Harmonic Self-Conditioned Flow Matching for Multi-Ligand Docking and Binding Site Design

Explore a comprehensive talk on advanced AI techniques for protein-ligand binding and binding site design. Delve into HarmonicFlow, an improved generative process for 3D protein-ligand binding structures, and FlowSite, which extends this model to jointly generate protein pocket residue types and molecule binding structures. Learn about the self-conditioned flow matching objective, the application of flow models to docking problems, and how these approaches outperform state-of-the-art methods in simplicity, generality, and performance. Discover the potential impact of these techniques on drug discovery, enzymatic catalysis, and energy storage applications. Gain insights into the technical aspects of 3D flow matching generative models, structure self-conditioning, and the use of harmonic priors for multi-ligand docking. Examine the performance of these models on the PDBBind dataset and their ability to recover residue types of known binding sites.

- Intro
- Flowsite: binding site design
- HarmonicFlow: flow matching generative models training
- 3D flow matching generative models training
- 3D flow matching generative models inference
- Making flow matching work for 3D structures
- Harmonic prior for multi-ligands
- Structure self-conditioning
- Docking on PDBBind dataset
- Extending HarmonicFlow to discrete residue types
- Recovering residue types of known binding sites
- Q+A