SC01: Biomedical AI Foundation Models and Applications in Drug Development

Artificial intelligence (AI) is transforming biomedical research and drug development, from accelerating molecule discovery to analyzing complex clinical data. Central to this transformation are foundation models: large-scale AI models pre-trained on vast data sets that can be adapted to a wide range of downstream tasks. A landmark example is the success of large language models (LLMs). There is a growing need for statisticians and data scientists to understand both the fundamentals and practical applications of these modern AI methods. This short course introduces foundation models and their applications relevant to drug development. The course is structured into three lectures. Lecture 1 covers fundamentals and historical evolution to build understanding of how and why foundation models work, which is essential for rigorous application in high-stakes biomedical and clinical development settings. Lecture 2 focuses on LLMs, a key engine behind modern AI and most current applications. We will demystify their capabilities from model architecture, pre-training and post-training paradigms, inference-time adaptation to more advanced AI agents, tailoring to biomedical and drug development context. Lecture 3 presents examples of biomedical foundation models spanning multiple modalities, including text, medical imaging, and omics, illustrating how general architectures adapt to domain-specific challenges and multimodal clinical realities. Participants will gain both conceptual understanding and practical insights for applying foundation models in drug development and biomedical research.