Jure Leskovec
Professor of Computer Science
Stanford University

Hosted by: Jian Ma, Ray and Stephanie Lane Professor of Computational Biology

Talk Title: From Relational Foundation Models to the AI Virtual Cell: Building the Next Generation of Intelligent Systems

Abstract: Foundation Models have transformed learning from unstructured data, but the world’s structured and relational data—driving decisions in enterprises and science—remain underutilized. I will present Relational Foundation Models (RFMs), a new class of pre-trained models that enable in-context learning over relational data, predicting outcomes such as engagement, churn, or fraud without supervision or feature engineering. RFMs combine table-agnostic embeddings, relational transformers, and SQL-like prompting to turn predictive AI from model-building into querying. Extending these ideas to biology, I will introduce the AI Virtual Cell (AIVC)—a large neural model that represents and simulates molecular and cellular behavior, offering a new paradigm for data-driven biological discovery. Finally, I will discuss Biomni, a general-purpose biomedical AI agent that autonomously plans and executes complex scientific workflows across diverse biomedical domains, acting as a virtual AI biologist that accelerates research and discovery. Together, these efforts outline a vision for foundation models that reason, simulate, and collaborate across structured, relational worlds—from enterprises to living cells.