P07-18

Lead generation of a V-ATPase inhibitor using molecular generative AI

Taiyo TOITA *1Kano SUZUKI2, 3Shoichi ISHIDA1Akira KATSUYAMA4, 5Satoshi ICHIKAWA4, 5Masateru OHTA6Mitsunori IKEGUCHI1, 6Takeshi MURATA1, 2, 3Kei TERAYAMA1, 7, 8

1Graduate School of Medical Life Science, Yokohama City University
2Graduate School of Science, Chiba University
3Membrane Protein Research Center, Chiba University
4Faculty of Pharmaceutical Science, Hokkaido University
5Center for Research and Education on Drug Discovery, Hokkaido University
6HPC- and AI-driven Drug Development Platform Division, RIKEN Center for Computational Science
7RIKEN Center for Advanced Intelligence Project
8MDX Research Center for Element Strategy, Tokyo Institute of Technology
( * E-mail: w245418f@yokohama-cu.ac.jp )

Vancomycin-resistant Enterococcus faecium (VRE) is a bacterium that causes nosocomial infections. VRE is resistant to many antibiotics, narrowing treatment options and spreading globally. Because of the health risks and limited treatment options caused by VRE, the World Health Organization (WHO) has listed VRE as one of the priority pathogens requiring new antimicrobials [1]. VRE grows predominantly under alkaline conditions after antibiotic administration by specifically expressing a Na+-transporting V-ATPase [2, 3]. We have discovered a hit compound that binds specifically to Na+-transporting V-ATPase and exhibits inhibitory activity. This compound contributes to inhibiting VRE growth by binding between the multiple subunits called the Vo region of V-ATPase. However, although this inhibitor is active in the small intestine, it is not effective in the large intestine. Therefore, there is room for improvement in the binding affinity and membrane permeability of this inhibitor. In this study, we performed lead generation for the V-ATPase inhibitor with a focus on improving binding affinity considering membrane permeability. For lead generation, we employed ChemTSv2 [4], a molecular generative AI based on recurrent neural networks and an exploration system with Monte Carlo tree search. When performing lead generation for the V-ATPase inhibitor, we considered several conditions to generate the structure. To fit into the small-sized binding pocket, we trained the generative AI using various datasets with limited molecular weight. In addition, we designed some functions that evaluate docking scores and interactions with surrounding residues to improve binding affinity. To keep the location of the part of the known inhibitor that contributes to the activity, we filtered by RMSD of the common structure. Here, we report the methods and the results.

[1] Willyard, C. The drug-resistant bacteria that pose the greatest health threats, Nature, 2017, 543, 15
[2] Murata, T. et al. Intracellular Na+ regulates transcription of the ntp operon encoding a vacuolar-type Na+-translocating ATPase in Enterococcus hirae, J. Biol. Chem, 1996, 271, 23661-23666
[3] Murata, T. et al. The ntpJ gene in the Enterococcus hirae ntp operon encodes a component of KtrII potassium transport system functionally independent of vacuolar Na+-ATPase, J. Biol. Chem, 1996, 271, 10042-10047
[4] Ishida, S. et al. ChemTSv2: Functional molecular design using de novo molecule generator, WIREs Comput. Mol. Sci, 2023, 13, e1680