Pharma & Life Sciences

AI in genomics and R&D: the evidence gap nobody talks about

The AI-in-healthcare debate lives in the clinic. But the deeper deployment is upstream, in research: prioritizing targets, filtering candidates, interpreting genomic variants. Decisions made there surface years later, inside a regulatory submission. And that is where the evidence gap gets expensive.

A target an AI prioritized, a variant classification a model suggested, a candidate a pipeline filtered out: if the program succeeds, those decisions become part of a dossier a regulator reads. Data-integrity expectations in GxP environments don’t exempt the discovery phase retroactively. What enters a submission must be attributable, contemporaneous, and reconstructable. The FDA’s draft guidance on AI in drug development points the same way: sponsors should be able to establish the credibility of AI models for their context of use, which in practice means showing what the model was, what it saw, and what it decided.

In research it sounds academic. In a submission review, a due-diligence data room, or a patent dispute, it is a liability attached to your most valuable asset. The molecule survives; the provenance of the decisions behind it has to survive with it.

Genetic data is the most protected data class in existence: a special category under GDPR, shielded by GINA in the US, residency-bound across the Gulf. AI touching genomic data needs two things in the same path: privacy enforcement on every access, and forensic evidence of every decision. Institutions usually build these as two systems. They are one problem.

every AI-influenced research decision recorded at the moment it happens: model and version, input provenance, the rule or threshold applied, the scientist who signed. Sealed immutably, independent of any single vendor or model. Not to slow research down, but to make its results defensible for the decade they’ll need to be. That is what we build.

― basis: FDA draft guidance, AI in drug & biological product development (2025) · 21 CFR Part 11 · EU GMP Annex 11/22 · GDPR Art. 9 (genetic data) · GINA (US)