P01-21

Generation of a suitable structure for SBDD by AlphaFold2 via Genetic Algorithm Parameter Search

Keisuke UCHIKAWA *, Kairi FURUI, Masahito OHUE

Department of Computer Science, School of Computing, Institute of Science Tokyo
( * E-mail: uchikawa.k.ac@m.titech.ac.jp )

Structure-based virtual screening (SBVS), which utilizes protein conformational information, has gained significant attention in recent years due to its potential to discover highly novel drug candidate compounds more effectively than other methods. However, a major challenge lies in the variability of screening accuracy depending on the conformation of the target protein. In this study, we focused on the combination of AlphaFold2, a representative method for protein structure prediction, and SBVS. We explored a method to improve the accuracy of SBVS using predicted structures by optimizing the parameters used in AlphaFold2 with a genetic algorithm.
Specifically, we used only shallow MSA (multiple sequence alignment) for prediction with AlphaFold2, and then explored how to introduce mutations based on docking scores using a genetic algorithm. As a result, we obtained predicted structures for CXCR4 that exhibited SBVS performance significantly surpassing that of the PDB structure. For KIF11, although the performance was slightly inferior to the PDB structure, we were able to generate predicted structures with performance that could not be achieved by the standard predictions of AlphaFold2. These results suggest that the application range of SBVS can be expanded by utilizing predicted structures from AlphaFold2.