Discovery leaves long-lived consequences
Model-assisted candidate selection and genomic analysis can shape work that later enters regulated development. If the model, data and thresholds behind an early choice cannot be reconstructed, the decision is difficult to defend.
Credibility needs provenance
Research teams should preserve the context of an AI-supported choice: the model version, relevant data, output and expert interpretation. A result alone does not tell a later reviewer why it was considered credible.
Privacy and evidence must coexist
Genetic data requires careful protection. A useful evidence trail records what was done and by whom without turning sensitive research data into an uncontrolled archive.
Basis
FDA draft guidance on AI in drug development (2025) · 21 CFR Part 11 · EU GMP Annex 11/22 · GDPR Article 9