Novel Bayesian classification models for predicting compounds blocking hERG potassium channels.

Area of Science:

• Computational chemistry
• Drug discovery
• Cardiovascular pharmacology

Background:

• Drug-induced long QT syndrome is often caused by blockage of the hERG potassium channel.
• Accurate prediction of hERG channel blockers is crucial for drug safety.

Purpose of the Study:

• To construct novel computational models for predicting hERG channel blockers.
• To improve the accuracy and efficiency of identifying potential cardiotoxic compounds.

Main Methods:

• Utilized Doddareddy's hERG blockage dataset (2644 compounds) divided into training and test sets.
• Constructed Laplacian-corrected Bayesian classification models using Discovery Studio.
• Validated models internally, on a test set, and externally using a separate dataset of 60 compounds.

Main Results:

• A Bayesian model incorporating molecular properties (Mw, PPSA, ALogP, pKa_basic) and ECFP_14 fingerprints achieved 91% global accuracy on the test set.
• The model demonstrated 90% sensitivity and 92% specificity on the test set.
• External validation showed 58% global accuracy, 61% sensitivity, and 57% specificity.

Conclusions:

• The developed computational model outperforms existing literature models for predicting hERG channel blockers.
• The novel model is suitable for large-scale prediction of compounds that may block hERG channels.
• This tool can aid in early-stage drug safety assessment and reduce the risk of drug-induced cardiotoxicity.