The Food and Drug Administration (FDA) hosted an insightful workshop on August 6, 2024 that brought together experts from diverse fields to explore the transformative potential of artificial intelligence (AI) in drug development. The event, held at FDA headquarters, drew hundreds of attendees eager to understand how AI could revolutionize the lengthy and costly process of bringing new medicines to market.
FDA embraces AI, citing over 300 AI-related submissions since 2016
Dr. Patrizia Cavazzoni, Director of the Center for Drug Evaluation and Research (CDER), set the tone in her opening remarks, revealing that the FDA has already received over 300 submissions containing AI elements since 2016. "We're seeing an increasing use of AI across the entire spectrum of drug development," Cavazzoni noted, emphasizing the agency's commitment to fostering innovation while ensuring safety and efficacy.
Multidisciplinary Expertise: The Key to AI Innovation in Pharma
The morning sessions focused on two critical aspects: the need for multidisciplinary expertise in AI development and strategies for leveraging existing data while creating new datasets.
In a panel on multidisciplinary expertise, Dr. Elazer Edelman of MIT and Harvard stressed the importance of what he termed "multi-fluency" – individuals with deep knowledge across multiple disciplines. "The obligation today is to teach the next generation to be multi-fluent, not just multidisciplinary," Edelman argued, suggesting this approach could accelerate innovation in the field.
Perspectives from different industries
Sam Glassenberg, CEO of Level Ex, offered a unique perspective from the video game industry. He demonstrated how techniques used in game development, such as procedural content generation, could be applied to create medical simulations and synthetic data for AI training. "The same tools we use to train doctors' neural networks, we can use to train artificial ones," Glassenberg explained.
Charles Fisher, founder of Unlearn AI, introduced the concept of "digital twins" – AI-generated patient simulations that could revolutionize clinical trials. Fisher highlighted the need for more AI researchers in drug development, stating, "The discipline that is currently missing is AI researchers, actually."
Digital Twins and In Silico Trials: Revolutionizing Clinical Research
The second panel explored the challenges and opportunities in healthcare data. Key points included:
- Itai Dayan of Rhino Health discussed federated learning approaches to enable analysis of distributed healthcare data while maintaining privacy.
- Wade Davis from Moderna highlighted the value of large-scale public datasets like UK Biobank in democratizing AI research in drug development.
- Merage Ghane of Ideas42 focused on the critical issue of bias in AI, emphasizing the need to consider potential biases from the very outset of projects. "AI gives us great insights into where the issues lie. When we see how it performs poorly in some samples, somebody should be asking why," Ghan noted.
Throughout the morning, a recurring theme was the need for clear regulatory guidance to balance innovation with patient safety. Dr. Cavazzoni announced that the FDA is developing new guidance on integrating AI into regulated work, adopting a risk-based approach focused on the context of use and potential consequences of AI applications.
Addressing AI Challenges: Data Quality, Bias, and Regulatory Uncertainty
As the workshop broke for lunch, attendees buzzed with excitement about the potential of AI to streamline drug development. Challenges around data quality, bias, and regulatory uncertainty remained at the forefront of discussions.
In silico trials and literature analysis
Luca Emili, CEO of InSilicoTrials, kicked off the afternoon session by introducing the concept of in silico trials, which use virtual patients to streamline drug development. "Cost reduction, time efficiency, and reducing ethical concerns are key advantages," Emili explained. He highlighted a recent publication, "Toward Good Simulation Practice," developed in collaboration with the FDA, as a step towards establishing guidelines in this emerging field.
Emili also showcased a practical application of AI in drug development, describing a system that dramatically accelerates literature analysis for identifying disease prevalence. "We moved a process that typically takes three to four months to just minutes," he stated, underlining the potential efficiency gains AI can bring to the industry.
Model performance and validation
The discussion then shifted to the critical issue of model performance and validation. Concerns were raised about current evaluation methods for AI models, particularly generative AI, which do not yet have good ways of evaluating the accuracy of gen AI models.
Dr. Subha Madhavan from Pfizer emphasized the importance of interpretability in AI models. "Interpretability and explainability are almost the same to me," Madhavan noted, stressing the need to not only explain how a model works but also interpret its outcomes for collaborators and end-users.
The Future of Drug Development: AI-Driven Efficiency and Safety
A key theme throughout the session was the challenge of encouraging stakeholder adoption of AI-driven approaches. Panelists agreed that focusing on end results and benefits, rather than the technology itself, is crucial for winning trust.
Regulatory considerations and approaches
As the discussion turned to regulatory considerations, panelists offered insights on potential approaches for the FDA:
- Providing incentives for adopting new technologies
- Developing guidelines based on use cases rather than specific technical details
- Taking a high-level approach to regulation, as suggested by Emili: "The more abstract you can be in defining some indications, the better it is"
Madhavan concluded the session with a call for collaborative learning: "How can we all learn together rapidly? And what are the tools that will enable us to learn rapidly?" she asked, suggesting mechanisms like challenges and competitions to engage the broader scientific community.
Balancing innovation and regulation of AI in drug development
As the workshop wrapped up, it was clear that while AI holds immense promise for revolutionizing drug development, significant challenges remain in validation, interpretation, and regulation. The FDA's ongoing engagement with experts in this field signals a commitment to navigating these challenges and fostering innovation while ensuring patient safety remains paramount.
