Drug Discovery Database Development

Background & Challenge

The RIKEN Center for Computational Science has been researching medium-molecule drugs that target protein-protein interactions (PPI), and has built a library of new compounds that inhibit PPI. To use this library for the design and search of new PPI-inhibiting compounds, we developed and operate the structured database "DLiP." Each compound in DLiP is labeled "active" or "inactive" for PPI based on experimental activity data collected from multiple public databases and academic literature, including ChEMBL and the Journal of Medicinal Chemistry. We are also developing "MIIDB-AI," a medium-molecule PPI database covering peptides and nucleic acids that DLiP does not cover, aiming for high-precision interaction prediction between targets and ligands through the application of AI technology.

Note: For inquiries about DLiP and MIIDB-AI themselves, please contact the operating organization, R-CCS Medicinal Chemistry Applied AI Unit.

Udzuki's Role

Implemented chemical structure search

Implemented compound search functionality based on chemical structure, including substructure and similarity search using SMILES/SMARTS

Implemented molecular descriptor calculation

Implemented calculation of molecular descriptors such as molecular weight and LogP using RDKit and other tools

Implemented drug-likeness search

Implemented search functionality using drug-likeness indicators used in drug discovery, such as Rule of Five, QED, and Fraction Csp3

Integrated multiple data sources

Built a unified, searchable database by integrating multiple sources in different formats, including ChEMBL, PDB, and academic literature

Contributed to a published paper

Designed and implemented the system to a standard suitable for publication as part of a paper in a peer-reviewed international journal