FDA Releases Two Discussion Papers on AI/ML
The agency covers two parallel topics: AI/ML in drug/biologics development and manufacturing
The US Food and Drug Administration (FDA) today published two discussion papers, providing an in-depth exploration of the role of artificial intelligence (AI) and machine learning (ML) in two pivotal sectors of the pharmaceutical industry: drug development and drug manufacturing. The papers reflect the FDA's evolving viewpoint on the subject with a combination of optimism and caution and a clear emphasis on transparent, interpretable, and effective uses of AI/ML models.
The first paper sheds light on the myriad roles of AI/ML in drug development, a field where these technologies are already redefining the landscape. The FDA acknowledges that AI/ML has become instrumental in the identification, selection, and prioritization of drug targets, in addition to screening compounds and designing innovative therapies. Moreover, the FDA recognizes AI/ML's far-reaching impact on optimizing preclinical studies. These studies, encompassing critical components like toxicity testing and pharmacokinetic modeling, represent an area where the FDA arguably has the most hands-on experience with AI/ML.
The FDA stresses that the incorporation of AI/ML into drug development is not devoid of challenges. AI/ML technologies require vast amounts of high-quality data for effective operation, a requirement that poses a significant challenge in a field where data is often scarce or incomplete. Furthermore, the FDA highlights the interpretability and transparency of AI/ML algorithms as a major hurdle, as their complex nature can make it difficult for regulators to evaluate their safety and efficacy.
The second paper focuses on the potential of AI in drug manufacturing. The FDA outlines how AI could drastically enhance efficiency and precision by predicting, controlling, and optimizing manufacturing processes in real time. The integration of AI is projected to boost production efficiency, minimize waste, and enhance product quality, potentially advancing the FDA’s long-standing interest in modernizing the process through a shift towards continuous manufacturing standards. In 2019, Dr. Janet Woodcock of the FDA provided a testimony before the House Committee on Energy and Commerce, Subcommittee on Health, on advanced manufacturing technologies, including continuous manufacturing. Dr. Woodcock underscored the importance of these innovations in improving drug quality, addressing shortages of medicines, and speeding time-to-market while enabling US-based pharmaceutical manufacturing to regain its competitiveness with China and other foreign countries and potentially ensuring a stable supply of drugs critical to the health of patients in the US.
One of the emerging themes in AI-driven drug manufacturing is the convergence of the Internet of Things (IoT), cloud, and edge computing. The orchestration of an interconnected IoT network, composed of sensors and instruments, can streamline data exchange, supplying AI models with a ceaseless stream of information. Such networks can be rendered more effective when combined with the capabilities of cloud and edge computing. These technologies can process data in real time, either at the source or in close proximity, amplifying the responsiveness and efficiency of AI algorithms, which can adapt to changes in the manufacturing process to optimize performance and efficacy. The FDA recognizes these innovations come with significant responsibilities. Similar to the case in drug development, the agency calls for transparent and explainable AI systems. Understanding the decision-making process of AI is crucial for its appropriate functioning and for fostering trust in its predictions. The FDA also discusses regulatory considerations, citing 21 CFR 211 Subparts D and J, which govern drug manufacturing processes in the US, and emphasizes the need for responsible and ethical use of AI models.
Through these discussion papers, the FDA provides an overview of the potential benefits and challenges of AI in drug development and manufacturing. While the agency's perspective will continue to evolve, their emphasis on explainable AI models, effective data management, and the potential of IoT and edge computing reflects on key issues currently lacking consensus in the broader community. The papers also affirm the FDA's commitment to ongoing stakeholder engagement, with the goal of establishing a future framework and more detailed guidances on the topic through continuous dialogue and collaboration, which includes an upcoming workshop with the Product Quality Research Institute (PQRI).