GA4GH Rare Disease Community Meeting

Agenda and Minutes 2023-2026

The GA4GH Rare Disease Community aims to promote global cooperation, data sharing and collaborative rare disease research through identifying the need for new standards, and/or implementing existing GA4GH standards

2026-06-24 - Data Collection for Rare Disease (Part 2)

2026-05-27 - Data Collection for Rare Disease (Part 1)

2026-02-24 - Rare Disease Phenotyping Challenges

2025-06-04 - HiFi Solves

Action Items

2025-05-07 - Monarch

2025-03-05 - TISLab

2025-02-05 - Phenotips

Rare Disease Risk Assessment & AI-Based HPO Extraction

UK NHS Integration

Matchmaker Exchange & Variant-Level Matching

2024-12-04 - ERDERA

2024-11-19 - RD Portfolio

2024-09-17 - GA4GH Connect: Towards a Rare Disease Portfolio

2024-08-07: RD Community Meeting

2024-06-26 - Mondo & DECIPHER

2024-04-22 - April Connect Session

2024-04-03 - Monarch KG + HPO

2024-02-29 - Grand Challenges in Rare Disease Webinar

2024-01-31

2023-11-22

2026-9- Follow up discussion after data collection presentations

Agenda:

1. Discuss previous presentations (GenomeConnect, Digital Cabinet, GlobalGenes, PhenoDB)

1. Possible community goal: Creating a standardized recommendation or a clear white paper example describing early, head-to-toe, thorough data collection, that is simple and patient useable

2026-06-24 - Data Collection for Rare Disease (Part 2)

Meeting Recording LINK

Passcode: @Maz$6ka

Attendees: Juliann Savatt, Ada Hamosh, Bryan Laraway, Andra Waagmeester, Elizabeth Terry, Gabriela Repetto, Hannah Blau, Heidi Rehm, Jonathan Berg, Julie McMurry, Lisa Chadwick, Maureen Mulia, Nicole Vasilevsky, Orion Buske, Sabrina Toro, Sharon Terry, Shiyu, Tony Brookes, Tzung-Chien Hsieh, Bradford Powell, Christine Von Raesfeld, Phillip Richmond, Mogomotsi Matshaba, Kshitiz Singh, Evan Connelly, Catherine Procknow

Meeting Summary: This meeting focused on reviewing data collection forms and methods for patient registries, with presentations from Genome Connect and Digital Cabinet. Juliann from Genome Connect presented their patient registry that has engaged over 14,000 participants since 2014, explaining how they collect genetic information through reported variants and health data through structured surveys that map to HPO terms. The surveys use skip logic to reduce burden and have a 72% completion rate, with participants able to provide updates annually and receive result updates back from the registry. Elizabeth from Digital Cabinet presented their secure, compliant platform that offers e-consent, customizable surveys, and roles-based access control for disease advocacy organizations and clinical trials. The platform includes features like conversational AI surveys, data de-identification, and mobile app development plans, with infrastructure owned by the company to ensure GDPR and HIPAA compliance. The discussion covered topics including participant recruitment, data governance, phenotype collection methods, and potential compensation models for data sharing, with participants expressing interest in both platforms and planning future discussions to determine next steps for their standardized data collection efforts.

GenomeConnect (Juliann Savatt)

• GenomeConnect is ClinGen’s online patient registry, launched in 2014 to engage people who have had genetic testing, regardless of diagnosis or result type. Participants consent online, in English or Spanish, to share genetic and health data with approved resources such as ClinVar, Matchmaker Exchange, ClinGen expert panels, and collaborating studies. The consent also enables recontact over time for additional phenotype questions, medical record requests, research opportunities, and potential variant/result updates.
• The registry collects two main types of structured information: genetic test results and health history. Genetic information comes from participant-uploaded reports or records requested from labs/clinicians, which are reviewed by GenomeConnect genetic counselors and assistants to extract reported variant details for structured sharing. Health data are collected through participant-completed surveys: a general head-to-toe survey using patient-friendly language, examples, and skip logic, with positive responses mapped to HPO terms on the back end. Participants are prompted annually to update responses, and the new platform seeds prior answers to reduce burden and improve longitudinal capture.
• The presentation emphasized that GenomeConnect is not intended to function like a large biobank for genome-wide discovery; its core value is adding case-level data around reported clinical variants. This includes unpublished cases, phenotype details, segregation information, and the ability to recontact participants when more detail is needed. Participants may also receive updates if variant classifications change, and they can connect with others based on shared genes or variants.

→ Discussion:

• Jules explained that GenomeConnect supports longitudinal updates through annual reminders and a newer platform that pre-populates prior responses, making it easier for participants to revise or add health information over time.
• Phenotype timing is captured through timestamps and follow-up questions about when a newly reported health issue began, though the depth of progression or severity data varies by survey.
• Participants can join if they have had genetic testing and can use the English or Spanish platform, though EU participation is currently limited because of GDPR. Recruitment mainly comes through lab reports, clinicians/genetic counselors, family referrals, search, social media, and other outreach.
• Jules clarified that GenomeConnect is not intended for broad complex-trait discovery; its main value is contributing case-level data, including phenotype and segregation details, to support variant interpretation and ClinVar submissions.
• The group also discussed whether LLMs could help analyze free text, improve phenotype capture, and reduce the manual workload involved in mapping participant responses.

Digital Cabinet (Elizabeth Terry)

• Digital Cabinet is a secure, participant-centered data platform for registries, distributed clinical trials, studies, and disease advocacy or research organizations. The platform was built around the principle that patients and participants should feel ownership and trust over their data. Its differentiators include strong privacy/security positioning, role-based access, compliance with HIPAA, GDPR, CCPA and related requirements, and ownership of both the software and physical infrastructure rather than relying on public cloud storage.
• The platform includes e-consent, survey design, data management, IRB/ethics workflows, administrator dashboards, and customizable branding for each organization. Users can participate as patients, caregivers, clinicians, researchers, or administrators, with each role seeing only what it is authorized to see. The e-consent process includes full consent text, plain-language summaries, digital signature, authentication, CAPTCHA-style validation, and voice-accessible options.
• A major focus was flexible, low-friction data collection. Digital Cabinet supports prerequisite surveys, branching logic, staged survey rollouts, file uploads, validated instruments, embedded videos, repeated/recursive surveys, and many question formats. It also includes conversational survey functionality through a controlled LLM process, intended to help participants complete surveys verbally or conversationally, especially where standard forms create barriers. Participants can delete all or specific data they entered, while administrators can access and export de-identified data, view aggregate dashboards, and move data to other systems such as REDCap without being locked into the platform.

→ Discussion:

• Elizabeth explained that Digital Cabinet is designed as a secure, participant-centered platform with role-based access, daily redundant backups, and infrastructure in San Diego and Dublin, with other data residency options possible.
• The platform is built to reduce participant burden through flexible consent and survey workflows, including staged surveys, branching logic, file uploads, validated instruments, and conversational survey options.
• Julie raised interest in computable e-consent and reusable consent approaches that could make participation easier across multiple studies while supporting interoperability.
• Elizabeth noted that mobile access is a major roadmap priority, especially for distributed clinical trials, participant-facing workflows, and easier return of results.
• The discussion connected Digital Cabinet back to the group’s broader needs: collecting standardized longitudinal phenotyping while maintaining privacy, participant control, accessibility, and the ability to export data into systems such as REDCap.

2026-05-27 - Data Collection for Rare Disease (Part 1)

Meeting Recording: Listen to the Zoom call Passcode: GkDFBf%7

Attendees: Tzung-Chien, Moni, Gabriela, Christine, Kahlil, Sarah, Sabrina, Julie, Brooke, Mrinal, Hannah, Emily, Hannah, Selena, Orion, Melissa, Zohreh, Ada, Andra, Nicole, Toyofumi, Ramona, Kaitlyn, Aimé, Catherine, Dianalee

RD Connect slides April 2026.pptx

Meeting Summary

The meeting focused on discussing tools for head-to-toe phenotyping in rare disease research. Zohreh presented RareX, a research platform of Global Genes that facilitates symptom-level data collection for rare diseases through structured surveys based on yes/no responses across body systems. Ada then demonstrated PhenoDB, a clinician-focused tool with 3,900 terms mapped to HPO that allows comprehensive phenotyping with 22 upper-level organ system categories, including both structural and functional assessments. The discussion covered key challenges including ontology mapping for terms with multiple examples, translation requirements for global implementation, and the need for better engagement from different stakeholders including patients, caregivers, and clinicians. Participants raised important questions about handling "unknown" responses, the difference between patient-reported and clinician-assessed phenotypes, and the potential for using facial image analysis for matching patients. The group acknowledged the need for patient-centered design approaches and discussed how to better communicate the value of comprehensive head-to-toe phenotyping to different user groups.

RareX Platform (Zohreh Talebizadeh)

• The RareX research platform was launched in 2021 by Global Genes. It facilitates data collection for rare diseases through symptom-based surveys
• Patient advocacy groups onboard participants and collect their feedback on the platform; Zohreh shared a recently published paper that illustrates this patient-driven approach
• Zohreh provided an overview of the RareX platform user experience, including account creation, consent, demographic information, and dashboard navigation.

• Head-to-toe-survey questions have yes/no or unsure responses across different body systems.
• RareX uses a branching logic system; Additional prioritized and validated surveys open up based on the user’s responses (e.g., answering yes to cardiac issues unlocks a cardiac-specific follow-up)
• The initial head-to-toe survey has a completion rate of about 50%; Follow-up sections have a completion rate of up to 85%.
• Zohreh highlighted the potential for RareX to capture unreported symptoms, longitudinal data, and cross-disease data for ultra-rare conditions;
• The group discussed language accessibility with Andra proposing Wikidata for translation purposes. Ada mentioned that HPO has already been translated into several languages, and noted that AI tools with human verification could help address this challenge.

PhenoDB (Ada Hamosh)

• Ada provided an overview of PhenoDB, a freely-accessible website for collecting and organizing patient phenotype information; The tool was built for clinicians, and designed to be completed in under 3 minutes.
• PhenoDB has a hierarchical format with 22 main categories that include both structural and functional components; The tool integrates HPO terms.
• Ada emphasized that the system helps clinicians capture comprehensive patient information and can flag potential genetic connections based on genotype data; In PhenoDB “unknown” indicates that the evaluator did not look at a particular feature; “Not abnormal” means the feature was evaluated and found normal.

Discussion

• The group discussed concerns about missing or incomplete annotations as they can affect analytical matching; Even non-abnormal findings could be meaningful for pattern recognition in certain cases.
• The discussion highlighted the differences between patient/caregiver self-reports and clinical assessments, with Melissa noting that previous research showed complementary and synergistic information between these sources; Patients also prefer survey formats over other modalities for symptom reporting.
• Ada emphasized the importance of systematic phenotype documentation to identify all potential issues rather than focusing on specific diagnosed conditions.
• Nicole suggested considering how patients present phenotypic information, noting that they may not think about symptoms in the same way clinicians do.
• Zohreh suggested developing age-specific use cases to demonstrate the value of comprehensive phenotype documentation to different stakeholders

• Tzung-Chien suggested integration of facial imaging and maintaing privacy through local analysis rather than sharing actual images.

Next Steps:

• Discuss how to organize the group and define goals ( e.g., minimum vs. comprehensive phenotyping standards, patient-centered design priorities

Journals:

• RARE-X a patient driven approach for collecting symptom and patient-reported outcome data in rare disease
• PhenoDB: a new web-based tool for the collection, storage, and analysis of phenotypic features
• Semantic integration of clinical laboratory tests from electronic health records for deep phenotyping and biomarker discovery

Links:

• Ontology of Biomedical Attributes
• Human Phenotype Internationalisation Effort

2026-02-24 - Rare Disease Phenotyping Challenges

Meeting Recording Listen to the Zoom call Passcode nW52%#.L

Meeting Slides: Phenotyping Challenges

Register for Connect HERE! The rare disease session will be April 16th at 2pm

Attendees: Ada, Kaitlyn, Christine, Ramona, Toyofumi, Zohreh, Heidi, Nicole, Adeepta Banarjee, Andra, Brooke Greinstein, Bryan Laraway, Catherine Procknow, Christine Von Raesfeld, Claudia Gonzaga-Jauregui, David Lagoce, Grant Wood, Jinny, Jonathan Berg, Jules Jacobsen, Julie McMurry, Karen, Kat Kratzer, Larry Babb, Les Biesecker, Lisa Chadwick, Lisa Matthews, Matthew Lynch, Monica Ferrie, Mousumi Ghosh, Mrinal Thomas, Norin Aman, Orion Buske, Sabrina Toro, Sarah Gehrke, Sarah Morton, Senkei Umehara, Solomon, Sonja Ramussen, Thushara Thomas,  Tzung-Chien Hsieh, Vanessa Vogel, Vibhor Gupta, Mónica Muñoz Torres, AC Desfaits, Danielle Boyce

Background: While computable standards (e.g., HPO, Phenopackets, Mondo) provide a strong foundation for rare disease phenotyping, challenges remain in capturing trajectory, severity, and granular laboratory/functional patterns - details that are often essential for rare disease interpretation. To work towards addressing this, we are planning a working session at GA4GH Connect in April to develop approaches for combining structured phenotypes with computable narrative elements, producing examples and early recommendations for potential product evolution. This community call will be an opportunity to align on shared challenges and experiences in rare disease phenotyping, and to refine the focus for the April Connect session.

Meeting Minutes:

Summary

The Rare Disease (RD) Community of Interest (CoI) meeting focused on improving rare disease phenotyping approaches so they are both clinically meaningful and computationally actionable. Participants discussed how current standards — including phenopackets, Human Phenotype Ontology (HPO), and Observational Medical Outcomes Partnership (OMOP) — often “flatten” clinical narratives, making it difficult to capture important contextual details such as time course, comorbidities, pertinent negatives, prior history, environment, and access to care. The group emphasized that effective phenotyping models must balance structured, scalable data with the richness of clinical narratives to better reflect real-world clinical reasoning.

Through discussion of use cases such as ALS, cardiomyopathies, and neurodevelopmental disorders, participants highlighted challenges in representing temporal progression, uncertainty, and nuanced phenotypic variation across patients. These discussions will inform a working session at GA4GH Connect in Montreal, where the community aims to explore practical approaches that better support collaboration between clinicians and bioinformaticians.

Phenopackets

• Phenopackets provides a structured representation of phenotypes including onset, severity, longitudinality, past medical history and more.
• However, two key gaps were flagged:

• Aggregation problem: “per-individual” makes it hard to compute disease-level patterns (core vs variable features, incidence, co-occurrence).
• Negatives problem: there’s no widely adopted, end-to-end mechanism for making pertinent negatives flow through tools and pipelines.

Human Phenotype Ontology Annotation (HPOA) files

• HPOA / cohort-style summaries are useful for grouping individuals under a disorder label, but they:

• Flatten clinical narrative and nuance
• Often omit co-occurrence patterns, disease progression, and clinical context (age, geography, access to care etc.)
• Generally do not encode what was assessed vs not assessed (missing denominators)

• Vignette-level detail is essential to preserve the “human picture” and clinical reasoning that tables alone fail to convey.
• Cohort curation approaches are one way to derive feature incidence and patterning, but the group was clear that incidence alone is not enough without timecourse/context.

Human Phenotype Ontology (HPO)

• Variations in phenotypic expression make exact term matching difficult
• Lack structured representation of co-morbidities, co-occurrence patterns, temporal dynamics (onset → progression → resolution), and group phenotype incidence
• Zohreh mentioned how patient reported outcomes (PRO)  are not easily mapped

• Unclear how to represent PROs in a structured ecosystem that interoperates with HPO and clinical data models.

Observational Medical Outcomes Partnership (OMOP)

• Danielle (ALS TDI) emphasized how integration of HPO and OMOP standards remains challenging and misses the detail needed for omics analysis and clinical relevance.
• Brooke (Epic) shared a concrete EHR implementation detail: while HPO terms can be documented, a status field (positive/negative/resolved) exists conceptually, but downstream tools often only accept positives.
• Sarah Morgan (ClinGen Heart Disease panel) emphasized that in practice, large amounts of relevant clinical data get discarded, even though it may be essential for interpretation and gene–disease validity.

• Sarah also provided an example of temporality/clinical nuance in congenital heart disease: what is “normal” vs “abnormal” can change by age (e.g., early-life findings that are physiologic vs pathologic), complicating coding.

• Jonathan Berg reinforced the same theme with a phenotype nuance example: “hearing loss” is not specific enough—episodic vs acute vs progressive, severity changes, and partial resolution are the details that matter and should be representable.

Disease Models

• Lack of comprehensive, structured, disease models

• No clear means of separating individual-level variation, core disease features and phenotypic disorder elements (and how they change over time)

• Questions to address

• How many patients are needed to define a disease?
• How do we represent mild variants and variable expressivity?
• How do we model uncertainty and lack of functional evidence (i.e., the role of only 5,000 protein-coding genes are known)?

• GPC is linking phenotypes to variants
• Sabrina and her group are working on disease models
• There is interest in building a disease “object” or “map” that inventories what’s known for a disorder

Tabular Phenotype Reporting

• Large phenotype tables have hundreds of patients x dozens of features, are often difficult to interpret and may not align with narrative case descriptions.
• Large phenotype tables lack denominators about feature assessment
• Tables fail to encode time course, co-occurrence, and clinical rationale.
• The group discussed bridging clinical reality and computation, building a structured inventory for a disorder that includes:

1. Phenotypes, genetic data, metabolic data, multi-omic, environmental info, etc.
2. Proper denominators
3. Negatives and comorbidities
4. Narrative vignette

RD-specific challenges

• Small patient numbers
• Large amounts of data are collected but not integrated (are discarded)
• Only a fraction of genes are well understood
• Genetic modifiers influence phenotype
• Complex conditions require nuanced representation
• Multi-omic data is not translating effectively to clinicians

Connect session

• Clarify what the current standards can already do
• Surface the key gaps clinicians keep hitting
• Test approaches on concrete disease examples
• Create actionable outputs

Next Steps:

• Engage clinicians and bioinformaticians from your networks, inviting them to Connect
• Send candidate use cases (disease + challenges) for the April Connect session, including:

• What data exists (EHR, tables, vignettes, omics, PROs)
• What representation fails (negatives, denominators, co-occurrence)
• What “success” would look like for clinicians + analysts

2025-06-04 - HiFi Solves

Meeting Recording 

Meeting Slides

Attendees: Marc Fiume, Maggie Clark, Angela Page, Arwa Farrukh, AC Desfaits, Ahmed Sherrif, Bradford Powell, Brooke Greenstein, Bryan Laraway, Catherine Procknow, Jillian Banfield, Julie McMurry, Kahlil Lawless, Katie Mullen, Mogomotsi Matshaba, Mónica Muñoz Torres, Sarah G, Yasmeen Kurdi, Zohreh Talebizadeh, Dalia Kasperaviciute, Ada Hamosh, Chloe Mighton

Background: HiFi Solves is a federated data network connecting renowned research institutions worldwide to build a real-time, globally representative database of rare genomic variants identified by HiFi long-read sequencing. Through this unique collaboration, HiFi Solves is accelerating innovation in the study of rare diseases and specific regions of the genome.

Meeting Minutes:

Action Items

• Present the Hi-Fi Solves project to the Matchmaker Exchange group.
• Consider how the Hi-Fi Solves platform could leverage the work being done in the GA4GH Genomic Knowledge Standards Work Stream.
• Share the recording of this meeting with the broader Rare Disease Community of Interest mailing list.
• Explore opportunities to collaborate with the Federated gnomAD project and the Variant Level Matching (VLM) project.
• Provide feedback on the types of questions and insights the rare disease community would like to see enabled on the Hi-Fi Solves network.
• Send save the date for the September brainstorming session.

Welcome & Housekeeping Items

• GA4GH is in the process of appointing new co-chair(s) for the community of interest.

• Angela discusses the pause in the Rare Disease meetings for the summer to regroup and think about the structure and strategy of the community of interest.
• The idea of turning talks into a resource library for the community is proposed.
• Angela mentions a save the date for a brainstorming session in the fall to discuss the next phase of the community.

Introduction to HiFi solves

• Marc Fiume, CEO of DNA Stack, introduces Maggie Clark, Director of Partners and Programs at DNA Stack, who has contributed significantly to HiFi solves.
• Marc explains the vision of HiFi solves as a global federated network to accelerate collaborative rare disease research.
• The project builds on earlier efforts like Beacon Network and Matchmaker Exchange, aiming to create simple push-button tools for data sharing.
• Marc emphasizes the importance of interoperable standards and the evolution of HiFi solves from earlier projects.

Overview of Omics AI Suite

• Marc introduces the Omics AI Suite, which includes Publisher, Explorer, and Workbench.

• Publisher is a data governance studio for connecting, organizing, protecting, and sharing data.
• Explorer is a federated data portal for discovering and accessing data sets.
• Workbench is a standards-compliant environment for conducting scientific analysis.

HiFi Solves and Its Implementation

• Marc explains the goals of HiFi Solves, including the use of GA4GH standards for data representation, discovery, and analysis.
• HiFi Solves is sponsored by PacBio, with all software and cloud computing costs covered by PacBio.
• The project aims to connect and make discoveries across global federated networks, focusing on complex variations identified by PacBio sequencers.
• Marc highlights the importance of privacy-preserving insights and the ability to share aggregate statistics without sharing raw data.

Day-to-Day Operations and Data Sharing

• Marc describes the day-to-day operations of connecting data to the HiFi solves network using Workbench and Publisher.
• The process involves running best practices pipelines and connecting data sources to the network.
• The importance of data sovereignty and the ability to install software components within partners' environments is emphasized.
• Marc discusses the technical challenges and benefits of federated data sharing.

Questions and Insights from the Network

• Marc demonstrates how researchers can use HiFi Solves to drive insights by executing SQL queries across the network.
• The platform allows for the sharing of allele frequencies and other insights without sharing raw data.
• Marc invites feedback on the types of questions the rare disease community would like to see available on the network.
• The potential for integrating phenotype data and using large language models for interpretation is discussed.

Feedback, Discussion & Future Directions Summary:

• Ada Hamosh suggests presenting the HiFi solves project to the Matchmaker Exchange group for additional input.
• Angela and Marc discuss the overlap with other projects like federated gnomAD and the Variant-Level Matching (VLM) project.
• The importance of harmonizing data across the network and the potential for using phenotype data is highlighted.

• Marc outlines the roadmap for HiFi solves, including the adoption of phenotype data and the implementation of matchmaker questions.
• The potential for using AI to support the final steps of variant interpretation is discussed.
• Angela emphasizes the importance of feedback from the community to refine the project.
• The meeting concludes with a plan to share the recording with the broader community and reconvene in the fall for further discussions.

Chat & Discussion:

• Ada: Could this be presented to the MME (Matchmaker Exchange) group?

• Marc: Yes

• Angela: Is this connected with federated gnomAD or the VLM project?

• Marc: No connection w/ federated gnomAD but would love to understand if we could use the technology. Same with VLM- would love to contribute in any way to help grow those networks

• Zohreh: Do you have any workflow for integrating availability or leveraging of a different type of patient-reported outcome data?

• Marc: The technology can adopt them, although we are not currently sharing that kind of data

• Marc: To encourage and enable data discovery, it's important to help researchers understand what data exists and what segments of a cohort might be relevant to their work. One example of this in action is with Target ALS, a partner organization focused on ALS research. They developed a dashboard that uses questions to generate cohort-level aggregate insights, allowing users to explore the relevance of available datasets before requesting access. Within just a few months of launch, hundreds of researchers had explored the data assets, and over 300 participants have now submitted access requests 30% of whom come from industry. Feedback from data users has been overwhelmingly positive, especially around how streamlined the discovery and access process is. This approach addresses a major challenge many data custodians face: a sharp drop-off between people visiting their site and those actually requesting access. By making it easier to explore what data is available and relevant, Target ALS has significantly improved engagement and utility of their datasets.
• Moni: Would you be okay sharing a little bit more about that level of granularity that researchers will have when they're trying to define a cohort? So, will they be built specifically on a gene, or variant types, or zygosity, specific HPO terms, or family history?

• Marc: The technology allows you to connect both genomic files and metadata tables about subjects. These tables are exposed as Data Connect endpoints- once search access is granted, users can explore all the associated data. On top of that, there's a second layer called "Questions," which lets data custodians define aggregate-level queries; essentially, SQL-based cohort queries. These can be used to create predefined, aggregative insights that others can use to explore the dataset without needing full access.

• Angela: Is the vision for this platform to serve as a solution for already-established consortia with harmonized data? Or is it intended as a broader network that anyone can join, requiring participants to bring their data up to a shared standard? In other words, is this a tool for existing consortia- or is the goal to build a global consortium through the platform itself?

• Marc: Yeah, we learned from Beacon the hard way that it was unrealistic to expect people to harmonize their data as the first step.  So we wanted to create a first step, that is a kind of come-all approach, regardless of however you've formatted your data, you can still connect to this network. Step 2 is really valuable to drive consensus across the collaborators in terms of how they want to process and collect their data, but that's not a requirement. We explicitly made that not a requirement to participate.
• Angela: And is that partly possible because of Data Connect and the way that Data Connect makes that possible?
• Marc: Yeah, Data Connect is essentially SQL on your data. So, if someone has data stored in one format, and another has variants stored in a different format, when you ask the question, each one can answer with their own implementation of SQL. But it's up to the implementer to interpret the question and respond following the standard protocol. We don't have to change the physical layer- how data is stored. We just have to change the logical layer, how the questions are answered and so, Data Connect is the engine that allows us to drive that.

• Moni: For your plans to incorporate MME, will consortium members be able to submit a cohort definition (say, patients with variants in gene X and phenotype Y) to the MME through HiFi Solves to find similar cohorts elsewhere?

• Marc: Similar to the allele frequency question, we imagine an input that allows you to put in your genes and phenotypes, and it would respond with a list of orgs that have matches.

• Marc: In a world where we have the ability to compute locally, both with Data Connect and with workflow execution, so we can do simple/complex processing, where could we take this? Is the dream scenario to build a zero-sided matchmaker where we're automatically identifying matches across the network? If we dream about where this can go, what would your magic wand do and create?

2025-05-07 - Monarch

Meeting Recording 

Presenter Slides 

Attendees: Nico Matentzoglu, Nicole Vasilevsky, Angela Page, Arwa Farrukh, Ada Hamosh, AC Desfaits, Beatrice Amos, Bryan Laraway, Catherine Procknow, Claudia Gonzaga-Jauregui, Chole Mighton, Jillian Banfield, Kahlil Lawless, MeeSiing Ngu, Moni Muñoz Torres, Marc (Orphanet Inserm), Sabrina Toro, Sarah G, Sarah Morton, Takatsuki Terue, Tony Brookes, Trish Whetzel, Zohreh Talebizadeh

Presentation Description: Rare diseases are among the most complex medical challenges and undeniably require global coordination. Terminologies and ontologies act as hubs, connecting information generated by a wide range of often isolated resources, each containing crucial insights about individual diseases. Many of these resources use their own terminologies or standard vocabularies to describe rare disease data, making the development of an integrated "ontology of terminologies" all the more vital. However, managing such a resource is challenging and demands the integration of diverse communities and perspectives on rare diseases. I will discuss how the Mondo team rises to this challenge through community curation, incorporation of externally managed content, and the management of rare disease designations.

Minutes:

• Looking into getting a new Rare Disease co-chair in next month or so

• Brainstorming what’s next for the rare disease community and how to best engage with the community

• Moni: Let’s work on that end-to-end case demonstration of GA4GH resources for diagnosing RDs

• Tudor and Neerjah have stepped down
• Marc Fiume to present June 4th: HiFi Solves (https://hifisolves.org)
• Think about what you want to get out of this rare disease community and from GA4GH

• Ex: other communities of interest have different formats where they have working groups that work on specific problems. Then they only have 2 meetings a year where they come together and share

• Nico Matentzoglu Presentation: Towards collaborative, decentralised ontology management for rare disease

• In a global challenge like Rare Disease, there will always be many different standards to describe disease data
• To connect all information together, we need

• Reusable Mappings
• An ontology of terminologies and ontologies
• A transparent collaboration culture which

• Welcomes and prioritises community contributions
• Enables community-managed content

• Groups like GA4GH are well positioned to encourage cross community collaborative editorial workflows

Q&A:

Nico: Any attempt to go and try to create a one ring solution is going to hamper our overall progress. We need to invest in solutions that allow us to work in a multi-ring environment very effectively and that's why I think especially investing into standardizing our mapping workflows, maybe even teaching each other the big question like what is an exact mapping? This could be the content of a great workshop even just to sort this out between our groups and if we can get to one point together and curate our mappings to a kind of common methodology, that would be a much bigger impact for a decentralized community than trying to impose a single ontology here, you have to use mondo, you have to- this is never going to work

Tony: I wonder if the underlying terminology semantics standards are now getting where they're getting so powerful that people, we're not providing the tools and the capabilities for people to easily use them to maximum effect. And I wonder if we're creating a gap^[a], if you see what I mean by taking the technology so far forward so fast. Maybe I'm wrong on that. It's just that within EJP, that was the sense I got- I'd like to ask that and pose that as a question to all of you.

Ada: If there is a way for whatever we're using, whatever works in whatever system, it actually be able to incorporate something that lets us get down to a granular level. Then we need some mechanism- you have to be able to find it, code it, and find that patient again.

How do we get these levels of granularity into and out of health care records?

Nico: how do we get these terminologies into Epic?

Melissa is trying to address with work with IMO Health

Moni: noted Heidi’s perspective - there is a desire for an integrative approach to help how rare diseases are diagnosed

Ada noted we need to get info into proper hands

Thinking about Mondo, there needs to be data harmonization

If we think about a framework for rare disease diagnosis- Mondo brings knowledge integration later/semantic integration, can integrate external knowledge bases

Can Mondo help give clinical validation?

What other places can Mondo be leveraged in our framework? How does it integrate with our other resources?

• Ada: lots of excellent work being done, from a GA4GH perspective- this community
• needs more clinicians. We need clinicians to validate the annotations and provide feedback
• Whatever we do today will probably be wrong later but that’s okay
• For a person who wants to know everything about a particular disease - they need different information than someone who is trying to diagnose a pt

• This level needs to be extractable

• Marc: Orphanet faces the same issue

• There are several issues to solve
• Quality of the dataset is a consideration

• Tony: We've got an exactly analogous problem with use conditions. We have a standard, we have ontologies and so on. We're using AI to take spoken text and written text around and use conditions to map that into that ontology and that data structure. It could be relevant here and help clinicians deliver what we need.

Chat:

Zohreh: Addressing a complex and evolving problem like this requires an iterative approach and mixed methods, and what you presented is a great example of that.

Moni: Aaaaaha!

Tony puts his finger on the issue we’ve been discussing on ‘handing GA4GH resources into the hands of the clinics.’ These two themes (RDs and clinical applications) are obviously connected.

Tony: In the Solve-RD program, most cases were annotated by <5 HPO terms  :-

Nico: What can we do to get more clinicians involved? I can only think of one thing.. Paying them 😛 For example as consultants.

Zohreh: Focusing on real-world clinical use cases, mapping the problem across all these integrated frameworks, and identifying a practical workflow for implementation in clinical settings is perhaps one approach.

Tony is using AI

2025-03-05 - TISLab

Meeting Recording

Presenter Slides

Attendees: Julie McMurry, Kahlil Lawless, Dilza Campos, Senkei Umehara, Karen Rothfels, Jonathan Berg, Shahim Essaid, Erinija Pranckeviciene, Sara G, Yasmeen Kurdi, Heidi Rehm, Dalia Kasperaviciute, Jillian Banfield, Sergi Beltran, Tim Heffron, Tony Brookes, Zohreh Talebizadeh, Beatrice Amos, Bryan Laraway, Kayla Socarras, Claudia Gonzaga-Jauregui, Neerjah Skantharajah

• What are the bottleneck in the diagnostic journey?
• Community white paper was put together to answer this: https://zenodo.org/records/14906644
• Dynamic vision for this virtuous cycle of learning healthcare system hinges on getting phenotypes right
• Without evidence, we can’t have evidence-based care
• Phenotyping in 2025 is not so different from 1907

• Reason: most of healthcare using flat lists, taxonomies e.g. ICD
• What we need: multi- hierarchy, multimodal graph based representation

• HPO

• Some of the clinical term has a patient accessible term that the patient can use to describe themselves
• It’s not complete, but it’s leveraging as much as the graph as possible

• Phenotype assisted diagnosis in action

• 2 patient have the same rare disease but present differently

• Patients and families should be empowered to work with us, however

• Patients are starting to have less and less time with physicians
• Struggled with continuity
• Patient reported information that has been traditionally accepted

• Phenotyping isn’t free!
• We need to know how much phenotyping is enough. What is the most valuable given what’s been entered. Whether something is negated or asserted will maximize annotation sufficiency
• HPO has a lot more coverage than Genome Connect
• Asked patients to use both tools to see the effect
• Use case: Wils’ tumor patients
• Diagnostic power is higher with layperson HPO
• In the future we should blend phenotyping power with specific questions, and have smarter ways of skipping questions more appropriately
• N3C: largest national public HIPAA-limited data set in US history
• The principles of harmonizing data and being able to analyse that information at scale can be applied to rare disease
• What can be ascertained from the EHR?: https://ojrd.biomedcentral.com/articles/10.1186/s13023-021-02061-3
• We need common modeling for diseases and phenotypes facilitate interoperability across domains and around the world
• We need all organisms to understand biology and human health
• Real world data is messy, but if you harmonize it and have a lot

Q&A

JB: Typically patients come in with concerns and so computationally it wouldn’t be hard to draw them through questions that would help to whittle it down to more information. Is that what you have planned for a patient/parent facing tool?

JM: Yes, that’s right. We’ve made mockups for these types of tools, but it’s about having the opportunity and grant funding to do it. Melissa and I are part of a start up whose aspirations include fixing this problem.

TB: AI? How many phenotypes would you need for this approach to work?

JM:There’s a lot of buzz around AI in rare disease. AI’s strong suit is not rare anything. The reason it’s good at pattern matching is because it has the luxury of trillions labeled data on cats (for example). See: https://www.youtube.com/watch?v=160F8F8mXlo

AH: The concept of physics in medicine is beautiful but we don’t know enough and there’s too much noise. Rare disease patients can present so differently

DC: Patients providing their own phenotypes is important, but many times the doctors don’t believe what the patients say.

JM: This is why initial phenotyping is so important because it can prompt the physician to ask the right questions.

TB: Doing the biochemical test is black and white so is the ultimate goal to determine those critical test and signals? Is that where this could lead us?

JM: Cautious yes. Resources are finite. Not just for chemical test, but also patient phenotyping - their time is a resource and we need to make sure the surveys they are completing isn’t a waste of time.Right now phenotyping isn’t anywhere in this landscape.

There’s an awareness problem in the reviewer community about rare disease. Amplifying that rare disease is a thing and it’s not something to be brushed away. That information needs to be relayed. Additionally, it needs to be clear that this is not a solved problem. We got reviews back that said WGS is standard of care, everyone has access to WGS. Something we can all do that this is not a solved problem and be responsible reviewers on panels.

HR: EPIC is launching better interactive phenotyping tools for physicians and intends to help consume the EHR record, deliver it to a physician and have them endorse (or not) that packet, send it to a clinical lab and have inability to weigh in on candidate diagnoses. Given the effort they are putting in this space, do you think asking them or a similar vendor would be willing to add this type of thing to their engagement with patients?

JM: Yes, that’s why we’re on calls with them. We’ve asked them about the patient side of that and their response has been non-commital.

TB: The ERN network in Europe have their CPMS system (clinical Patient Management System), to enter rich patient descriptions and seek help with diagnoses. It sounds like your precision phenotyping approach might be useful there?

2025-02-05 - Phenotips

Meeting Recording

Presenter Slides

Attendees: Neerjah Skantharajah, Nicholas Owens, Orion Buske, Tudor Groza, Greg Costain, Nicole Vasilevsky, Bryan Laraway, Zohreh Talebizah, Heidi Rehm, Kayla Soccarras,  Nico Matentzoglu, Julie McMurry, Larry Babb ,Leslie Matalonga, Giselle Kerry, Sarah Morton, Claudia Gonzaga-Jauregui

Minutes

Actions and Upcoming Events

• Register here for the Rare Disease Day webinar - Advancing Rare Disease Knowledge through Global Collaboration
• Register here for the Phenotips Speaker Series - The Undiagnosed Hackathon: Driving Diagnosis Through Global Collaboration
• Register here for GA4GH Connect taking place from April 1-4 in Boston, Massachusetts

Introduction

• The January RD meeting was a joint session with the Clin/Pheno WS (see meeting notes here)
• Sean O’Neil provided an overview of methods to evaluate large language models
• Potential formation of working group within RD and Clin/Pheno communities to further explore this area
• Next step is a Connect session on this topic in April in Boston

Phenotips
Presented by Orion Buske

• Phenotips integrates with hospital electronic health records (EHRs) to streamline genetic testing and risk assessment.
• Focus areas: pedigree management, structured phenotyping, genetic test integrations, and workflow optimization.
• Key advantage: supporting structured phenotype data collection for clinical and research use.

Rare Disease Risk Assessment & AI-Based HPO Extraction

• AI-driven tools for Human Phenotype Ontology (HPO) term extraction are improving phenotype accuracy.
• LLMs show superior performance in recognizing implicit phenotypes, particularly in neurodevelopmental and behavioral conditions.
• Challenges remain in integrating AI-derived phenotype data with SNOMED, OMOP, and ICD-10.

UK NHS Integration

• Phenotips is actively integrating with NHS genomic services.
• Supports multi-disciplinary team (MDT) workflows, patient-reported data collection, and genetic test ordering.
• Upcoming developments include structured phenotype and genetic data exchange using HL7 FHIR and Phenopacket API.

Matchmaker Exchange & Variant-Level Matching

• Phenotips contributes to the Matchmaker Exchange (MME) for gene-phenotype matching.
• A new initiative led by Heidi, Nara, and Jessica aims to enable variant-level matching within MME.
• Discussion on how Phenotips could align with these efforts.

Undiagnosed Hackathon

• An annual event where clinicians and bioinformaticians collaborate to solve undiagnosed rare disease cases in real-time.
• Past successes:

• 2023 (Sweden): Diagnosed 4 out of 11 cases.
• 2024 (Netherlands): Diagnosed 10 out of 21 cases.

• The 2025 edition will take place at Mayo Clinic in September.
• Orion is hosting a panel at the end of February with key organizers to discuss outcomes and future plans.

Key Questions & Discussions:

• OMOP Interoperability: Need for better integration between OMOP (research data model) and Phenopackets (clinical interoperability standard).
• SNOMED & ICD-10 Mapping to HPO: Monarch Initiative is working on mapping, with an initial subset to be released soon.
• Future Talks: Orion invited to present in the VLM group on one-sided variant matching.

2024-12-04 - ERDERA

Meeting Recording 

Attendees: Catherine Procknow (EPIC), Sergi Beltran (ERDERA), Erinijia (Vilnius University), Brooke Greenstein, Sarah Morton (Boston Children’s), Aida Falcon, Zohreh Talebizadeh (Global Genes), Karmen Trzupek (Global Genes), Kayla Socarras, Claudia Gonzaga-Jauregui (UNAM, CEPAL), Jillian Banfield (CIHR), Kahlil Lawless (Illumina), Monica Munoz-Torres, Nicholas Owen, Nicole Vasilevsky, Yasmeen Kurdi, Neerjah Skantharajah, Beatrice Amos, Julie McMurry

Minutes (to be added)

Next upcoming community call

📅 On January 15th, 2025 at 16:00 UTC, we’ll be hosting a collaborative joint call with the Clin/Pheno Work Stream! This session will explore benchmarking exercises to evaluate the use of LLMs in Rare Disease diagnostics and determine how effectively LLMs can organise knowledge and support diagnostic processes.

We invite the GA4GH community to help shape this initiative and address these key questions:

• What datasets should we use?
• How can AI tools best support Rare Disease diagnostics?
• What are the key metrics for evaluating LLM performance with knowledge graphs?

European Rare Diseases Research Alliance (ERDERA) Overview - Sergi Beltran 

• ERDERA continues and expand the work of the European Joint Programme for Rare Diseases (EJPRD) & other initiatives related to data infrastructure & patient needs in the RD space

• EJPRD concluded on August 31st, 2024, it’s main objective was to create a research & innovation pipeline “from bench to bedside”, ensuring rapid translation of research results into clinical applications & uptake in healthcare
• Their virtual platform (second level access) employed Beacon to query resources

• ERDERA is active beyond just Europe, with 37 participating countries, a 100 million euro budget, 130 participating institutions

• Including 24 European Reference Networks (ERNS), each focusing on a specific RD, providing clinical & research expertise

• ERDERA contains 3 primary objectives

1. diagnosis established or enrolment in systematic research in average within 6 months after coming to medical attention
2. New effective therapies for RD approved in Europe + beyond the majority of which addresses diseases without approved options
3. Better understanding of the impact of RDs on patients, families, & society to improve quality of life

• ERDERA contains 25 work packages, across different domains including 1) coordination & strategy, 2) RD Funding, 3) Clinical Research Network, 4) Inter(national) capacity alignment [where GA4GH collaborations fall], among others

• Clinical Research Network - Diagnostic research

• This network has 3 work packages: Diagnostic Data Availability, Genome re-analysis research pipeline, Genomic innovation to shorten time to diagnosis
• Main aim is to bring together data (exomes & genomes) that have not yet been diagnosed in either a distributed or federated approach to crack these cases
• WP6: concerns data organization & harmonization, with the input data including genomic/metadata/clinical/phenotypic/pedigree data for comprehensive interpretation

• This data is harmonized with ontologies & standards (OMIM, HPO Phenopackets, etc)
• This data then goes through the data analysis & data interpretation task forces (clinical experts)
• Whatever is successful is translated to implementation guidelines for the community
• This effort is intended to span over 7 years, with the initial 3 years having secured funding

Moving from distributed to federated approaches

• Federated approaches will help to

• lessen the burden of centralizing large quantities of data
• work around technical/legal/ethical challenges of genomic data sharing
• access data that is generated from healthcare and not research

• The goal is to harmonize data at the source, by leveraging GA4GH standards and European data infrastructure in development

Diagnostic research workstream

• Largely based on SolveRD, a project that recently wrapped in 2024
• Working with 48 partners across 22 countries, they aim to house 100k datasets by the end of ERDERA

• These patients are still undiagnosed despite having data generated

• Data Services hub within ERDERA > includes several work packages (finding & accessing data ecosystem, data readiness services, data sharing & analysis services, knowledge bases/ontologies/semantic models for RD research)
• Distributed analysis workflow (leveraging Solve-RD data infrastructure)

• Data will first be harmonized in a standardized way through RD-Connect GPAP (genome phenome analysis platform) developed at CNAG
• Data is then deposited at EGA for long-term archival (phenopackets, PED files, FASTQ, BAM, gVCF, transcriptomics, epigenomics, proteomics, ect )
• To control the data where it sits, RD3 database is employed

• Expertise Services Hub 

• Provides regulatory support akin to GA4GH REWS + methodological support (statistics)

• Education & training in RD research

• Provides training program for patients across the RD research pipeline, in addition to training to empower RD diagnostic/clinical/data research
• EU curriculum on RD research aims to establish a higher education program culminating in an academic diploma with credits
• (Inter)national capacity alignment

• Positioned to participate with other global initiatives like GA4GH
• Contains 3 primary tasks: 1) Nation Mirror Groups promotion & national alignment, 2) fostering engagement of underrepresented countries, 3) ERDERA Global collaboration

Discussion + Q&A

CGJ: Mexico is trying to join efforts in Latin America similar to the RD networks established in Europe. Will the efforts of ERDERA be Europe specific or include other parts of the globe?

SB: we are open to other experts in the world, provided we are meeting legal considerations

NS: considering becoming a GA4GH GIF project may be helpful to assist with the development of implementation guidelines and to gain visibility in the GA4GH network

SB: A lot of the groups participating in ERDERA are familiar with GA4GH. One challenge experienced in EDERA was trying to implement standards like Beacon to non-genomic data

NV: you mentioned you’re using ontologies like OMIM for disease annotations. Have you considered using Mondo at all? Mondo has a comprehensive coverage of RD in the ontology.

SB: I’m not very familiar with it and need to look into it more. I’m not sure how much they overlap but we’ve been working with Orphanet for many years.

NV: Mondo is co-developed with the HPO and integrates 17 disease terminologies, including Orphanet and OMIM. Mondo aims to be more comprehensive in covering RD outside of just Europe, which is Orphanet’s speciality.

YK: Connection re: Beacon network in EJPRD. The Zero Childhood Cancer Program in Australia experienced similar limitations with Beacon endpoints (non genomic data). To overcome these limitations, they built their own Beacon UI from the ground up and the documentation is supposed to be open source and available soon.

2024-11-19 - RD Portfolio

Meeting Recording

Attendees: Jillian Banfield, Jaime Delgado, Beatrice Amos, Moni Munoz-Torres, Claudia Gonzaga-Jauregui, Tudor Groza, Nicole Vasilevsky, Diya Uberoi, Heather, Zohreh Talebizadeh, Jaime Delgado, Sergi Beltran, Bryan Laraway, Dalia Kasperaviciute, Marc Fiume, Aime Lumaka, Senkei Umehara, Jessica Seegobin, Neerjah Skantharajah, Yasmeen Kurdi, Ada Hamosh

Minutes (to be expanded & circulated)

Rare Disease Portfolio

• Most GA4GH standards are applicable to the rare disease domain but this portfolio attempts to piece the products & standards together in a coherent picture that’s applicable to a particular context  
• Tudor gave a quick recap of the RD Connect session in Melbourne (see link in agenda)
• Tudor then walked through the current working draft of the portfolio

Discussion with attendees

Moni: what I envision is figuring out how we answer that question. Is there a way to generate a tool or pipeline that leverages all the products? One tangible product would be able to guide that answer. Is there a workflow we can offer the RD community through this portfolio?

Sergi: one need in the community is how to best apply/implement these standards. For example just because 2 initiatives use Beacon, doesn’t mean they can query for the same things. Coming to some consensus would be helpful so that we can work interoperably.

• An inventory of current implementations on RDs is a good starting point.

• GA4GH’s current implementation registry can be accessed here.

• GA4GH tech team is currently working to updating this registry into the new year, which can be used as a resource for the portfolio

Jaime: complementary view would be useful - including some requirements from RD use cases would be helpful to limit the scope of the portfolio

Diya: From a REWS perspective, could we put together some sort of roadmap or guidance document that would outline the use of this portfolio and the actors involved. While GA4GH consent clauses for example are general, we need to consider the extent to which our recommendations are applicable to different jurisdictions.

Clin/Pheno Work Stream & Rare Disease COI joint meeting: January 15th 2025

• This session will focus on a benchmarking exercise designed to evaluate the use of Large Language Models (LLMs) in rare disease diagnostics. The goal is to determine how effectively LLMs can organize knowledge and support diagnostic processes, enabling the development of tools that leverage these advanced AI systems.

• We aim to involve the community in shaping this initiative by addressing key questions:

• What datasets should we use?
• How can AI tools best support rare disease diagnostics?
• What are the key metrics for evaluating LLM performance with knowledge graphs?

• One proposal is to design benchmarking data for systems that query a knowledge base (or graph) with an AI component, helping us assess how well LLMs use structured data to answer specific questions. This could include tasks such as mapping phenotypes to potential rare disease names.
• Stay tuned for meeting details that will be shared via email!

2024-09-17 - GA4GH Connect: Towards a Rare Disease Portfolio

GA4GH Connect September 2024 - Rare Disease Community Agenda.docx

2024-08-07: RD Community Meeting

Meeting Recording

Attendees: Ramona Walls, Zohreh Talebizadeh, Mallory Freeberg, Kahlil Lawless, Monica Munoz-Torres, Tony Brookes, Julia Foreman, Sabrina Toro, Bryan Laraway, Kayla Socarras

Intro Slides

Minutes

Sli.do Results

How can members contribute to the RD Community?

• Join us at monthly calls
• Present your work
• Spearhead a study group
• Join an emerging study group

What is the best next step we can take to collect use cases related to data sharing practices and tooling?

• A combination of targeted tech session + focus groups is a good idea

Managing Sharing and Using Data to Create Solutions for Rare Diseases
Presenter: Ramona Walls

What is C-Path?

• Critical Path Institute catalyzes innovation and partner in medical product development worldwide
• Operates on a consortium based model
• Develop drug development tools to accelerate medical product development (e.g.clinical trial simulation tool)
• CPath is a public-private partnership which allows it to act as a unique neutral convener to bring together pharmaceutical companies, patient partners, academics researchers etc.
• Aims to overcomes the valley of death within drug development
• Working with data is foundational to everything CPath does, but we lack good natural history data, especially in rare diseases
• Trustworthy patient level data are rare, unstandardized, and siloed

How does CPath Work?

At a high-level…

• De-identified study data goes into CPath, they curate and standardize the data and put it into an integrated database which can then inform the development of actionable medical product development tools

Getting deeper…

• Input data include clinical trial data, natural history, patient registry, real-world data etc.
• Data is ingested which involves data contribution agreements, secure data transfers
• We do responsive curation where we create data dictionaries, map data to standards, load it to databases, and publish it to a platform
• We integrate the data into a knowledge graph using external ontologies
• The data is published to the RDCA platform (data and analytics platform)
• Many of these areas are manual including the mapping and analysis work
• The CPath team is looking to automate some of these processes, which amy involve the use of AI
• Automation is especially necessary now as their is a huge increase in patient level data
• Standardized vocabularies are no longer enough, now that we are getting data that isn’t always clinical
• Portal for query and access request - how to query that without security risks is a consideration
• GA4GH standards are not currently used but just now wrapping up with a consultant to determine a way forward for working with genomic data so GA4GH will become more important then - Nicole will keep GA4GH and CPath connected!

RDCA-DAP: https://portal.rdca.c-path.org/

• Develop use cases, build platform and ingest data
• Serves the entire RD Community
• Have had rare disease and orphan consortia within CPath
• We now have an ALS project and new RD consortia coming in
• Provides home for all different RD Programs within CPath
• Workflow involves data ingestion, curation and data management, fair data services, and research workspaces
• RDCA-DAP leads task forces covering disease within different areas of the body including brain, liver, bones, kidney etc.
• Success of the platform: Use the shared database to support the development of the first drug ever created for Friedreich's Ataxia!

Q&A

TB: To what extent is this data being used?

RW: From the consortia, the data we get in has all been used. We only take in data once we have a consortia that is ready to use the data. RDCA Dap is a new model where we’re acting as a repository. We are actually certified as a NIH trusted repository. People are coming and using it but don’t have the number of uses, in the high hundreds. The number of workspaces is ~50. We don’t know exactly what people are doing unless they come and tell us. The task forces are important because it illustrates what we can do with the data, and hopefully that would drive adoption.

TB: You can start doing analysis of the users and their data use behaviors. E.g. What is being sued and what isn't? That information can lead to more attention going to certain creation efforts

RW: Agree! Another thing we are looking at is what are the characteristics of the datasets that have led to success? We are hopeful that we can use some AI methods to get at that question.

DECIPHER: Sharing rare disease phenotype-linked variant data for diagnosis and research (to be updated)

DECIPHER supports data sharing and linking

• Decipher platform is meant to be a resources where information can be deposited safely
• Takes individual location records and combine it with different annotations
• DECIPHER is one of the platforms that is part of Matchmaker Exchange - a federated discovery network that allows for query of variants
• There are over 49000 anonymous records that can be viewed in aggregate
• Represents over 59000 variants
• Have a collection of over 197000 phenotypes pulled from HPO
• There are more than 320 depositing centers than spans 45 countries
• Among these is 7 consortia
• The platforms also allows clinicians to contact each other

What can I do/see in DECIPHER: https://www.deciphergenomics.org/

• Varying levels of visibility

• Private [...]

• DECIPHER supports many types of genetic variation
• Supports genome browser visualization
• Supports granular phenotype display
• Decipher supports variant interpretation interfaces
• Once you deposit a patient record each variant has it’s own page
• Supports protein structure visualizations
• Supports ensembl variant effect predictor
• Supports matching patients

• Informs diagnosis
• New phenotypes associated with disease
• New variant-disease pairings

• DECIPHER links to disease management resources
• DECIPHER acts as a holistic approach to understanding genomic medicine
• Both share a vision for safe and secure data sharing
• Also have links to the clinical genomics laboratory community!

Free online course! Interpreting genomic variation: overcoming challenges in diverse populations - materials will be available for a year after the course is done: https://www.futurelearn.com/courses/interpreting-genomic-variation-overcoming-challenges-in-diverse-populations

Q&A

MM: How is DECIPHER capturing phenotypic data? And how are you organizing family history data?

A: we are not capturing them in anything as structured as Phenopackets. We are capturing an array of phenotype terms. We are not capturing health record information. We are looking at supporting Phenopackets in the future. With family history, we don’t have pedigrees because that would make it identifiable. We do have certain mappings that can be linked to mother/father, but that info isn’t captured in a pedigree.

DECIPHER may be a good example/use case to address the issue of variant representation of family history in easy that allow you to anonymize - connection with Clin/Pheno

MF: Planning to have a Beacon set up over DECIPHER in the next year or two. Work towards that will also necessitate that connection with clin/pheno

TB: Co-lead the beacon group, beacon is a framework for defining an API, but the actual API you have to make some decision to set it up. When you talk about putting a beacon on top of the data, you have to specify which beacon. GDI? EJP RD? What will it look like?

MF: Part of a project called paradigm. But we’re also connected to MME so however we use the beacon will need to be aligned to how it’s being used in MME. We are definitely looking towards groups in GDI

TB: GDI hasn’t made their spec yet, EJPRD is running. Let’s keep in touch to make sure we aren’t duplicating Beacons

2024-06-26 - Mondo & DECIPHER

Meeting Recording 

Attendees:

Minutes

Check out the Rare Disease Community’s latest blog post titled “Uncovering and overcoming common data sharing challenges in the Rare Disease landscape”

Updates on the Rare Disease Community Survey

• Goal of the survey: to learn about the group’s diverse data sharing use cases and challenges, to help guide the future direction of focus areas and activities, and contribute to the development of a best practices package / toolkit  
• There was low uptake of the survey, with 7 current respondents
• Refer to the recording (linked above) to see an overview of the results
• Next step: engaging the community at upcoming meetings via Slido to learn how we can better tailor and circulate the survey to increase uptake

Mondo Disease Ontology: Building a Community-driven Disease Resource - Sabrina Toro

• The Monarch initiative leverages information from research and health records to support disease diagnosis and treatment, specifically by enabling integrated data comparisons to make inferences pertaining to disease
• The data Monarch uses comes from many different sources that use different standards to represent disease information (ex - NCI, Orphanet, OMIM, ClinGen, etc)
• To address this discrepancy, Mondo was created by evidence based merging of equivalent disease terms from several different standards
• Mondo covers a broad scope of diseases and reconciles classifications and mapping between terminologies (open community driven resource), hierarchy of relatedness, cross reference IDs of corresponding concepts in other sources
• Mondo includes typical ontology elements like a unique permanent ID, definition, synonyms and their sources

Representation of rare diseases in Mondo

• Many discrepancies exist when defining “rare”: different countries use different cutoffs of prevalence, and some countries have additional criteria such as life-threatening or chronically debilitating diseases, should subtypes of disease count as “rare”, etc
• Mondo harmonizes and integrates all rare disease concepts, according to rare disease authoritative sources like Orphanet, NORD, and GARD
• Mondo also includes metadata, allowing users to choose which rare disease criteria is appropriate for their use case
• Please refer to the meeting recorded (above) to see in depth examples and explanations
• Upcoming work for Mondo includes: expanding link-outs to external curated sources, including new sources based on user requests,

Next steps / ideas to consider

1. Kahlil Lawless offered to help with survey redesign: klawless@illumina.com
2. One of the core issues that this group could try to shed some light on is an exhaustive list of reasons for why certain standards are not being adopted. Lets say, why does not everyone

1. Use phenopackets
2. Use ORDO / Mondo for Rare disease annotations
3. Use HPO for phenotyping (maybe there are good reasons not to use it?)

1. Having a good list of “barriers to adoption” would help this group to devise better strategies to foster adoption

3. Approaches to the survey: casting a wider net, narrowing the focus

2024-04-22 - April Connect Session

Please refer to the Connect Agenda for the meeting recording, slides, and summary

Rare Disease Connect 2024 Agenda

2024-04-03 - Monarch KG + HPO

Meeting Recording

Chairs: Nicole Vasilevsky, Tudor Groza

Attendees: Kshitiz Singh, Shawn O’Neil, Sergi Beltran Agullo, Zohreh Talebizadeh, Kahlil Lawless, Peter Robinson, Sabrina, Nara Sobreira, Monica Munoz-Torres, Bryan Laraway, Coline Thomas, Ashley Hobb, Kayla Socarras, Renan Martin

Regrets:

Minutes

Introduction from leads

• Practices and challenges in data sharing in the Rare Disease domain: fill out the survey here to help inform the future direction and activities of this community!
• Upcoming Connect Rare Disease session: Advancing Data Sharing Practices in the Field, Monday April 22nd, 3:00 pm - 4:30 pm (CEST)/ 9 am - 10:30 am (EDT), look out for calendar invitation!

Shawn O’Neil - The Phenomics Assistant ( Experiments in Knowledge Graph Enhanced Generative AI)

• Monarch integrates different biomedical data from different ontology & data sources, which are aggregated into a knowledge graph that shows disease centered relationships among the data (ie - genes and phenotypes)
• Users can also access/query the monarch data graph programmatically via the API that Monarch provides
• Plugins/GPTs allow for the integration of AI resources like ChatGPT in apps of your choosing, as Monarch has integrated
• How the Monarch Initiative GPT works: Monarch provides a few APIs that the plugin can pull from via natural language instruction, for example from a disease ID input, GPT can return associated phenotypes and genes (alongside links to resources and publications when available)
• Monarch built their own interface which doesn’t require a subscription to ChatGPT+ to use: Phenomics assistant link
• When you give AI access to tools that have a knowledge graph behind them, it increases the proportion of fully correct answers that are returned: see associated publication here
• Future considerations: providing AI with more powerful tools in addition to AIs, Shawn posed a question: what is the best way to evaluate AI models in this space?

Q&A

• Q: Is it possible to integrate information about current curative treatments for a disease (specifically rare and genetic diseases) OR potentially connect to promising preclinical research or clinical trials?
• A: yes absolutely, there are many tools that are posed for open text search (ie publications) vs structured knowledge graph queries.Sample publication of a tool.

Peter Robinson - HPO (Human Phenotype Ontology) Updates

• Each term in the HPO describes a clinical abnormality (anything a physician may use to aid in a differential diagnosis), there are 250 000+ annotations for genetic diseases
• HPO is used at the backbone for many tools such as exomiser

• HPO has gone international! There are translations in 10 different languages
• HPO has been working with the community to improve the ontology in specific clinical areas. This year they will focus on Nephrology, ophthalmology, immunology, mental health, & dermatology
• Phenopackets schema: agreement about how to structure observations (HPO terms and others) to represent a clinical course, enables bioinformaticians to not worry about what the data format is Phenopackets GA4GH Standard, Phenopacket Store
• C3LS software:Clinical Consensus Clustering for Lumping and Splitting - example of using phenopackets

• Too much lumping (grouping diseases into broader categories) will obscure biologically/clinically important differences
• Too much splitting (narrowing disease categories) will yield too many categories with insufficient information about any given one

• HPO is working on a python package to guide optimal lumping and splitting in the context of clustering diseases/associated data
• Peter provided examples of how the algorithm works, which can be viewed in the session recording

2024-02-29 - Grand Challenges in Rare Disease Webinar

Meeting Recording

Meeting slides

Webinar Summary:

Marc Hanauer presented on Orphanet, a unique online resource gathering and improving a vast knowledge base on Rare Diseases, to improve the diagnosis, care, and treatment of patients with Rare Disease. Orphanet addresses major gaps in knowledge within the Rare Disease landscape, by providing specific domain descriptions of Rare Diseases and a plethora of data tools, serving as a key actor in the field. An additional functionality of Orphanet is their new platform Orphadata, which provides users with comprehensive and computable quality data sets related to Rare Disease from the Orphanet knowledge base. Users of Orphadata can also access APIs that can help implement data into their own systems, Orphapackets, an annually revised nomenclature pack, data visualization tools, and terminologies for mapping services.

Switching gears, Matthew Bolz-Johnson from EURODIS presented on the patient experience within the Rare Disease diagnosis journey. Individuals with rare conditions live at the intersection of multiple vulnerable groups, which together have an accumulative impact on poor mental health. Stressors associated with rare conditions include but are not limited to 1) the diagnostic odyssey, 2) living with uncertainty, 3) complex interacting emotional and physical symptoms, and 4) stress and strain on relationships. To help navigate stressors, psychosocial protection factors such as being a part of a connected community and having a positive support system have been shown to positively influence mental health. EURODIS is currently working on a position paper which calls for the implementation of programs to promote mental health and psychosocial support for individuals living with rare conditions.

Article links

• The importance of psychosocial support at diagnosis (2022)
• Consensus Statement of Good Practice (2022)

2024-01-31

Meeting recording

Chairs: Nicole Vasilevsky

Attendees: Ada Hamosh, Bryan Laraway, Charles Steward, Dr. Ravi Bhargava, Heidi Rehm, Kahlil Lawless, Karmen Trzupek, Leigh Carmody, Leslie Matalonga, Nico Matentzoglu, Sabrina Toro, Sergi Beltran Agullo, Wolgang Pernice, Zohreh Talebizadeh, Reger R. Mikaeel, Emily Hartley, Ian Braun, Anne O’Donnell-Luria

Regrets:

Minutes

Data sharing survey updates

• Purpose is to understand needs of this community to come up with best practises
• Question to discuss today is who the audience is beyond who’s already involved
• We’re going about targeted outreach via the outreach sheet, which we invite the community to add contacts to

ClinGen curation of ClinVar - Heidi Rehm

• ClinGen curators review old/outlier submissions in ClinVar
• The goal is to not formally classify a variant but to address obvious issues (if a submission is old, lack of evidence to support claim, etc)  
• The star system is also considered (higher stars are weighted more than lower star rankings) but equal star ranking flags will be resolved
• Feedback from laboratories was collected on 107 SCVc, 94% were agreed upon
• Now on ClinVar there are labeled ‘flagged submissions’ which will not contribute to the annotation of the variant
• They prioritized genes for which ClinGen has an expert panel, the degree of the outlier, and requests can also be taken from the community
• Future goal: if 2 classifications for the same gene regarding 2 different diseases are submitted, an expert panel will validate this and approve 2 classifications of a gene with sufficient evidence

Medical Action Ontology (MAxO) development and tool implementation for the annotation of Rare Disease  - Leigh Carmody

• Like HPO, provides structured vocab for medical actions, one focus is for clinical annotations
• Commonly used ontologies include Mondo for disease annotation, and HPO for phenotypes but MAXo wanted to differentiate between different treatment/investigations
• The tool they developed is called POET > a tool to capture disease & disease phenotype annotations, not open access and requires login
• Omim terms are also linked within the POET tool
• There are about 300 MAxO POET annotations available covering 125 diseases
• They’re working on diagnostic annotations directed at phenotype terms

OMIM - Ada Hamosh  

• OMIM separates discussions of gene entries and phenotype entries but there are links if applicable
• Each gene/phenotype entry is assigned a unique 6 digit number (MIM) which is stable (unlike gene or phenotype names)
• Structured full text descriptions that is curated by OMIM staff
• Genomic coordinates are linked on OMIM entries and provided by NCBI
• There are linked Phenegene graphics (visual representation of phenotype/gene relationships) in both linear and radial views
• New things in the works include OMIM search builder, custom lists, molecular mechanism series (not yet publicly available)
• GeneScout > tool for searching genomic coordinate ranges

2023-11-22

Meeting recording

Chairs: Nicole Vasilevsky, Tudor Groza

Attendees: Caterina Lucano, Dalia Kasperaviciute, Charles Steward, Fowzan Alkuraya, Sergi Beltran Agullo, Heidi Rehm, Paul Muller, Alex Tsai, Jacqui Beckmann, Monica Munoz-Torres, Mogomotsi Matshaba, Beatrice Amos, Karyn Megy, Ada Hamosh, Zohreh Talebizadeh, Marco Roos, Ariel Pradipta

Regrets:

Minutes

TG: *review of perspective paper outline*

FA: Lack of uniformity between people phenotyping. We should seize the opportunity to make phenotyping as uniform as possible. Documenting facial dysmorphic features should be an agreement in tools to minimize subjectivity. Perhaps we are in a position to make that suggestion within the paper.

TG: Do we want to touch on AI?

HR: Goal of the paper - what our role with GA4GH is in this space. Trying to improve and change and encourage data sharing in this space. Want to make sure that there is an objective in this paper that isn’t just saying that ‘this is why it’s important’ without saying how we are going to make it better. For example, are we going to be suggesting standards that don’t exist today - e.g. do we need a different ontology (probably not, but just an example). What are we going to do to push our field to get phenotypes and standardize. What will we do to change the field.

TG: Our next attempt to model the structure of the CoI. Where we sit in GA4GH is as this group. We are in the exploratory space. This space can be materialized that goes deeper into the topic that we find to be important for this space. Trying to see if we can marry some of the topics discussed. Is there a way to make the topics more fine-grained to give this group something more theoretical…not a “here’s what needs to be done” so much as it is “here’s what the challenges are” that our CoI can then work towards addressing.

HR: This group should think about what would be more impactful for our field. A small number of people very deeply phenotypes vs a large number of people phenotype in a limited way. For MME there’s very few people phenotypes, but through data sharing you can go into deep exchange. That whole process would not have happened if we hadn’t brought the community together to match and connect. Some of the things that still need to get off the ground is just basic phenotyping. Could counter with no, we really need to have an impact on therapeutic treatments. If there is a paper written, we need to justify why we would focus on deep vs limited phenotype. For example, if you have limited resources, you should focus on this and here’s why.

AH: Agree with Heidi. Would argue that standardized, high-level, is very really useful. If an ophthalmologist is putting information in they can get into detail about the eyes for sure, but some standardized phenotyping is important. Specificity is disease specific. AI facial recognition technology - is it up to par for people who aren’t white? That worries me.

FA: jury is out, but data shows it does work. Looking to find characteristics like “prominent cheeks” when some people don’t know how to identify that.

AH: Happy to have a picture that takes standardized measurements, but not sure if it’s ready for prime time.

FA: But it works the other way. There’s this bias in the minds of many people. Once they know the diagnosis they force themselves to see certain phenotypes. But AI is objective.

MR: Sergi can comment on how relevant this is for the upcoming RD meeting. Can we start with a good definition of deep phenotyping in the paper and determine if there are certain standards that we need for it.

AH: Very much in the eye of the beholder and the study. Would need to define one for this.

SBA: New European project submitted that will start some time next year.  Have looked at how many HPO terms have been considered deep or informative enough.

KM: Deep phenotyping - the need for it might depend on what you require. Some deep phenotyping is not useful enough to be used in ClinVar. If you do deep level phenotyping on a rare disease can you then identify the patient. And is that then a reason you might not want to do it. That’s why we didn’t put deep phenotyping in ClinVar at the start. Big mistake because now it’s not as useful as it could be.

AH: What we need is standardized searchable phenotyping as part of the clinical record.  This will ease the transition to research and allow greater understanding.

MR: FWIW for a semantic perspective on this I encourage considering submitting something to SWAT4HCLS 2024 (see https://swat4ls.org). (Deadlines will be updated this week.)

HR: There’s the clinical application, then there’s matchmaker exchange where you don’t need any phenotyping, but the number of times people don’t respond at all to a match inquiry. You can put a high level phenoptye in and you can state some things even if you dont have direct consent. These are some ways we can guide people to using phenotyping. Deep vs comprehensive phenotyping.

FA: I think negatives are just as important, and it’s our obligation to guide the community to document negatives. Cna we provide high-level guidance on which negatives are important? People provide better phenotpying and aprt of that is including info  on negatives.

DK: In the paper would be good to outline use cases of different scenarios (e.g. clinical care, research project). In those scenarios you have different challenges of what you want and don’t want to see. Could also include recommendations of what could be done. Another aspect, where phenotyping and standardizing, is when you don’t want to see the variant. Often … don't want to see certain incidental findings. Because the phenotypes are so no-standardized you pull out everything and then you get comments on why you pulled out certain phenotypes. To be able to highlight or hide the list of what phenotypes definitely don’t match and you shouldn’t put efforts towards.

AH: We should continue this conversation. What is breadth, what is depth, what is permissible to share. You were talking about variants. A variant is only relevant in the context of a phenotype.

HR: To Dalia’s comment about what lab’s don’t want to see and blocking info on incidental findings. The way most US labs do it, and seen different ways of doing it based on the country, the US gives consent for what you’re giving back or not. If you find something incidental you have clear guidance on whether or not you give that back or not. I feel it’s dangerous that as we get in the world of exome and genome sequencing, we should tackle that whole discussion of what we do and do not recommend. Do we get REWS in to tackle that specific topic on consent. I would rather labs not be blinded to things that could be useful (DK +1)

TB: this idea of broad vs deep, surely we’re talking about comprehensive data. In an ideal world we’d have full broad, fully deep. Since we’re never going to have everything we need to be clear about what our objective is going to be.

TG: Gut feeling from today is that the paper is premature.

[...go back to recording…]

HR: define the use cases, clinical care, MME, clinvar submission, data sharing for raw data, etc. what is the critical level of phenotyping needed to do that? What consent level is needed to do each of these? What can be done without consent? Include definitions: deep, comprehensive, broad, etc.

If we create this, we can later write a paper about it.

Toolkit for RD data sharing. Talk about the ways our community can participate in data sharing. There is overlap with the above. Can currently share by submitting to ClinVar, MME, etc. Document the different ways to share data

Heidi has seen groups wanting to check, I’ve implemented this or that - suggests a ‘checklist’

For example, DECIPHER database around phenotyping - can we get their framework and allow others to use it

Work to define the ways we can share data

What phenotyping is needed for these things?

Are you a good citizen of rare disease? Provide a checklist, like submitted to ClinVar, etc.

AH: The various things that the RD community needs to filter into the REWS working groups.

REWS can tackle questions about consent and HR noted that REWS is working on this question currently

Question re: phenopackets using HPO terms- is there a way to say I don’t know about a specific organ system?

Moni noted this can be requested to phenopackets. The schema is not prescriptive, we can recommend anything we need

Tony:

Grid - list of use cases against types of data to see where common interests or challenges are. Add into that what Ada said about having one of those columns be a questionnaire or checklist to say what systems have been checked. In parallel we need a grid on data sharing because there are a lot of data sharing use cases as well such as access, analysis, federated access, physical sharing, sharing minimal data or sharing of aggregate data All these aspects of data sharing should go on a parallel list of what data we need and how we need that data. For the use case, do we need discovery, or a federated access solution, or access to minimal data? This would give us a full picture of what data we need and how we need this data

TG: how can we provide recommendations by means of the use cases and utility that is being sought so the reader can go there if you need that.

Action items

• Development of grid system outlining use cases against types of data

Resources and Links

Rare Disease Launch

10th_Plenary-Rare_Disease_Community_Workshop-220819

Rare_Disease-Contact_Sheet-220707

Rare_Disease-Community_Charter

20XX-XX-XX: Template Agenda

Chairs:

Attendees:

Regrets: