Small Molecular Drug Screening Based on Clinical Therapeutic Effect

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Virtual screening accelerates drug discovery by reducing experimental time and costs.
• Drug multi-classification can enhance virtual screening efficiency by predicting drug classes.
• Accurate classification of drug molecules is crucial for identifying potential therapeutic agents.

Purpose of the Study:

• To develop and validate a drug multi-classification model for efficient virtual screening.
• To group drug molecules based on therapeutic effects and mechanisms of action.
• To predict potential attributes of unknown drug compounds and identify new uses for existing drugs.

Main Methods:

• Collected 1019 drug molecules from various databases.
• Quantified molecular structures using molecular descriptors and fingerprints derived from SMILES.
• Employed the Kennard-Stone method for data set division.
• Compared five classification algorithms and a fusion method to identify the best performing model.
• Utilized the best model for validation set prediction.

Main Results:

• Achieved a prediction accuracy of 0.862 and a kappa coefficient of 0.808 on the test set.
• The highest classification accuracy on the validation set reached 0.873.
• Identified a reliable molecular set for predicting drug attributes.

Conclusions:

• The developed drug multi-classification method significantly improves virtual screening efficiency.
• The model can accurately predict drug classes, aiding in drug discovery and repurposing.
• This research provides a valuable reference for simultaneous multi-class drug virtual screening.