PASS Targets: Ligand-based multi-target computational system based on a public data and naïve Bayes approach

Area of Science:

• Computational chemistry
• Pharmacology
• Drug discovery

Background:

• Accurate estimation of drug-target interactions is crucial for drug discovery and toxicity assessment.
• The ChEMBL database version 19 provides a rich source of data for training predictive models.

Purpose of the Study:

• To develop a computational tool for predicting interactions between drug-like compounds and protein targets.
• To leverage structure-activity relationships (SAR) for enhanced prediction accuracy.

Main Methods:

• Utilized a Bayesian-like method implemented in PASS software.
• Trained the model on 589,107 chemical compounds from the ChEMBL database.
• Employed leave-one-out and 20-fold cross-validation for accuracy estimation.

Main Results:

• The developed tool, PASS Targets, can predict interactions with 2507 protein targets.
• Achieved high prediction accuracy of approximately 96% (AUC ROC) during cross-validation.
• Demonstrated an average AUC ROC of 90% on an external test set of 700 drugs and 206 targets.

Conclusions:

• PASS Targets is a validated computational tool for predicting drug-target interactions.
• The tool shows significant potential for accelerating drug discovery and toxicity assessment processes.
• The high accuracy suggests the reliability of SAR analysis for predicting compound-target relationships.