A comprehensive support vector machine binary hERG classification model based on extensive but biased end point hERG...

Area of Science:

• Computational Chemistry
• Pharmacology
• Drug Discovery

Background:

• The human ether-a-go-go related gene (hERG) potassium channel is a critical target for preclinical drug discovery toxicity screening due to its role in cardiotoxicity.
• In silico models are essential for efficiently screening large compound libraries for potential hERG blockers.

Purpose of the Study:

• To construct and validate a robust in silico screening model for predicting hERG channel blockers.
• To identify key molecular descriptors that contribute to hERG channel blocking activity.
• To refine existing models and improve the accuracy of cardiotoxicity prediction in drug discovery.

Main Methods:

• Utilized the PubChem hERG Bioassay data set (1668 compounds) to build an in silico screening model.
• Employed a combination of 4D fingerprints (4D-FP) and 2D/3D VolSurf-like molecular descriptors.
• Developed a binary classification model using a Support Vector Machine (SVM) and validated it with internal and external datasets.

Main Results:

• The optimal SVM model, incorporating Lipinski's rule-of-five and molecular lipophilicity, achieved 95% accuracy, 90% sensitivity, and 96% specificity in 10-fold cross-validation.
• The external test set (356 compounds) showed an overall prediction accuracy of approximately 87%.
• Identified significant structural features contributing to hERG blocking effects, aiding in 3D structure-activity relationship interpretation.

Conclusions:

• The developed in silico models are robust and accurate in predicting a compound's predisposition to block hERG ion channels.
• These models can significantly aid in preclinical drug discovery by identifying potential cardiotoxic compounds early in the screening process.
• The study provides insights into structural features associated with hERG channel blockade, facilitating rational drug design.