step: working more directly with phenotypic data, case counts of variant by phenotype are basis of many diff forms of variant evidence, can address that with HL7 FHIR and Phenopackets to ask two questions:

1. What phenotypes do you see in your cohort for this variant
2. Given this phenotype and variant, how many patient observations do you have

Approaches for federated data query are Data Connect and WES starter kit implementation in conjunction with container at siloed dataset to specify parameters to a container that is located remotely

Someday: Pedigree analysis, co-segregation analysis, one of most powerful forms of variant evidence

Complex computation not viable with beacon query, pedigree subgroup of Clin-Pheno work stream developing standard to address this in future, moving in right direction but not quite there yet

Pedigree data is hands down identifiable, debatable whether one can look at an image of a pedigree, but great information, commercial testing labs are using this line of analysis more and more in conjunction with cascade testing

Q&A

What would be your vision out of the cancer community group? DPs who can query through Data Connect or Beacon, sending workflows via WES, which of these projects represented would implement this?

Should/can we engage the national initiatives (NI) in this?

RIKEN wanted data represented in research cohorts, won’t be the only nation that has this wish

Are there national initiatives prepared to participate?

We have technology, we have data holders in the NIs, are they in a position to start being data providers

Take idea of container to next level, 150 different country genomic efforts (beyond just DNA, transcriptome, proteome etc) - how to share workflows that contain containers

Get everything into repositories, make sure tested, and then share those, share results on aggregated level

Commercial vendor as a compiler of workflows, use GA4GH standards to get it done

What is in the container - what do you ask of the dataset? Looks at vcf input, takes genotype data as input and asks two questions - where are there unknown variants that co-occur with known pathogenic variants, and what is frequency of each over the

Enabling Passports to work with ERACommons ID and dbGaP:Anne Deslattes Mays

Want to go faster! Answer specific driving cancer question, access data world wide

At last call lincoln suggested talking about making ERA commons a means to access data

Containers as smallest element, one function in larger workflow, all pieces containerized so can be used

More standards to help see what data is out there

Tens of cases for intracranial germ cell cancers in US, many more in Asia

Lifebit is compiler of workflows, compiled on nextflow, many different workflows just in time fashion to ask question

Share aggregated results on to know whether variant, allele or isoform level does what we think

GEL - biostatisticians, clinicians, together

US has no medically embedded data initiative per se

Ga4gh take ID of you and datasets you’re authenticated for, pull direct from github

Q&A

Should that discussion be part of DURI? Need passport discussions to get out of “what ifs” and into real concrete use cases

Some possible now, GEN3 - 4 NIH cloud platforms, CRDC, CGDC - not actually using passport yet though

Though ERA commons is very NIH focused

Is there an opportunity in GEL to find scientific use case, send compute to GEL, federate something with them

NEED to go international, diversity doesn’t exist in datasets in single countries

Technical people discussing technical things, but serious cancer questions we could answer now - what would it take to cross borders?

Figure out how it’s done now and how we can improve on it

FASP - can I do a federated compute? Yes, but gone where there are GA4GH implementations, but not driven by scientific question, data in these two places I need to ask a real question of

Getting to those global examples is a key goal of this group

Research community has been developing tools, infrastructure, for many years; GA4GH comes from this community; now seeing in Europe at least that whole design and control is from health, far from our environment

Coming from one side developing nice things and standards, need also to connect initiatives that are racing to control / set rules for use of data - coming from health

Okay on genomic part, research controls that, but clinical data is key

Pharma companies want to come in and access the GEL data

N3C

H3Africa

Not a technical problem needed to be solved, we have the technical bits

ACTION: (now and forever!) document the use cases - this has been done through slides and elsewhere, how do we establish that as an accessible body of information?

Where are there gaps that GA4GH technical staff can help move this forward?

March 2, 2021

• Meeting Recording
• Detailed Minutes

Attendees [Name(Affiliation)]: Melissa Cline (BRCA), Angela Page (GA4GH), Peter Goodhand (GA4GH), Digant Shah,  Sharon Plon, Lincoln Stein (ICGC/EUCANcan), Alex Wagner (VICC), Christine Williams (OICR), Ian Fore (NCI, NIH), Hidewaki Nakagawa, Jonathan Dursi, Jennifer Lee, Stephanie Li, Amanda Spurdle, Jordi Rambla, Augusto Rendon, Amber Johns

January 25, 2021

• Meeting Recording

Attendees [Name(Affiliation)]: Melissa Cline (BRCA), Anne Deslattes Mays (Science and Technology Consulting, LLC), Jan Armida (SPHN), Gary Saunders (ELIXIR), Angela Page (GA4GH), Lindsay Smith (GA4GH), Peter Goodhand (GA4GH), Digant Shah,  Sharon Plon, Lincoln Stein (ICGC/EUCANcan), Alex Wagner (VICC), Christine Williams (OICR), David Torrents (BSC), Ian Fore (NCI, NIH), David Higgins (CHOP/Kids First), Salvador Capella-Gutierrez (BSC/ELIXIR-ES), Michael Brudno, Hidewaki Nakagawa, Alvaro Ferriz, Alona Sosinsky (GEL), Allison Heath (KF/CHOP), Michael Baudis, Christina Yung

Notes: 

What this group does: ID’ing cancer applications for GA4GH, not just to use standards for the sake of it but to drive a need

Mining Germline Variants Co-occurrences

Federated analysis for cancer variant interpretation

Presenter: Melissa Cline

• Algorithmic access to data we might not be able to analyse/share directly

• Proof of concept analysis with BioBank Japan

• shared docker container for them to apply to protected patient level cohort, sharing variant level data, being analysed by ENIGMA for interpretation of co-occurring variants
• Looks like it will impact several VUS, will be in ClinVar (and publication) relatively soon
• Container: on Dockstore and integrated with WDL; per variant lists co-occurrences, freq. Of cases, and freq. Of control

• not sufficient for re-ID w/o a high false positive rate; info BB Japan was comfortable sharing
• Sufficient for qualitative assessment of data by ENIGMA

• Next steps:

• Integrate container with add’l GA4GH workflow standards (WES), GA4GH phenotype standards
• Apply in more scenarios, with more cancer genes

Discussion

IF: Are any collaborators represented in the GA4GH cancer group already? If not, how do we bring them to the table?

This is exactly the kind of use case for federated workflow to run WES within GA4GH Connections (//FASP)

SP: tried something more complicated in ClinGen with whole series of queries, labs resistant to amount of work it took; how much work does it take on BBJ’s end?

MC: Obviously configuration is needed to integrate our container with their data, in a perfect world all data would already adhere to standards to simplify integration; in practice customization is needed, I don’t see that going away immediately;

The work it took on their end was to give us good description of their data and its format so we could adopt our container appropriately, they executed the container

Since this was a proof of concept, they executed multiple versions of container until we got the output that was assessed as meaningful

LS: Incredibly useful work. If you’ve already docker and WDL-ized the WF and have it on dockstore, should be possible to launch on any WES-supporting sites, eg., Terra; I’m not clear that that’s what your user community would want? Wouldn’t they want to run the docker container locally so it has access to their local data, vs. figuring out how to get local data onto Terra for example, so docker container can reach it. How is WES going to be used in this particular context?

MC: so far our 1 user of this container, has in fact simply wanted to run the container locally rather than on a cloud platform; maybe we don’t need WES if having a locally available container is good enough

IF: a key thing that seemed to be important was that compute needed to be done where the data was (in Japan); sounds like they ran it for you locally; you gave them the docker container and they ran it.

This hits a theme that we want to drive a lot of this from what we can do now, and ask ourselves the question how can that inform GA4GH standards and help us do better what we can do now; vision for something like this might be that you want to run WES there at Riken, remotely, and do that at other federated locations as well; that’s the vision - need to confirm whether that’s a useful vision.

Goes beyond everyone working with local data

LS: federated analysis / cloud based analysis makes most sense when you have a very large reference data set that not everyone wants to have a local copy of; eg., big copy of dbsnp or million cohorts dataset; or it’s very compute intensive task and local compute isn’t sufficient; or when there is a well-established ecosystem of multiple workflows working on top of a common set of standards and data and you don’t want to configure each of them locally. Maybe this is a first step.

HN: we BBJ and also Riken has very personal information, genotype, pedigree info, cancer, all info cannot be shared outside the country and outside of BBJ; so it is impossible to share that data in the cloud system, we are still discussing how to share such information in public cloud; BBJ is very conservative to sharing.

Pediatric intracranial germ cell tumor

Presenter: Anne Deslattes Mays

In the time of the genome project, more standardization than today

Data can’t move, 90+ genome projects happening world wide

In 2019, write a layer on top of google cloud platform, needed to know how to spin up machines; but reproducible, access TCGA, spin up docker VM

In 2020, we do this differently, have a platform agnostic nextflow workflow, aggregate data can be stored in Zenodo; from this you can reproduce the entire analysis from that point on, large compute against 9K gtex files needs to be done in cloud

New Gtex rules limits ability to access , not accessioning new seq data in SRA; ability to search for data using bioproject and biosample

Teaching people with no prior command line experience how to do this kind of analysis, teaching to trainees in cancer centers; “dry bench skills for the researcher”

Gone are the days when we can toss solutions over a fence, not everyone can be a system administrator.

We are solving problems now in FAIR and rigorous manners

SRA explorer - put in general terms, eg., cancer, cell line, rnaseq, some are access controlled, some not; I write a nextflow, access data in reproducible manner

We shouldn’t inhibit that

SDOs typically create standards, standards drive commerce, SDOs provide testing and certification

What is the test? WES is on terra? Is WES the standard? WES can write WDL, I write Nextflow

Use case

• MC’s case is exactly what we’re doing
• Playing a game of go-fish, I know what’s in my hand and Japan and other countries know what’s in their hands; just need to have “do you have xxx”
• Docker container shipping is state of the art; I’m sending nextflow workflows, has all config, adjust for multiple platforms; does not’ work on azure now, on prem at Jax,
• Intracranial germ cell cancer has very low incidence in the US, but it’s 5-8 fold greater incidence in Japan and others
• We’re testing the immunogenicity of the data; do we have, eg., HLA antigen presentation in the cancer?

Discussion

IF: we have a round table here, with people who may have data relevant for the use case; “go fish” sounds like Beacon; when we were looking for this particular variant, we looked around in Beacon. That kind of activity seems worth pursuing as a use case for Beacon, Search

ADM: also include Nextflow Workflow, using cloud OS, open data science platform, all tools are on github, can be run on any platform. We’re at point where tech for even on-prem data can be made cloud accessible, whatever cloud you’re on, using AWS outposts, equivalents

Separate auth/acc from datasets and workflows

I look to GA4GH for setting the standards and how to test for them, I expect companies I work with to adhere to those standards

SP: more clinical question, very rare tumor, original pub from a number of years ago when Ching was at Baylor; ADM: with his collaborators, going to datasets he’s blind to, send the NF WF for more samples we don’t have access to

IF: this is a rare cancer, crossover between RD and cancer?

DT: germline so many technicalities on genome analysis, number of ind’ls, BRCA so penetrant. Beginning alos of federated learning; sending a docker somewhere and get result is prob how we’ll have to do that; don’t know how far along we are on agreeing on one system (nextflow, wdl)

CY: in the perfect world that were’ moving toward, any workflow language won’t matter, engine can support all of them in any compute environments; how much work do we want to put in to make it happen; still feels a bit segregated that some engines can only run certain languages

LS: workflow won’t matter in long run, in short run it’s very difficult bc feature sets supported by wf managers is diff and challenging to port from one WF language to another; would be great to have a common language like post-script, running in background, we don’t worry about it. I think will be a while to get there and wouldn’t be on any of the horses right now, tech changes very quickly

ADM: except in Europe; US is very myopic toward WDL

LS: Nextflow is my strong preference, but didn’t come to my radar screen till about two years ago

Kids First Data Resource Center

Presenter: Allison Heath

Kids First is at intersection of RD and pediatric cancer, more and more is in that RD use case

Data resource started 2-3 years into start of program

Child cancer and structural birth defects

Increasingly we know that across developmental landscape, utility in cross analysing/ understanding these different diseases

Portal includes significant amount of data, 20 studies released in 2020

Some overlap between pediatric cancer and structural birth defects

Over 20K more participants coming through

Intracranial germ cell cancer coming in (ADM’s use case)

Down syndrome and pediatric cancer defects, lots of overlap

Way to have intersecting platforms and ecosystem, standards are key

For us the number of people looking t specific diseases is not enough; need people looking at other diseases; common disease, looking for interesting variant, more eyes looking at them

Genomic Workflows

• We do all WF in CWL, already a pain point of goal of having anyone bring their own WFs
• Working hard on the idea of functional equivalence and what that means
• Somatic WGS/WXS - tried to make it modular so if you want a piece in CWL only need to port the one you’re interested in; other techniques like that
• RNA-seq, mainly WGS, focus on family based cohorts, cancer tumor and rna seq as well
• NIH Common data ecosystem
• Interoperability across NIH resources, RAS in collaboration with DURI work stream, australia biocommons, elixir
• Lots of data in dbgap and SRA researchers want access to, how do we have right standards for DRS, passports, etc so people can use the tool they want to rather than the only one data is available in

Kids variant workbench

• Alpha stage
• Data continuum challenge, wide spectrum from those biobanking to those addressing clinical outcomes
• Different tools / standards support along that continuum
• Runs on AWS, open source things;
• Focused on annotations, pulling all this together in way that is useful for people was most work, tech, big data, platforms can handle that; but what do people really want to do with this, so when they show up in makes sense
• ID’d our own use cases for variant QC, filter out lower to focus on higher quality
• First question is “do other datasets have phenotypes I’m interested in?” AND “how do I do this in bulk?
• Our goal is really to focus on high level, what do people want to do commonly, how do people bring own datasets, stay aligned even if view differently, so can exchange under the hood

Other modalities

• Genomics but also increasingly proteomics, esp in pediatric cancer

• Rare tumors behave like other tumors but don’t have key variant

• Imagingg: files are there, but want a more integrative experience for folks wanting to dot hings like radio genomics

Clinical data flows

• Opportunity to use data to inform real time things like molecular tumor boards, how do we intersect these on behalf of kids with poor prognosis where time matters
• FHIR to connect to clinical systems, API to create apps and different tools, intersect research and clinical data flows lots of opps
• EMR, imaging, how do you get data out, de-ID, use in research
• NCPI FHIR working group
• Is this a path to interop with dbGaP?

Children’s Brain Tumor Network + PNOC (clincial trial arm)

• WG and RNAseq characterization
• Releasing PNOC integrating with landscape of CBTN data, what goes into clinical trial
• How do we do this and how do partner with the right standards
• Clustering uses CBTN as backdrop, see patients most similar, access in real time

• Zero Childhood Cancer (zerochildhoodcancer.org.au)

• Scaling up to sequence every child with childhood cancer in Australia, WGS and RNA seq

• International Pediatric Cures Project (ipc-project.eu)

• Federated learning, high level analytics

Final Discussion

• One thing we were looking for back at Plenary was what use cases could we take forward to 2021 plenary with demonstrable use of GA4GH standards to help the cancer problems we’ve talked about
• Need to process what we had here and work out next steps / action items / next meeting
• Things we haven’t even touched on, eg., cancer phenopackets initiatives
• ADM: Whatever we do must include Japan and all international partners we can think of, address this problem head on
• PG: the goal here is for the cancer community to do more together than it does as individual DPs with us, coming together to find areas of common interest
• Suggestion to meet more frequently, possibly at next Connect meeting (roughly 1 month away) or possibly before then

October 1, 2020

• Meeting Recording
• Slides
• Minutes