Contextual AI for Precision Medicine
Context shapes biology and medicine: cellular environments modulate protein function, patient genetics alters disease risk, and intervention timing determines therapeutic outcomes. Many medical AI models still treat context as an afterthought, failing to generalize to new cell types and patient subgroups, which leads to misdiagnoses and suboptimal treatment decisions. In this talk, I will describe the foundations of contextually adaptive medical AI and the new opportunities it creates. I will first introduce PINNACLE, a multiscale contextual model that learns cell type specific protein representations. PINNACLE systematically interrogates therapeutic targets’ behavior across hundreds of cellular contexts, facilitating context-aware target nomination for autoimmune and neurodegenerative diseases. I will then present SHEPHERD, a few-shot method that integrates patients’ clinicogenomics with biomedical knowledge to diagnose thousands of rare diseases. Within the Undiagnosed Diseases Network, SHEPHERD improves diagnostic efficiency by twofold. Finally, I will describe CLEF, a controllable generative model that simulates cellular and patient trajectories under alternative interventions and temporal contexts to support counterfactual treatment design. Trained on 200 million routine laboratory measurements, CLEF enables earlier detection of autoimmune disorders. Overall, my research advances medical AI that scales to infinitely many contexts, producing predictions validated across cells, patients, and the full spectrum of disease.
Dr. Michelle M. Li is a Berkowitz Postdoctoral Research Fellow in the Department of Biomedical Informatics at Harvard Medical School. She earned her Ph.D. in Biomedical Informatics from Harvard University, and B.S. in Mathematical and Computational Science from Stanford University. Dr. Li develops medical AI models that dynamically adapt their outputs based on the contexts in which they operate, from cells to patients to clinical workflows, thereby enabling AI-assisted rare disease diagnosis and therapeutic target discovery. Her research has been recognized through the National Science Foundation Graduate Research Fellowship (2021), Albert J. Ryan Fellowship (2023), Harold M. Weintraub Graduate Student Award (2024), and UCLA Emerging Genomic Scientist Fellowship (2025). Dr. Li is a founding organizer of international conferences and workshops on foundational AI algorithms (Tutorials Chair at Learning on Graphs Conference), AI for drug discovery and development (NeurIPS Workshops), and clinical AI integration (Symposium on AI for Learning Health Systems).
Hybrid, but in-person attendance is encouraged.
https://bit.ly/MichelleLiHarvard