A simple method to improve the odds in finding 'lead-like' compounds from chemical libraries

Area of Science:

• Computational Chemistry
• Drug Discovery
• Medicinal Chemistry

Background:

• Virtual screening is crucial for identifying novel drug candidates.
• Existing methods often require complex calculations or extensive datasets.
• There is a need for simple yet effective virtual screening approaches.

Purpose of the Study:

• To propose a straightforward virtual screening method utilizing chemical characters from 2D chemical structures.
• To introduce and define a new concept, 'trait', as a global chemical character represented by 2D molecular descriptors.
• To evaluate the efficacy of molecular fingerprints and traits in predicting drug-likeness for specific pharmacological activities.

Main Methods:

• Calculating local (molecular fingerprint) and global (trait) chemical characters from 2D chemical structures.
• Training the model on a database of clinically used Japanese drugs to learn drug-specific fingerprints and traits.
• Applying the trained model to predict drug-likeness in external chemical databases containing compounds with relevant pharmacological activity.
• Assessing prediction performance using the enrichment factor.

Main Results:

• The proposed method achieved practical results despite its simplicity.
• An enrichment factor of 66 was obtained for beta-adrenergic blockers.
• The method successfully covered approximately 57% of active molecules in the tested chemical databases.

Conclusions:

• The combination of molecular fingerprints and the novel 'trait' concept offers a simple and effective approach for virtual screening.
• This method demonstrates significant potential for identifying drug-likeness and accelerating drug discovery.
• The approach is particularly promising for targeting specific pharmacological activities with high efficiency.