Collection and preparation of molecular databases for virtual screening

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Traditional drug discovery is linear, often failing due to non-drug-like molecules identified late in development.
• In vitro and in vivo assays show poor correlation, necessitating improved early-stage filtering.

Purpose of the Study:

• To develop and validate a novel binary Quantitative Structure-Activity Relationship (QSAR) model for assessing drug-likeness.
• To create a prioritized, drug-like molecular database for enhanced virtual screening in early drug discovery.

Main Methods:

• A binary QSAR model was derived and used to filter molecules from 36 catalog suppliers.
• The model's performance was compared against existing filters, including those used by the ZINC database.
• The validated model was applied to a large dataset (4,972,123 molecules) to generate a refined subset.

Main Results:

• The derived QSAR model demonstrated superior performance compared to commonly used filters.
• A significantly refined database of 2,920,551 drug-like molecules was generated.
• The new database effectively prioritizes compounds likely to succeed in later drug development stages.

Conclusions:

• The developed QSAR model offers a more effective method for in silico prioritization of drug-like compounds.
• The resulting prioritized database holds significant potential for improving virtual screening efficiency.
• This approach facilitates the discovery of molecules with a higher probability of success in drug development.