PoseBusters: Evaluating AI-based Docking Methods for Physical Validity

Explore a critical analysis of AI-based protein-ligand docking methods in this 34-minute conference talk by Martin Buttenschoen from Valence Labs. Delve into the development of PoseBusters, a Python package designed to evaluate the physical plausibility of molecular structures generated by deep learning-based docking methods. Learn about the importance of assessing these methods beyond just RMSD to native binding modes, and discover how PoseBusters compares five deep learning-based docking methods with two established standard docking methods. Gain insights into the current limitations of deep learning approaches in outperforming classical docking tools and understand the potential for improving future predictions through the use of molecular mechanics force fields.

- Intro + Motivation
- Posebusters
- Results
- Conclusion
- Q+A