Drug Discovery Computational Chemistry Platform UnitHigh-precision In Silico Screening Using the Latest Modeling Theories and Statistical Theories

High-precision in silico screening technologies

Drug design based on the three-dimensional structure of proteins (SBDD) is useful for searching for new drug seeds. however, in conventional protein-ligand docking, the flexibility of protein structure and other docking conditions were not sufficiently considered. These docking conditions should be properly optimized for successful in silico screening. The Drug Discovery Computational Chemistry Platform Unit has been constructing a highly accurate in silico screening system (Figure 1.) that makes maximum use of the three-dimensional structure of the target protein and information on known inhibitors.

in silico screening system used in Drug Discovery Computational Chemistry Platform Unit
Figure 1. in silico screening system used in Drug Discovery Computational Chemistry Platform Unit

PALLAS is a semi-automated docking condition optimization system that selects protein structure sets and other docking settings suitable for detecting diverse ligands by examining many conditions such as protein structural variations and the presence or absence of solvent molecules. By docking using the selected optimum conditions, prediction performance can be significantly improved.

For the final selection of compounds for biological assay, we employ Artificial Intelligence (AI)-based prediction named MUSES using novel protein-ligand interaction descriptor named aPLIED. The aPLIED provides protein-ligand interaction energy values atom-by-atom. The MUSES system performs machine learning to detect subtle differences in the interaction pattern between active and negative (decoy) compounds. The machine learning prediction model using aPLIED showed significantly higher performance compared with a commercially available docking score (GLIDE score) or a literature-known descriptor (PLIF).

For highly accurate activity prediction based on molecular simulation, the FMO-PBSA method, which combines the FMO method and solvent effect prediction. The FMO method enables first-principles quantum chemistry calculations for proteins. We also combined FMO-based quantum chemical interaction energy values with AI. Figure 2 shows the drug discovery technologies based on quantum chemical calculation (FMO method).

Click on the following image to enlarge

Drug discovery technologies based on quantum chemical calculation
Figure 2. Drug discovery technologies based on quantum chemical calculation (FMO method)

Molecular design from hit to lead/lead optimization

In drug discovery research, it is essential to improve not only the bioactivity for a target protein but also off-targets, pharmacokinetic (ADME), and toxicity (T) profiles. Improvement of the ADMET and off-target profile is a rate-determining stage in the early stages of drug discovery. In collaboration with AMED's next-generation drug discovery AI project (DAIIA), our platform unit develops integrated drug discovery AI platform including AI models for the above-mentioned ADMET and off-target profile based on the latest AI technologies and learning data provided from pharmaceutical companies and applied the AI models to real drug discovery targets.

Drug design scheme collaborated among RIKEN DMP platform units
Figure 3. Drug design scheme collaborated among RIKEN DMP platform units

Contact us

to TOP