AI-driven fragment-based drug design of a grain yield enhancer in rice
Tatsuki AKABANE *1, 2, Tomoki YONEZAWA3, 4, Yugo SHIMIZU1, Kazuyoshi IKEDA1, Etsuko KATOH5, Naoki HIROTSU2
1Center for Computational Science, RIKEN
2Graduate School of Life Sciences , Toyo University
3Faculty of Pharmacy, Keio University
4Lifematics Inc.
5Graduate School of Food and Nutritional Sciences, Toyo University
[Purpose]
Fragment-based drug discovery (FBDD) enables efficient exploration of diverse chemical structures with relatively few candidates, followed by optimization through structural elaboration. Artificial intelligence (AI) tools, including generative models for small-molecule design, have advanced medicinal chemistry, yet their application in agrochemical research remains limited.
THOUSAND-GRAIN WEIGHT 6 (TGW6), a hydrolase of IAA-glucose, negatively regulates rice grain yield. We hypothesized that chemical inhibition of TGW6 could enhance yield of rice. In this study, we performed fragment screening to identify TGW6-binding molecules and designed antagonists using AI-assisted methods.
[Methods]
Fragment screening was conducted using 19F NMR and biochemical assays. Compounds containing hit fragment structures were selected from commercial libraries. Virtual screening with docking simulation and molecular dynamics simulation were conducted based on a crystal structure of TGW6 using Flare (Cresset, Cambridgeshire, U.K.). Effects of the selected compounds on yield components in rice were assessed in planta. Molecular generative AI (ChemTS) was employed to design novel compounds starting from hit fragments and based on 3D similarity against initial active candidates.
[Results and Discussion]
Two hit fragments were identified from over 1,000 molecules. Substructure search retrieved approximately 8,000 analogues, of which nine were prioritized by docking score and RMSD. Field trial showed several compounds significantly improved the yield components in rice.
Molecular generative AI generated approximately 200,000 molecules based on each hit fragment. Docking simulation predicted that some generated analogues had stronger TGW6 binding than both hit fragments and initial active candidates.
[Conclusion]
AI-generated analogues, the superior predicted binding affinity, are considered as potential TGW6 antagonists increasing rice yield components. Therefore, it is predicted that FBDD combined with AI-based design enables generation of potential agrochemical yield enhancers. This approach demonstrates the utility of integrating AI-driven FBDD into agricultural research.