An NIH Bridge2AI Initiative Qualitative Study: Ethical Implications of Functional Genomics Data Generation and Downstream AI Uses

Danielle M. Pacia – Research Associate, Research Department, The Hastings Center; Ian Stevens – The Hastings Center; Jean-Christophe Bélisle-Pipon – Simon Fraser University; Vardit Ravitsky – The Hastings Center

Abstract: The fast-evolving field of functional genomics focuses on better characterizing the relationship between genotypes and phenotypes at a molecular level, such as protein-protein interactions within and between cells. For example, although we understand that BRCA1 is often linked to cancer development, little is known about the biological processes through which the genotype translates into the phenotype (the disease state). One such effort in functional genomics research is the Cell Maps for AI (CM4AI) Consortium, which is part of the larger NIH Bridge2AI (B2AI) Initiative. This initiative seeks to generate new datasets while concurrently developing best practices for machine learning (ML). CM4AI, using several different imaging approaches, will create a library of "maps" to better characterize cellular function across different disease contexts and train future ML tools, including those used for precision medicine interventions.

Using a modified grounded theory approach, a team of three researchers analyzed key stakeholder interviews to examine decision points in the AI lifecycle and the challenges of managing collaborative, multidisciplinary research efforts. Through these interviews with AI developers, data scientists, and other B2AI stakeholders, we explored the ethical, legal, and social values related to functional genomics and AI research.

This paper presentation, drawing from our qualitative study, will present the ethical, legal, and social values related to functional genomics research and data generation practices for downstream AI use, including Problem Identification and Prioritization, Data Accessibility/Usability, Data Sourcing, Transparency/Explainability, Generalizability, and Bias. Insights regarding the practice of multidisciplinary collaboration in generating AI-ready datasets will also be presented.

Keywords: Functional Genomics, Bridge2AI, Qualitative Research