MFBind - A Multi-Fidelity Approach for Evaluating Drugs in Generative Modeling

Explore a 55-minute conference talk on MFBind, a multi-fidelity approach for evaluating drugs in generative modeling. Dive into the challenges of current generative models for drug discovery and learn about a novel method that integrates docking and binding free energy simulators to achieve an optimal balance between accuracy and computational cost. Discover how MFBind utilizes a pretraining technique and linear prediction heads to efficiently fit small amounts of high-fidelity data. Examine the extensive experimental results demonstrating MFBind's superior performance in surrogate modeling and its ability to enhance generative models with higher quality compounds. Follow along as speaker Peter Eckmann from Valence Labs presents the background, methodology, results, and conclusions of this innovative approach. Engage with the Q&A session to gain further insights into this cutting-edge research in AI-driven drug discovery.

- Intro + Background
- MFBind
- Results: Surrogate Modeling
- Results: Generative Modeling
- Conclusions
- Q+A