MF-LAL: Drug Compound Generation Using Multi-Fidelity Latent Space Active Learning

• DOI: 10.48550/arxiv.2410.11226
• Publication Date: 2024-10-14
• Authors: Peter Eckmann; Dongxia Wu; Germano Heinzelmann; Michael K. Gilson; Rose Yu

Current generative models for drug discovery primarily use molecular docking as an oracle to guide the generation of active compounds. However, such are often not useful in practice because even compounds with high scores do consistently show experimental activity. More accurate methods activity prediction exist, dynamics based binding free energy calculations, but they too computationally expensive a model. To address this challenge, we propose Multi-Fidelity Latent space Active Learning (MF-LAL), modeling framework that integrates set oracles varying cost-accuracy tradeoffs. Unlike previous approaches separately learn surrogate model and model, MF-LAL combines multi-fidelity into single framework, allowing more higher quality samples. We train novel learning algorithm further reduce computational cost. Our experiments on two disease-relevant proteins produces significantly better than other approaches.