Presentation Information
[1Yin-A-55]Implementation of Molecular Complex 3D Structure Prediction Models for Accelerating Drug Discovery Research
〇Shotaro Maedera Maedera1, Shogo Suga1 (1. Daiichi Sankyo Company, Limited)
Keywords:
AI,Biology,Structure Biology,Pharmacology
Recent molecular 3D structure prediction models, including AlphaFold-3, have achieved high-precision prediction of complex structures formed by biomolecules such as proteins, nucleic acids, and small molecules. They accelerate the drug discovery process by expediting intermolecular interaction analyses. However, no system existed to perform complex structure prediction using these models in a secure environment in Daiichi Sanyo. Therefore, to establish an environment where researchers can smoothly utilize commercially available AlphaFold-3 reproduction models, we implemented the models and developed a GUI application. We deployed Boltz-2, Protenix, and Chai-1 on Tokyo-1, GPU computing environment. Evaluation using publicly available data demonstrated favorable results when comparing Boltz-2 with conventional methods. Furthermore, focusing particularly on nucleic acid-small molecule complexes, we implemented an application executing Boltz-2 structure predictions via GUI and visualizes the results on Google Cloud Platform. Through these activities, complex structure prediction models have been introduced and made readily accessible to researchers within our organization.