Ensemble of structure and ligand-based classification models for hERG liability profiling.

Area of Science:

• Pharmacology
• Computational Chemistry
• Drug Discovery

Background:

• Drug-induced cardiotoxicity is a major safety concern in early drug development.
• Blockade of the human ether-à-go-go-related potassium channel (hERG) frequently causes cardiotoxicity and fatal arrhythmias.
• Accurate prediction of hERG liability is crucial for safe drug development.

Purpose of the Study:

• To compare the performance of ligand-based (LB) and structure-based (SB) computational models for predicting hERG-related cardiotoxicity.
• To evaluate the effectiveness of combining LB and SB features for improved prediction accuracy.
• To develop reliable tools for early-stage identification of potential cardiotoxic drug candidates.

Main Methods:

• Development of LB and SB classifiers using the Random Forest algorithm on a training set of 12,789 hERG binders.
• SB models utilized docking and rescoring calculations.
• LB models were based on physicochemical descriptors and fingerprints.
• Internal validation via ten-fold cross-validation and external validation on a separate test set.

Main Results:

• The ligand-based model performed best on the training set.
• The combined LB and SB model demonstrated superior performance on the external test set, outperforming individual LB and SB models.
• Both LB and SB approaches, particularly when integrated, showed satisfactory predictive performance.

Conclusions:

• Combining ligand-based and structure-based features significantly enhances the prediction of hERG blockers for novel chemical scaffolds.
• The developed predictive models offer valuable tools for early-stage screening of drug candidates to mitigate cardiotoxicity risks.
• Integration of diverse computational strategies is key to overcoming limitations in predicting drug-induced cardiotoxicity.