
e-HAIL Event
Molecular Simulations for Chemical Foundation Models
Large-scale models for bioactivity–chemical foundation models–promise to integrate diverse evidence and facilitate a broad range of tasks in drug-discovery. The playbook for scaling that has worked for large-language models with web-scale data, however, is limited by the naive use of physical, chemical, and biological empirical evidence of drug-like molecules. In this talk, I will discuss strategies for leveraging molecular simulations to enable theory to “get more” from available experimental data. Specifically, I will discuss three projects, 1) ChemPile, our effort to curate chemical foundation model training data; 2) Docking to Novel pocKets (DoNK), leveraging protein design to find an orthogonal basis of binding sites; and (3) Docking to Known pocKets (DoKK), towards building structure activity relationships for all target based activities reported in the literature.
Zoom information will be sent to e-HAIL members.