Title : Artificial intelligence and digital twins in drug product development: A regulatory science perspective
Abstract:
Background & Rationale: Regulatory agencies including the FDA, EMA, and ICH are increasingly recognizing the value of Artificial Intelligence (AI) and Digital Twin (DT) technologies in pharmaceutical development. FDA's Emerging Technology Program and guidance on Pharmaceutical Quality for the 21st Century (PAT/QbD) have opened pathways for sponsors to prospectively incorporate computational models into CMC submissions. Despite growing regulatory receptiveness, practical frameworks for integrating AI/DT outputs into ANDA and NDA filings — particularly for complex sterile injectables and combination products — remain underexplored in the literature.
Objectives: This presentation examines AI and digital twin technologies through a regulatory science lens, with three core objectives: (1) to evaluate how AI-generated predictive models for critical quality attributes (CQAs) — such as particle size, viscosity, and chemical stability — can be documented and defended within CMC Module 3 eCTD submissions; (2) to explore how digital twin simulations of sterile manufacturing unit operations (e.g., media milling, fill-finish, lyophilization) can serve as process understanding evidence under ICH Q8/Q10 principles; and (3) to assess FDA's current expectations for validation, risk management, and transparency when AI/DT tools are used in development decision-making.
Regulatory Framework & Industry Application: Drawing from 15+ years of CMC development and NDA/ANDA authorship experience — including sterile injectable suspensions, prefilled syringes, and autoinjector combination products — the presenter maps AI/DT applications to specific regulatory touchpoints. Machine learning models built using JMP (SAS) and Design Expert, trained on DoE-generated datasets, were used to establish design spaces and define proven acceptable ranges (PARs) for injection formulations. These outputs were incorporated into Module 3.2.P.2 (Pharmaceutical Development) sections, with model metadata, training data summaries, and validation statistics included as supporting documentation. Digital twin simulations of media-milling scale-up (bench to commercial) were similarly anchored to ICH Q8 process understanding requirements, enabling reduced physical batch campaigns while maintaining regulatory defensibility. Risk management of AI/DT model uncertainty was addressed through structured FMEA-based approaches aligned with ICH Q9, ensuring that model limitations were transparently communicated to reviewers.
Findings & Regulatory Implications: Key findings indicate that AI/DT outputs, when properly validated and documented, are accepted as supplementary process understanding evidence by FDA reviewers, particularly when paired with traditional experimental data. Predictive stability models enabled proactive specification setting, materially reducing the frequency of FDA Complete Response Letters and Information Requests in reviewed submissions. Digital twin scale-up evidence was successfully used to support commercial manufacturing without additional PPQ batches in select programs, representing both a regulatory efficiency and resource savings. Remaining challenges include the absence of standardized AI model reporting formats in eCTD, variability in reviewer familiarity with computational methods, and evolving FDA guidance on AI/ML lifecycle management for drug manufacturing.
Conclusion: AI and digital twin technologies offer compelling regulatory science advantages when applied thoughtfully within established ICH/FDA quality frameworks. For sterile injectable and combination product developers, these tools can shorten development timelines, reduce batch consumption, and generate richer CMC narratives — provided that model validation, uncertainty quantification, and risk management are rigorously addressed. Proactive engagement with FDA's Emerging Technology Program is recommended for sponsors seeking to incorporate novel computational methods into IND/NDA filings.
Keywords: Artificial Intelligence, Digital Twin, Regulatory Science, CMC, QbD, ICH Q8/Q9/Q10, NDA/ANDA, Sterile Injectables, FDA Emerging Technology, Process Understanding, eCTD Module 3.
