Comment from EIGEN Bio

Comments Submitted by Eigen Bio RE: Docket No. FDA-2024-D-5095 Title: Draft Guidance for Industry — Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products Eigen Bio is a biotechnology company developing AI-powered platforms for the discovery and optimization of RNA and biologics therapeutics. We are pleased to provide the following comments in response to FDA’s draft guidance. 1. Clarify Scope to Include Preclinical AI (especially for RNA/Biologics) Reference: Section II, Scope The draft guidance currently excludes drug discovery but does not clarify treatment of AI models used for preclinical prediction of efficacy, safety, or manufacturability, especially for RNA therapeutics. These models are often foundational to IND-enabling studies and regulatory pathways. Recommendation: Clarify that AI models used to generate data informing IND submissions (e.g., toxicity prediction, off-target risk, or delivery efficiency) are within the scope of this guidance, even if initiated during discovery. 2. Add Explicit Guidance for Federated Learning Models Reference: Section IV.B, Lifecycle Maintenance Federated learning enables collaborative model improvement without centralized data pooling — especially relevant for institutions with sensitive genomic, clinical, or proprietary manufacturing data. However, federated models require unique lifecycle validation, versioning, and auditability of decentralized weight updates. Recommendation: Include a carve-out or appendix section detailing best practices for federated learning validation, data protection standards, and expectations for auditing federated AI model performance over time. 3. Create Accelerated Pipelines for Orphan Drug and Drug Repurposing AI Models Context: Regulatory Innovation Opportunity (not currently addressed in detail) Drug repurposing and orphan indications can benefit significantly from AI-based predictive modeling (e.g., transcriptomic alignment, off-target analysis, patient stratification). Yet current guidance does not propose any fast-track regulatory framework for these lower-risk or underserved domains. Recommendation: Propose an AI Fast Track pathway for: Predictive models used to identify efficacy signals in rare diseases. Repurposing models for already-approved drugs. Cases where AI models help design targeted trials with limited patient populations. This would align with the spirit of accelerated approval and expanded access, while still maintaining FDA oversight. 4. Balance Between Deep Learning Accuracy and Explainability Reference: Section IV.A.4.a, Model Architecture Many state-of-the-art deep learning models (e.g., for RNA folding or protein binding prediction) require multi-layered embeddings that may reduce interpretability. In some cases, enforcing explainability may degrade performance and limit the ability to identify nonlinear biological correlations crucial for therapeutic success. Recommendation: Acknowledge that accuracy and explainability may be inversely correlated in some contexts. For high-risk models, allow explainability to be partially substituted by: Robust validation on external test sets, Calibrated uncertainty scores, and Lifecycle monitoring of output confidence under real-world deployment drift. 6. Centralized AI Advisory Point for Early-Stage Sponsors Reference: Section IV.C, Early Engagement For companies developing AI-native platforms not clearly fitting within existing FDA engagement channels (e.g., not yet at IND or manufacturing), a centralized AI engagement point would greatly enhance communication. Recommendation: Establish a cross-center AI Engagement Office or Portal to coordinate feedback for: Foundational AI models, Cross-cutting algorithmic validation issues, Sponsors without a clear existing FDA review team. Closing Remarks FDA’s leadership in offering structured AI guidance is commendable. We encourage the agency to continue refining the framework to address emerging technologies, data privacy constraints, and the practical trade-offs inherent in advanced AI systems. Eigen Bio looks forward to working collaboratively to advance safe, efficient, and AI-enabled therapeutic development. Respectfully, Sang Lee Eigen Bio sang@eigeninsights.com