R-CCS Cafe - Special Edition (Jul 19, 2024)

Title:
Accelerating Middle Molecule Drug Discovery with Evolving Next-Generation AI Platform and Databases
Abstract:
Despite the remarkable advances in AI technologies in recent years, significant challenges
remain in their application to next-generation drug discovery. These challenges include a
qualitative and quantitative lack of data in drug discovery research related to new modalities,
difficulties in domain adaptation of AI models, and the need for efficient use of computational
resources. We are developing an AI drug discovery platform that utilizes the powerful
computational resources of the supercomputer “Fugaku” to improve the efficiency of
medicinal chemistry processes and contribute to the discovery of innovative drugs.

The main goal of my research is to develop AI models for targets that are difficult to treat with
conventional small-molecule drugs, such as protein-protein interactions (PPIs), to support
the acquisition of lead molecules that will become novel drug candidates. In this seminar, I
will present my research results on the latest AI drug discovery technologies and their
validation.

I have constructed a new database of small-to-medium-sized compounds with high drug
properties and novel chemical structural information. This has enabled the development of
predictive models using deep learning and iterative screening that effectively utilize small
amounts of experimental data. Experimental validation through viral infection inhibition has
confirmed significant improvements in hit rates and led to the discovery of new inhibitors.

Additionally, we focus on developing high-quality compound design methods using medicinal
chemistry AI that leverages pharmaceutical patent data. This approach can generate novel
patented molecules in the known pharmaceutical chemical space. We intend to accelerate the
generation of research results with HPC and discuss the potential contribution of AI to
middle-molecular drug discovery.