Data Harmonization Theme Team

Neuroscience Community of Interest

Background

Key Resources

Team Members

Agendas and Minutes

Meeting: 2025-06-25

Meeting: 2025-02-03

Meeting: 2024-12-10

Kick-off Meeting: 2024-11-06

Minutes

Background

Who are we? The Data Harmonization Group, led by JB Poline (McGill), is one of three specialized theme teams formed by the Neuroscience Community to address critical challenges in data sharing, integration, and interoperability within the field of neuroscience. This theme team plays a pivotal role in the effort to streamline the linking and harmonization of distributed datasets, enabling researchers across different institutions to efficiently share and access valuable data. This is an OPEN group (sign up here). A list of current team members can be found here.

What do we want to do? The core question this group seeks to address is: How can we effectively harmonize distributed data across diverse research settings? Rather than doing the harmonization work for researchers, the group's approach emphasizes empowering researchers—those who know their data best and have the strongest incentives to harmonize it—to take charge. This means providing tools, frameworks, and common standards that allow them to harmonize their data for shared use.

What is the problem? The neuroscience field has seen few easy-to-install solutions for data harmonization thus far. Several critical issues remain, including:

• Granularity of search versus retrieval: How much detail is available when searching for data, and how granular is the data when retrieved?
• Generic versus specific use cases: Balancing the development of generic tools that can be applied across many studies with the need for specific solutions tailored to certain datasets.
• Breaking down modality silos: Ensuring that harmonization spans from raw to processed data, across different data modalities (e.g., imaging, genetics, behavior).

How will we achieve success? The key to successful data harmonization is flexibility—creating adaptable tools that empower researchers to harmonize data according to their specific needs. The Data Harmonization Group will focus on making distributed data integration more accessible and impactful by:

1. Leveraging and promoting existing projects like Neurobagel and Nipoppy, which already offer frameworks for federated data queries and processing pipelines.
2. Developing repositories of resources that provide researchers with the tools and standards they need to harmonize their data effectively.
3. Facilitating collaboration by bringing together groups and individuals facing similar data challenges, enabling them to share insights, exchange best practices, and potentially launch cross-institutional projects aimed at linking datasets.

Key Resources

Team Members

Agendas and Minutes

Future meeting notes will be available here: Neuroscience Community Meetings

Meeting: 2025-09-24

Attendees: Francis, JB, Nicole, Saskia, Rae, Kelly, Kaitlyn

Actions:

• The group will continue to refine and expand the ontology/standards list, potentially adding more details on the purpose, domain, and adoption of each item.
• JB to follow up with his team on the work they are doing to catalog assessments and potentially present an update at the next meeting.
• JB to reach out to Lyuba Zehl to possibly present at the next meeting (October 29th, 2pm)
• Saskia to present on the metadata schema used at the Allen Institute for a future meeting
• Nicole to report back if additional information about medication ontologies is obtained
• The group will explore inviting external experts to present on specific ontologies or standards that could be helpful, such as in the area of medication-related ontologies.

Overview: The Neuro Community's Data Harmonization Theme Team discussed the need for a structured approach to data harmonization, focusing on cognitive and behavioral assessments. Key points included the importance of ontologies like Pheno packets, HPO, and MONDO, and the need for a landscape analysis of existing standards. Saskia de Vries introduced the Allen Institute's AIND schema for data acquisition, while Jean-Baptiste Poline highlighted the need for a catalog of cognitive assessments. The team agreed to refine the ontology list, explore potential collaborations, and schedule future meetings to present and discuss these initiatives in detail.

Ontology list: need to further classify and add details as described below (see new spreadsheet)

• Explain what each one is and how it is used

• Provide explanation of how these were adopted

• Group by domain and coverage (assign a type to each)

• Identify which are compatible with each other

• Separate ontologies, schemas, and other tools
• Identify pros/cons of each group
• Add in whether their organization has implemented any of these

Topics of interest: need to Identify experts that we could reach out to that could present and/or provide guidance

• JB: want to create an ontology of assessments (similar to cognitive atlas)

• Starting from ReproSchema (https://github.com/ReproNim/reproschema)

• Kelly: Cognitive, neuropsychological assessments, and behavioural tasks are areas of harmonization of interest.
• Saskia: interested in HED tag (hierarchical events descriptors) works well for behavioral events, but might not work well for higher level lab.

• Pydantic used.https://docs.pydantic.dev/latest/ 
• Saskia to demo their new version 2.0 at next meeting

• Nicole: looking into medication ontologies

→ Upcoming presentations:

• Demo from Saskia et al.
• Presentation from JB
• Lyuba Zehl from EBRAINS (OpenMinds)

Meeting: 2025-06-25

Attendees:  

Moni Torres (Uni Colorado, Monarch Initiative), Jaime Delgado, Nicole Schoer (CAMH), Kelly Shen (SFU), Jaime Delgado (Uni Barcelona), Rae Price (OBI), Ali Demir (OBI), Francis Jeanson (OBI),

Regrets: JB Poline, Jessica Seegobin

Meeting Recording

https://us02web.zoom.us/rec/share/b_WMZRvAzqtrq71RymfJ2FOT8pV0lEfyrdouaG3fELB9eap2qBVQBS8SatQZKXmD.djJbKZSYXtW5JlWm?startTime=1750878045000

Passcode: iNz88Zp+

Minutes

Moni presentation on standards for data formats (slides here: https://docs.google.com/presentation/d/1FvcyGSAs9558eP8j-DFH2OPw-xZEj3sYGrophLxkWRw/edit?slide=id.SLIDES_API195769423_0#slide=id.SLIDES_API195769423_0 )

Need robust data interoperability.

Phenopackets: GA4GH, ISO, FAIR, current integration with FHIR and OMOP for clinical applications

Nuance between Schema and Ontology

        Schema defines the data structure and relationships between elements. Elements in a schema are often ontology terms.

        Phenopackets is a “schema”

Human Phenotype Ontology (and ontology)

        Characterizes over 18K standards terms related to phenotypic abnormalities.

        Developed with input from experts from various areas of expertise

MONDO Disease Ontology

Integrated many existing ontologies and dictionaries and removes inconsistencies and biases, and adds semantic mapping for computational reasoning.

What can we do for Neuro?

Phenopackets could help link concepts in fMRI for example

HPO in itself could have a vocabulary for phenotypic annotation

MONDO could be used to unify the neuro disease classification. And has hierarchical strucutrue to link different cohorts.

Example: Linking Electrophysiology to Phenopackets

Ongoing effort to integrate HPO and Mondo into EHRs which could support the clinic

Discussion:

EU have had workshops on extending the HPO ontologies to adapt to domain specific areas.

Adopters: https://docs.google.com/presentation/d/1_8Kce_3bL2wD1P4_yMypqt4lGtKgLt2lANWZWGOuKcs/edit?slide=id.g329ef3178fa_1_1141#slide=id.g329ef3178fa_1_1141 

Opportunities to connect with other members of HPO and MONDO

        E.g. Sabrina Torro developer of MONDO

Actions

Meeting: 2025-04-30

Attendees: Francis Jeanson, Neerjah Skantharajah, Jessica Seegobin, Jaime Delgado, Azadeh Kushki, Justina Zych, Kelly Shen, Kyle Weber, Patrick Ray, Brian Lee, Carol Thompson, Karen Onyeneho

Regrets: JB Poline

Meeting Recording

Minutes

• Main conclusion from the session was that the group should identify use cases and then specific requirements/standards that would fulfill them
• Some need are specific to neuroscience, some things are disease/domain agnostic - making that distinction will be important going forward
• Identify security problems that the community feels are relevant and then analyze what we can do on top of current solutions
• A potential challenge to address: ontologies - to start can look at current work within GA4GH and INCF, as well as LOINC and Snomed
• Learning about GA4GH standards - best practices to support the knowledge gap
• Finding ontology standards across domains - examples of project in the GA4GH realm what the possibilities are of how different domains can be linked
• Next meeting topic could focus on ontologies:

• Bringing in Phenopackets and discussing HPO mapping
• Monarch Archive MONDO ontology

• Ontologies discussion could lead into data security use cases

Actions

• Reach out to Phenopackets and Monarch teams to present at next meeting

Meeting: 2025-02-03

Attendees: Kyle Weber, Aida Falcon, Ali Demir, Tristan Glatard, Bukky, Francis Jeanson, Edward Xu, Mojib Javadi, Moyez Dharsee, Nadia Farrag, Kimberly Ray, Ali Khan, Kajal Sinhmar, Patrick, Moyez Dharsee, Brian Lee (UCSC)

Regrets:

Meeting Recording

Presenter Slides

Minutes

Presentation Description: OBI collects a wide range of multi-modal data across a variety of brain disorders and patient populations via large scale multi-site research studies. Both health control and patient data are collected for complex disorders that often have co-morbid conditions. So far OBI's Brain-CODE platform has made data available for secondary use on a study by study basis. But we're seeing a growing interest in the ability to combine data from multiple research studies to form sophisticated cohorts of individuals that could be analyzed together. As a result, we've been actively developing an updated harmonized metadata model with high level cohort information. Initially based on the SPARC SDS model, our data model captures study dataset description information, high level participant information, and individual participant records/samples information in standardised JSON format. To facilitate cohort discovery, we have built a Data Discovery Portal and we’re seeking to integrate this with Federation APIs capabilities such as the Neurobagel API which will further allow researchers to build much larger cohorts of data for advanced analytics and AI modelling.  

OBI’s Next Gen Metadata Model

Presented by Francis Jeanson

• Neuroscience research involves a variety of data modalities - clinical, neuroimaging genomics, molecular, wearable/sensors
• Brain-CODE supports collection storage, federation, sharing and analysis of this data throughout large research programs
• There are 1000+ Brain-CODE users
• Research programs include CAN-BIND, Transcendent, POND Network, CP-NET and CALM.
• Across these studies there are data harmonization tools that are used across studies to increase interoperability and we have a mechanism for sharing data externally so that external individuals can request access
• Brain-CODE enables:

• Multi-modal data collection
• Data linking (using homomorphic encryption)
• Common Data Elements - a subset of standard measures across all OBI studies
• Data sharing

• Q: NIH suggested having global identifiers (i.e. sharing hash of info that’s shareable from patient) is that what you mean by data linking?

• A: That’s a type of linking (global unique identifier), and you need to use standards that are as stable as possible, but variables can change so that’s prone to challenges, but it’s an interesting concept. Brain-CODE’s particular tech is baked on encrypted health care numbers.

• Q: Does that generate constraints because of the possibility of reidentifying patients?

• A: Yes, the other risk is reidentification. It’s about aggregating sufficient info about the individual so you can conduct more significant and insightful research. One way to do that is through informed consent for participants so that they know their data can be linked, but that it will be done in a way that prioritizes privacy.

• Brain-CODE has had 249 controlled requests and 312 open requests

• The controlled access needs research ethics approval and Brain-CODE has their own data access committee
• Requests have come from Canada, US, China, Brazil etc.

• Working with Indoc team to make data releases available for request from the international community
• There are a number of controlled datasets that have metadata which describes high level info about the study (e.g. study description, collaborator names, version number, size of data, # of participants)
• More recently, we’ve been generating DOIs and ___IDs
• In some cases we have worked with researchers to create videos to help describe the study
• Once logged into the platform you can make a request for access of the data
• This portal allows you to explore the datasets and select the modalities you are interested in
• You can also select the instruments you are interested in exploring
• With each study release, things have worked a little differently - some prefer to show their data in a more customized way. Dashboard have been used to show users a customized view determined by the researchers

• The challenge here is that users may want to browse across and combine findings from two or more studies/cohorts, but since they are bespoke that can prove difficult
• There’s an opportunity to develop an approach whereby researchers can search across datasets and combine data from multiple different datasets to create virtual cohorts - this is the Next Generation Metadata Model!

• Cohort Discovery (see Federated Cohort Building GA4GH WG)

• Cohorts refer to groups of individuals and their data who share specific characteristics relevant to one or more research questions. Building cohorts across federated data repositories would allow researchers to aggregate more data for their studies, reuse data more easily, and better collaborate.

• To accomplish this we must shift our focus from study-centric to patient-centric
• Instead of collecting Subject ID, Assessments,  Omics, Consents for studies we can collect demographics, diagnoses, conditions, provenance for patients
• The key dimensions, from Brain-CODE’s perspective, to prioritize for harmonization is demographics, diagnoses, modalities, instruments, tasks and visit timelines
• Have started to look at datasets available for POND, ONDRI, and CANBIND and mapping key dimensions between OBI categories and other vocabularies (e.g. SNOMED, LOINC)
• Currently working on mapping demographics, diagnoses, modalities, instruments, tasks and visit timelines to a normative code
• Working on the structure of the metadata - it has three major layers
• Dataset > Participants > Samples

• Based on SDS SPARC data model which is under review by INCF

• Dataset will have high level info about the dataset/study, different arms in the study, collaborators, research groups etc.
• The participant metadata will collect basic demographic info (age, sex, ethnicity, with flexibility to add fields if needed, diagnoses
• Sample will include the type of data being collected/modality, it is a single point of data collection about that individual (e.g. assessments, imaging, sequencing of a biosample)
• See slides for version 3 draft of metadata for cohort discovery
• With this metadata model we can start to aggregate results into a cohort discovery platform
• The cohort discovery platform can filter cohorts by modality, age, and other factors to help explore the cohort data
• Next step is to see how data can be fetched from external datasets - working with Neurobagel to help achieve this

• Neurobagel has core ways to filter and it can be aligned with the metadata model
• This will enable Brain-Code data to be discoverable by external researchers

• OBI’s neurobagel node would load studies from other sites into their Data Discovery Platform (DDP), and data from the DDP could be uploaded to the OBI NEurobagel node which allows others to discover OBI data since nodes are connected, and on the other direction, make OBI data discoverable on in Neurobagel
• Exploring ways to automate this
• BBMRI in the EU : rules of access

Discussions

Q (Tristan): For imaging, how does the SPARC standard compare to BIDS. and do you anticipate that they can convert easily between the two

A: They play slightly different roles. The SDS is the surface layer and is able to point to the imaging data. Indoc has helped prep the imaging data with SDS. BIDS will be useful when a request has been placed. On the core discovery side we wanted to have that extra layer that spans across the modalities. Data is not yet being pulled to make those data releases. The first step is to load the metadata and request access. Still have some testing to do for the pull of the data.

Q (Patrick):  How do you see the development of systems that propagate access control and consent, given your approach of starting with content metadata and working toward interoperability?

A: We don’t have a built-in metadata schema, but we use the Digital Use Conditions project, which defines a JSON schema that can reference ontologies like DUO (GA4GH) and ICO. This structure can be exposed as an endpoint in a repository - potentially linked via a DOI- surfacing machine readable access conditions. These conditions consider factors like user profile, jurisdiction, and intended use. While users currently need to verify compliance themselves, future tools could automate matching and streamline the experience.

Q: Applied at study level or participant level?

A: will probably be at the study level for OBI but may depend on the insititution.

Q (JB) Suggest that we work collaboratively with others to develop layers of interrogation for authorization and access. Most of the Neurobagel nodes are not visible because of who has access to the data. Would recommend that those layers of authorization are developed by consortia so that they can be adopted across projects.

A: If there could be a standard mapping for translation between Neurobagel model and in house model that would be great.

Q (Patrick): As people begin to share data you can look at the intersection of data and overlap of content to help solve this problem. It may be geared more towards a framework than a standard.

Meeting: 2024-12-10

Attendees: Brian Lee (UCSC), Saskia de Vries, Sebastian Urchs, Ali Demir, JB Poline, Vidhya Venkat, Ali Khan, Dara Cannon, Edward Xu, Mette Peters, Moyez Dharsee, Kelly Shen, Karyn Onyeneho, Neerjah Skantharajah

Regrets:

Meeting Recording

Neurobagel Slides

Minutes

• Neuroscience community leadership meeting yesterday - happy with the direction of this group
• Neuroscience Community to propose a connect session
• GA4GH Connect taking place in Boston from April 1-4
• Next meeting will focus on future directions

AIND Metadata Schema

• Focus on neurophysiology (Allen Institute)
• Allen Institute has a strong mission to practice open science and share data, specifically a lot earlier in the process
• This mission requires a robust metadata schema
• Received feedback that participants had a difficult time determining what the experiments were for in the (dandy?) sets
• Good metadata enables flexible reuse
• Schema defined on github - https://github.com/AllenNeuralDynamics/aind-data-schema
• 7 major classes in the schema

• Data_description - administrative details
• Subject - somewhat generic for broader application to other species
• Procedures - procedures done to animal or tissue prior to data collection (e.g. injections, craniotomies etc.), what material is being injected, who did the procedure, which can be anonymized
• Instrument/rig - instrument used for data collection, does not have day to day operational information
• Acquisition/session - details about how the data is acquired at that specific time (e.g. laser power), for physiology you might have information about where each probe is located or the behavioural stimulus.
• Processing - processing code and parameters for data processing
• Quality control - recently added! Quality control evaluations that can be either automatic or manual. This is useful because people want to use good data!

• Processing and quality control can be appended after the data collection

Q: MP - after sacrifice do any of the mice have spatial transcriptomics [check recording] data?

A: SdV - Yes [check recording]

MP: related to IDs whether that’s the mice of the reagents - are you utilizing RIDs?

SdV: Anywhere where I can use a persistent identifier I will use it!

• Persistent Identifiers currently incorporated or going to be incorporated in AIND metadata schema
• ORCiD
• ROR
• MGI
• RRID
• EMAP - for mouse anatomy
• Addgene
• NCBI
• protocols.io
• Allen Brain Atlas
• Internally we document all of our protocols, so when we publish our protocol we link to that
• Q: MP - Do you mint DOIs to datasets?

• SdV: yes, we put our data in a public bucket before we have a complete dataset. As a single asset that was acquired yesterday it won't have a DOI. When we are publishing a paper and have analysed a group of data assets we will put them in central repositories.

• Q: SU- When giving your data to the user it will go on dandy. DANDI doesn’t restrict the domain/range? You can use any vocabulary to describe reagents. Are you being consistent in your own data on the vocabulary you’re using?
• SdV: we put things on dandy when they’re associated with a paper. Using DANDI is a duplication but it is helpful when we are working on a group of things together. Schema in NWB is not very extensive so there’s a lot of info that doesn’t fit into that schema so we are finding other ways to make sure that all of this info ends up in the NWB files, even if that just means JSON files ending up there.
• Doc DB

• Metadata gets ingested when we upload to Doc DB and we can query against this DB which allows us to identify assets in specific metadata fields
• Links to outputs and visualizations
• Using natural language to query the DB
• GAMER: Generative Analysis of metadata Retrieval
• Planning 2.0 release in Q1 2025

• Simplify some classes
• Cleaning up some of the pedantic logic
• Adding extensions

• Build more ingest and query tools
• Q-JB: Was the schema built from scratch? Or pulled from others?

• SdV: For the imaging data we looked extensively at existing standards for OME. So a lot of the classes and fields that pertain to in vitro microscopy data …]. For physiology/behaviour data we leaned into info that we thought was important/kept track of in our own work

Neurobagel

• Perspective is looking from the outside in
• We don’t have data
• We are interested in how to connect people who have data with people who want data
• How do we streamline and simplify existing solutions?
• How do we create a common data model to facilitate interoperability between individual sites?
• Nipoppy: protocol for the individual dataset
• Neurobagel - answering the discovery question
• We try to come up with a simple data model centred around a research participant
• Everyone can find out where the diagnosis is in their respective column, and if we make it easy enough they can click that column
• For the purpose of discovery, we want to point to an ontology (e.g. SNOMED)

Questions

• How do you want people to find your data
• We’ve been building a data platform and mostly people will engage through our data platform which will take a few different forms - having credentials into a cloud based computing platform
• MP: Do we need a global data directory :)

• In the early days of the internet there was such a thing, but we eventually got to a point where it became to much and so we developed a search engine

• How do i find and take advantage of the neuro resources?!

• It’s a motivation of Neurobagel

Next Meeting

• Could we get two more presenters for the next round and we will also add in some time for where we are going and create a precise deliverable for where we are going

• Karyn Onyeneho
• [tentative] Mette Peters

Kick-off Meeting: 2024-11-06

Attendees: Mette Peters, Brian Lee, Karl Klein, Saskia de Vries, Kajal Sinhmar, Faramarz, Sebastian Urchs, Azadeh Kushki, Karyn Onyeneho, Stephen Arnott, JB Poline, Neerjah Skantharajah, Edward Xu, Alexander Barton, Ali Khan

Regrets:

Meeting Recording

Minutes

• Problem: we do not have access to enough data to make significant findings!
• GA4GH is supporting the neuroscience community to see what can be done that is useful and will help the community address data harmonization challenges
• The goal, in addition to connecting people, is to identify areas of work that can be impactful and useful for the community
• Neerjah summary on where we come from (GA4GH X INCF)

Roundtable of Introductions and Gaps

• Edward Xu

• Attending on behalf of kevin cobane?
• From University of Ottawa Heart Institute
• Involved in the data portal project

• Mette Peters

• Senior advisor at the NIH/NIA
• Supporting the division of neuroscience in data science and infrastructure related questions
• Have acted as the director for a large Data Coordination Center supporting data sharing from multiple ICs and non-profit foundations
• Come from the -omics world, but what interested her is that neuroimaging does not exist in a vacuum so how can we establish data linkages, and harmonize data across these different data domains, and also developing data models to use data in different ways

• Karl Klein

• University of Calgary, Hotchkiss Brain Institute
• Research area of focus - epilepsy genetics
• Challenge with differences in documentation, especially in epilepsy field

• Alex Barton

• CAMH - managing brain imaging data
• More data standardization makes his job easier 🙂

• Karyn Onyeneho

• Geneticist by training with a background in chronic diseases
• Chair of NIAGADS and Senior Advisor for Genomic Data Sharing for the NIA across its extramural programs. https://www.niagads.org
• Joined NIA and serves as a senior advisor for genomic data sharing
• Leads implementation of federal policies centers around data sharing
• Ensures that national institutes adhere to federal policies
• Serves as the chair of the data access committee and oversees data access requests
• Interested in this group because while genomics is being used worldwide, the utility and effectiveness has not been maximized for individuals with indigenous ancestry
• Also interested in determine social constructs across ancestry

• Azadeh Kushki

• Associate Chief o Data Science at Holland Bloorview
• Analytics Lead for Province of Ontario Neurodevelopmental Disorders Network (founded by OBI / Ontario gvt)

• Saskia de Vries

• Associate director data and outreach at the Allen Institute for Neural Dynamics (brand new website, content still in development)
• Building a platform to support open and reproducible data sharing for colllaborative analysis workflows primarily in systems neuroscience space
• Light sheet imaging
• Lacking std in neurophysiology / dynamics
• Interested in metadata standards for flexible reuse

• Brian Lee

• UC Santa Cruz
• worked on UCSC genome browser
• “sspygene” : https://sspsygene.ucsc.edu/ 
• Not in production : a window of opportunity for the metadata standard
• BIL / Dandy / NIMA for repositories
• Interested in development of metadata standards
• Also open to implementation/trying out products/standards that this group creates

• Faramarz

• Data engineer collaborated with a previous OBI project that worked with genomic data
• Interested in data harmonization and standardization generally

• Ali Khan

• Associate Professor at University of Western Ontario
• Expertise in developing image analysis for MRI
• Quite involved in pre-clinical studies and we have a recently funded project called TrRIDENT that spans across organoids in order to develop pre-clinical drug evaluation pipeline
• Looking for metadata standards to adopt for that project

• Stephen Arnott

• Works at Indoc Systems (Toronto)
• Building health data ecosystems for various groups
• Indoc works with OBI
• Supporting the neuroimaging platform with BRAIN-CODE that also collects demographic data
• Works across Canada, US and Europe which is why they’re interested in data harmonization standards that can support these efforts

• Sebastian Urchs

• Works with JB, leads the neurobagel.org project
• Software developer that develops tools for harmonization of phenotypic tabular data for cross site discoveries
• Some institutes have a lot of infrastructures and others don’t, so making something pragmatic/social engineering project that can help these groups work together and facilitating that through federated discovery is of interest

Brainstorm: What is the key problem that you would like to solve? What gaps can this group address?

Brian

• Getting the information from the labs, especially when dealing with dozens of labs
• Even getting a thumbs up/down response on metadata fields that they believe are required or not

Karyn

• Would like to see less burden on investigators seeking access to genomic data
• Reducing barriers to discovery
• Diversifying the availability of data - how “common” are CDEs? Would like to investigate this on the country level

Sebastian Urchs

• Brian's point re: thumbs up/down, and Karyn’s comment on CDES - what is it that you want to do with this information? Is discovery the barrier?

• KO: All of the above. Accessibility, discoverability and harmonization are key. Having access to that data that can lead to scientific discovery is of great importance. ¼ NIA sessions at a recent event focused on global data sharing and the barriers faced. The value of data once it’s accessible is important and that’s a recurring theme we’ve gotten from PIs.

Brian

• Getting information before data are acquired / generated

Mette

• For data users coming from multiple data silos, it probably isn’t feasible to come up with a standard that rules them all - so do we want to discuss the use of AI in data harmonizing, generative CDEs, and tools to enable semantic mapping among different metadata models and CDEs

Karl

• You complete your metadata differently every time depending on the use case, which makes it very hard to standardize

Saskia

• One of the key things we’re trying to do is put a lot of our standardization efforts, particularly acquiring metadata, early in the process during the acquisition of data
• AI will definitely be a tool that’s helpful in this space. Right now we are relying on AI for querying metadata as opposed to ingesting it
• Trying to lean more into existing ontologies and persistence identifiers, because then it doesn’t matter as much if the names of our metadata fields differ
• Determining what ontologies are underdeveloped and require more development or more use

Mette

• This is where AI can be used. It can pull on knowledge from existing databases and be used for semantic matching

JB

• What incentive do people have for annotating data? If people know what the data will be used for that can be helpful

Mette

• Review the docuthon for people who weren't there

Next Steps/Wrap Up

• At the next meeting , we can present 3-4 slides on tools that we are working on and later on determine some deliverables

• Neerjah to get a list of interested presenters - what is their project about, what problem are they trying to solve?

• Solicit some insights from DaMaSC
• Everyone to review the docuthon document: Hackathon - Data Harmonization Worksheet

Chat:

• Mette: asking about spatial transcriptomics data mode (Brian?)
• Brian (chat): This is a little hard to see, but here is the GitHub BICAN GitHub ticket for adding a Image-based Spatial Transcriptomics model being worked on in BICAN (separate from my SSPsyGene group, but we’ll likely use it):  https://github.com/brain-bican/metadata-schemas/issues/49
• Brian: Another group that has very rich metadata is HuBMAP and SenNet using CEDAR from Stanford to make a systematic approach to metadata modeling and acquisition.  They have a rich spatial transcriptomics schema. This spreadsheet is a useful overview of terms they use (or used):  https://hubmapconsortium.github.io/ingest-validation-tools/field-schemas.xlsx