Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products Guidance for Industry and Other Interested Parties; Draft Guidance

The commenter provides a detailed technical breakdown of machine learning (ML), deep learning (DL), and generative AI (GenAI) to clarify their differences and applications in data science. They suggest specific guidelines for the draft guidance, including distinguishing between GMP and non-GMP operations and outlining a multi-step process for model development and evaluation.

The commenters, who are academics from the University of Colorado, Stanford, and Harvard, support the FDA's draft guidance for its "context-of-use" framework and risk matrix. However, they argue that the guidance should be strengthened by mandating disaggregated performance testing for subgroups, specifying lifecycle maintenance requirements, and including independent external validation and community involvement.

Bhasker Sambar, a Senior Manager in External R&D with experience in pharmaceutical development, provides technical feedback to improve the draft guidance. He recommends expanding the scope to include combination products, extending process examples, clarifying requirements for third-party AI models, and addressing cybersecurity risks.

Nexus Concordat Inc. supports the draft guidance but proposes expanding it to recognize a specific category of "compiler-verified synthetic twins" based on biochemistry-closed opcode sets. They argue that these substrate-grounded AI systems offer deterministic, bit-for-bit reproducible results and should be subject to tailored credibility-assessment expectations based on their architectural properties rather than just behavioral evidence.

A pharmaceutical or biotechnology sponsor (implied by the use of "sponsors" and "industry" context) supports the draft guidance for using AI in regulatory decision-making. They suggest that the FDA provide further clarification on how to document human accountability and adequacy determinations when AI models are integrated as automated, embedded components of regulatory workflows.

The commenter suggests specific revisions to the draft guidance to improve clarity and readability for sponsors and stakeholders. They recommend relocating and rewording the explanation regarding why the hypothetical examples only cover the first three steps of the framework.

The commenter is asking clarifying questions regarding the FDA's requirements for using AI-predicted design spaces and risk-based reductions in traditional DOE trials. They do not take a clear position for or against the proposed guidance.

Dr. Lamia Anwar, representing the Egyptian Drug Authority, supports the draft guidance as an important step toward responsible AI use in drug development. The commenter suggests specific improvements to the guidance, including clearer scope definitions, objective risk scoring, explicit bias mitigation strategies, and better alignment with international harmonization standards.

CIMCON Software, a company specializing in AI and GRC software, supports the FDA's draft guidance on the use of AI in regulatory decision-making. They provide specific recommendations to enhance the guidance regarding 21 CFR Part 11 compliance, third-party risk management, explainability, data governance, fairness, and the unique risks associated with LLMs and Generative AI.

Eigen Bio, a biotechnology company, supports the draft guidance and suggests several improvements to enhance its scope and practical application. They recommend including preclinical AI models, providing specific guidance for federated learning, creating accelerated pathways for orphan drugs, and establishing a centralized AI advisory point for early-stage sponsors.