AGENDA

Diversity in Datasets
25th April 2022 at 4pm BST, 11am EDT, 8am PT

(dial in numbers below)

Meeting recording [passcode x.t71b9%]

Zoom Details:

Join Zoom Meeting

https://us02web.zoom.us/j/82351738156?pwd=aDZyMStVMjk5c09PSktBUGUzMGFFZz09

Meeting ID: 823 5173 8156

Passcode: 493379

Attendees:

1. Maili Raven-Adams
2. Chuck
3. Lindsay Smith
4. Anna Lewis
5. Shuba Gopal
6. Alessia Costa
7. Ariana Gonzalez
8. Maxine Mackintosh
9. Elena Ghanaim
10. Angela Page
11. Diksha Srivastava
12. Tina Hernadez-Boussard
13. Yvonne Bombard
14. Sarah Nelson
15. Emi Casas-Silva

Minutes:

[Activity] Please take 3 minutes to think and type out your answer to the question: What does diversity in datasets mean to you.

SG: Representation of people across the world from all walks of life

AL: Capturing the data we need to understand what drives outcomes, so that we can intervene as effectively as possible to improve lives

AP: To me, diversity in datasets means that collectively, all the world’s genomic data is representative of the global human population with respect to gender, ancestry, geography, etc.

AC: Data that is representative of different subjects

ECS: Diversity in datasets to me means that we work towards have as broad representation as possible from people around the world representing as much diversity in ancestry in the genomic data collected and stored.

EG: Diversity in datasets means diversity of biomedical datasets on a few dimensions: Datasets are diverse in the geographic communities from which research participants are recruited; diversity in the racial/ethnic (social constructs) make-up of research participants; diversity in the ancestry (not entirely straightforward) as well) of research participants.

MRA: A way towards ensuring genomic research is applicable to everyone, and not just subsets of the population (s)

DS: Diversity in datasets should enable a world where everyone receives equitable genomic health outcomes, irrespective of one’s background.

MM: The degree and type of variation across the axes that matter in your data allow you to most comprehensively and robustly answer your question

SN: There are several different axes of "diversity" - the first ones that come to mind (especially for genomics and biomedical research): genetic diversity, including across global populations; gender diversity

C: greater opportunity to discover association between variants and proximal transcript regulation

AG: It means the possibility fulfill representation of data worldwide. It would show a common and communitary effort to show unity and create bonds to help every country concerning their reality and possibilities.

YB: All have the meaningful opportunity to participate in research and genomic medicine

TH: Take a step back and look at the broader issue we’re interested in. Many datasets remain inapplicable to communities around the world. People use these public datasets to do research around the world, meaning they are more likely to have subjects from European ancestry. Produces biases that are more beneficial to certain persons, making it a challenge for interpretation.

MRA: Some unanswered questions - what is diversity? What can be considered representative? How we can measure that datasets are diverse, and does the answer change depending on context? Is there a role for standards and regulatory oversight? Are there ever any good reasons for unrepresentative data?

If you have resource or any other fundamental questions, please let us know!

AL: There has been a big focus on how we categorize participants and their samples. Diversity and the ways in which we categorize people are hard to separate. In the US, there has been a lot of focus on race-based health disparities, which has lead to a lot of thinking about the use of race in health. Diversity in genomics has a messy history. The Human Genome Diversity project was branded the “Vampire Project” by the World Council of Indigenous communities.

The idea of admixture does not solve this problem, but compounds it.

If we’re interested in understanding adverse health outcomes, including health disparities, we need to ensure diversity is in the factors that we think about.

TH: We need to have a greater emphasis on population parity - that datasets we use and data consent, representation bias, sample to population inference, transfer across populations. Do we want accurate representation, or ideal representation? We have historical biases in our systems, might want to think more about ideal representation, what data do we want to represent rather than what is available in a particular population or setting.

AL: I know for some, this might be the first GA4GH REWS work stream you’ve been involved in. We’ll aim for some kind of deliverable, like a white paper or policy doc and that has to be approved by the steering committee. We also have the option of a publication alongside of that.

MRA: Envisaged deliverables - defining the concepts that underpin this topic, like diversity and representation. What does that mean in different settings and how does it apply internationally. Then we want to translate into actionable recommendations, touching upon data acquisition, usability of the dataset, measurability, transparency in data collection.

AL: Everything we do has that international focus. Any reactions from anyone? The directions we take are very open.

ECS: If we have recommendations for researchers, think about the feasibility of how to measure and implement those recommendations. How could we provide information on that as well.

TH: Great point, many of the datasets we use are not representative. Need to ensure we use them responsibly. Generate the conversation on points that need attention. Reporting what is actually in your data, who it was acquired from, etc. is the first step in being transparent.

MM: Been involved in Standing Together, aims to create a measure or standard for diverse in health datasets. Not genomics exclusively. Most comprehensive work is sheets for health datasets. Recommend everyone read the pre-print linked in the chat. Second, homogenous populations can be really important for some research questions. Look at known-unknowns, how far into the future of that use can we go.

SG: Touch on the data acquisition piece. Many groups were eager to try and bring more racial diversity in US datasets, but then ran into challenges because the history of medical research with minority populations is horrific, and there’s a reason many do not want to participate. Acknowledging that we’re trying to address that. Culture sensitivity is a big piece.

AL: GA4GH had an output last year about community engagement that digs into some of these points.

TH: It’s about building trust. We really should include the stakeholders being involved in every aspect of research questions and design. Diversity in the stakeholders involved in the data collection is really important.

AL: Need to think about how ideas of diversity cross borders.

MM: Levels of diversity. When we looked at the global problem at lack of diversity, once you look into certain countries at the local level its less likely to be as large of a problem (ancestry)

AG: In Latin America, we’re still some steps back, now living the reality that social/health system are incorporated like genetic reports. Don’t have a data collection or privacy law dedicated to genetic information. To talk about the use, some countries are a little behind. The majority of data is part of private laboratories because they have the resources.

AL: Do you have an impression of whether the topic of diversity in data comes up in Argentina?

AG: It does, but in a small sphere. Very few places that can access or use genomic testing. The work we have to do is a lot right now. A lot of tests go outside of the country. Difficult to engage. We don’t even have a biobank. Not discussed as much as it should be.

MRA: As we look at these concepts of defining, could do an empirical study to see how that differs across countries, which could be an opportunity to bring in more voices.

TH: Do some due diligence and reach out to some countries/institutions.

Next Steps:

Already touched on who is missing from this discussion. Would be great if everyone could reach out to one group they see missing.

In the next 1 or 2 meetings, have presentation from 1-2 members on the work that you are doing. Aim for a high level alignment of points. Then move to the Google Group to capture ideas.

AL: Maxine present on some of the links she has shared.

TH: Have Ariana give a presentation about her current situation? Need to overcome the barriers of how to even collect that data?

AG: Would be happy to!

AL: Other ideas on how to organize time for subsequent meetings?

MRA: Move to a different time, so next one will be a different time, will utilize that.  Will send a follow up email tomorrow.

MM: Synchronous or asych meetings?

TH: Prefer synchronous meetings.

MRA: Can do what works for us - Google Docs, Slack, Email, etc.

AL: Think about what GA4GH could do in this space. Have a particular thought about our unique contributions internationally.