Rare but Common: Generative AI’s Potential on Data, Evidence, and Insight Generation in Rare Diseases

PURPOSE: This panel explores the potential of Generative AI (GenAI) in addressing the unique challenges of rare diseases. The discussion focuses on three key areas: 1) Data: Unlocking data through GenAI and large language models (LLMs) to enrich research data on rare diseases. 2) Evidence: Developing rare disease-specific knowledge graphs to better represent natural history and disease pathways. 3) Insights: Generating actionable insights to address unmet patient needs and support early trial recruitment, ultimately improving patient outcomes and healthcare decision-making. DESCRIPTION: About 10% of individuals worldwide are affected by rare diseases (more than 7,000 rare diseases). Rare diseases, while individually uncommon, are collectively more prevalent than often realized. For example, over 60% of oncology drugs are approved based on specific biomarkers, redefining certain cancer subtypes as rare diseases. The evolving evidence underscores the complexity of clinical research for rare diseases, compounded by limited patient populations, scarce evidence, and high treatment costs. Generative AI may offer a new opportunity to address these challenges. Dr. Xu (Yale) will open the session with an overview of GenAI advancements, focusing on how LLMs can generate more data for rare diseases and their phenotypes from clinical notes. Dr. Wang (Tulane) will discuss leveraging LLMs and knowledge graphs to extract data from clinical, literature and social media, enabling a deeper understanding of disease history and pathways. Dr. Weng (Columbia) will present the use of fine-tuned LLMs for early diagnosis of rare diseases and genetic testing recommendations using RAG models. Finally, Dr. Wang-Silvanto (Astellas) will provide an industry perspective on the alignment of robust data and evidence to address unmet needs, inform early trial recruitment and support decision-making in pharmaceutical settings. The session concludes with an interactive Q&A and digital polling to encourage audience participation.