Comment from NEXUS CONCORDAT INC.

VIA ELECTRONIC SUBMISSION May 12, 2026 Dockets Management Staff (HFA-305) Food and Drug Administration 5630 Fishers Lane, Rm. 1061 Rockville, MD 20852 RE: Docket No. FDA-2025-D-6131 General Considerations for the Use of New Approach Methodologies in Drug Development; Draft Guidance for Industry To Whom It May Concern, Nexus Concordat Inc. submits this comment in response to the FDA's draft guidance, General Considerations for the Use of New Approach Methodologies in Drug Development. I. Commenter Credentials & Standing The undersigned is the inventor of record for four U.S. provisional patent applications (Nos. 63/939,190; 63/962,385; 63/988,485; 64/034,536) directly relevant to the architectural questions raised in this draft guidance. My expertise is grounded in biochemistry, molecular biology, and clinical laboratory practice subject to CLIA standards. Furthermore, I bring concurrent direct-care experience as a Certified Nursing Assistant in long-term care, where the practical implications of medication management and data integrity are a matter of daily clinical observation. II. Executive Proposals This comment presents two primary proposals for the final guidance: 1. Architectural Credentialing as a Distinct Regulatory Tier The draft guidance's validation framework assumes that the underlying models are inherently reproducible and auditable. However, AI/ML NAMs built upon commercial Large Language Models do not hold these properties by construction: training data is undisclosed, weights are not byte-stable, and inference is non-deterministic. We propose that the final guidance distinguish between: Tier 1 (Substrate-Level Credentialed): Architectures satisfying 21 CFR Part 11 and ALCOA+ requirements by construction. Tier 2 (Procedurally Wrapped): Approaches achieving auditability through external governance layers applied to a non-deterministic core. The Regulatory Analogue: This mirrors the institutional accountability framework imposed by CLIA, where structural requirements must be present within the testing method itself rather than appended as an afterthought. 2. Mental Health and Psychiatric Medicine as the Canonical Use Case Diagnostic chains in psychiatric medicine are fundamentally linguistic, utilizing instruments such as the HAM-D, PHQ-9, and C-SSRS to operationalize DSM-5-TR criteria. While the inference is constructed from language, the treatment is selected for its biochemical action on the underlying neurochemistry. A substrate-grounded AI/ML NAM that computes neurochemical dynamics through ordinary differential equations—rather than natural-language tagging—provides a biochemical attestation floor that linguistic evidence alone cannot provide. We propose that the final guidance recognize domains resting on linguistic interpretation of subjective report as primary candidates for the Tier 1 acceptance pathway. III. Summary of Substantive Comment The attached full comment maps the credentialing-chain proposal to the FDA’s four validation principles (Context of Use, Human Biological Relevance, Technical Characterization, and Fit-for-Purpose). It includes seven specific recommendations for the final guidance and provides a working substrate-level reference implementation for review. IV. Conclusion I respectfully request that the FDA consider these proposals, particularly within the principles of Technical Characterization and Fit-for-Purpose, to ensure that the evolution of AI in drug development remains anchored in verifiable biological and mathematical truth. Sincerely, Marjorie Gayle McCubbins, BSc Founder and Chief Executive Officer Nexus Concordat Inc. Hamilton, Alabama