AI and ML in Predicting Biopharmaceutical Product Profile Approvability

In a fascinating session, “Applying Machine Learning and Artificial Intelligence for Predicting Product Profile Approvability” session, industry leaders dissected the evolving roles of artificial intelligence (AI) and machine learning (ML) in pharmaceutical regulatory processes at the 2024 DIA conference. Dr. Romi Singh, an influential figure in global drug registration and AI/ML application in regulatory sciences and Founder and Principal Advisor at GRA Advisors, chaired the session.

The panel included Amin Osmani, CEO of Cedience Inc.; Subha Madhavan, PhD, Vice President and Head of AI/ML at Pfizer; and Lily Li, JD, founder and president of Metaverse Law. Each brought a unique perspective on AI and regulatory frameworks, ensuring a multifaceted discussion on AI’s application and regulatory challenges in the pharmaceutical industry.

Cloud Computing and AI for Regulatory Predictions

Amin Osmani, CEO of Cedience Inc., started the panel by exploring how his company uses cloud computing to enhance regulatory predictions. Osmani highlighted that while AI’s textbook definition focuses on replicating human intelligence, the real value lies in its practical applications. “Artificial intelligence is not just about mimicking human intelligence,” he explained. “It’s about creating tools that solve real-world problems efficiently.”

Osmani’s team incorporates cloud computing to analyze historical approval data, identify patterns, and forecast the probable outcomes of new drug applications. “By harnessing cloud resources, we’ve been able to create adaptive models that can quickly process and analyze data, making our predictions both faster and more reliable,” he added. This approach accelerates the regulatory process and ensures more robust data handling and storage capabilities, catering to the needs of life sciences organizations dealing with high volumes of sensitive information.

Pfizer’s AI-Driven Approaches

Subha Madhavan enthralled the audience with a detailed look at Pfizer’s use of AI and ML. “Our focus is on enhancing R&D productivity,” she stated, offering a glimpse into Pfizer’s innovative AI applications. One of the most compelling examples was Pfizer’s global challenge of using AI to generate clinical study reports (CSRs). The challenge aimed to determine whether AI could outperform humans in generating initial drafts of these critical documents.

Madhavan explained that this challenge involved 21 companies globally experimenting with fine-tuning models using proprietary tools and open-source models like GPT-3. The results were promising, with AI-generated drafts showing 80% factual accuracy. However, she pointed out that human intervention was still necessary to ensure the clarity and precision of the final documents. “AI could rapidly generate drafts, but humans were indispensable for refining the nuanced language and ensuring the accuracy of complex medical data,” Madhavan noted. Through this challenge, Pfizer learned that generative AI’s power lies in its ability to expedite routine tasks, allowing experts to focus on higher-order cognitive functions such as analysis and interpretation.

She elaborated on the importance of distinguishing between analytic AI and generative AI, emphasizing that while analytic AI helps classify data and predict outcomes based on historical trends, generative AI can create content and interpret large datasets. For example, analytic AI assists in sorting patients by conditions or predicting treatment efficacy, while generative AI predicts the next word or image, generating reports or scientific illustrations. Integrating both forms of AI could dramatically transform drug development processes, making them more efficient and insightful.

Legal Dimensions of AI Regulation

Lily Li provided essential legal perspectives, emphasizing the need for a unified regulatory framework to guide AI development and application. “We need a common federal body of law to define AI,” she asserted, highlighting the risks of fragmented regulations that could hamper innovation. She compared the current regulatory uncertainty to the early days of the automotive industry, where varied rules initially impeded progress until standard regulations were established. Without comprehensive legislation, she warned, AI technology might face similar hurdles that could inhibit its growth and deployment in critical sectors like pharmaceuticals.

Li spoke about the proliferation of AI legislation, such as California’s comprehensive privacy laws, including stringent AI regulations. Li said these can sometimes lead to excessive rules, stifling innovation. For instance, the California Consumer Privacy Act (CCPA) included clauses allowing consumers to object to automated decision-making, resulting in extensive new regulations that many argue exceed the original legislative intent. “We need thorough yet balanced legislation that protects consumers without overburdening innovators,” Li insisted. To support her point, she referenced examples from the data privacy domain where a lack of clear, unified regulations led to overly burdensome compliance requirements, suggesting the same risk for AI technologies if not appropriately regulated.

Practical Applications and Case Studies

Dr. Singh illustrated the practical application of AI through a compelling case study from his career at Pfizer. He shared how sentiment analysis of regulatory meeting minutes was utilized to predict product success. The project involved analyzing 20 years of regulatory documents, including meeting minutes and email communication, through a specialized app that scored sentiments to gauge the likelihood of regulatory approval. By quantifying sentiment trends over time, the team could identify positive or negative shifts in regulatory feedback, providing critical insights for strategic decision-making.

Using sentiment scoring, the team could track the sentiment trends over time, identifying positive or negative shifts in regulatory feedback. For instance, a notable drop in positive sentiment scores around 2009 and 2013 coincided with critical input from the FDA, which eventually impacted the product’s marketability despite its approval. Conversely, a spike in positive sentiments in 2018 reflected favorable regulatory reviews, aligning with the product’s subsequent launch. This case study underscored how AI could be leveraged to reduce guesswork and increase predictive accuracy, making regulatory processes more efficient and transparent.

Future Directions and Challenges

The session concluded with a forward-looking dialogue about the future of AI in regulatory science. Panelists emphasized the need for ongoing evaluation and standardization to ensure AI-generated outputs are both reliable and useful. Subha Madhavan mentioned the necessity for robust evaluation metrics and continuous improvement cycles, while Lily Li stressed the importance of a regulatory framework that can adapt to technological advances. The panel discussed the importance of collaboration between regulatory bodies and tech innovators to pave the way for AI advancements.

Interactive DIA Q&A sessions allowed attendees to engage deeply with the panelists, discussing topics such as AI’s role in clinical trials, ethical considerations in AI deployment, and the future of AI in drug development. Questions reflected the audience’s keen interest in leveraging AI to streamline and enhance regulatory processes. Panelists responded with actionable insights, reinforcing the importance of a collaborative approach in tackling the challenges and maximizing the benefits of AI technologies.