The Study on the hERG Blocker Prediction Using Chemical Fingerprint Analysis

Area of Science:

• Pharmacology
• Computational Chemistry
• Drug Discovery

Background:

• Blockage of the human ether-a-go-go-related gene (hERG) potassium channel by small molecules can lead to severe cardiac side effects.
• Early screening for hERG channel activity is critical in the drug discovery pipeline to mitigate potential cardiotoxicity.

Purpose of the Study:

• To develop and evaluate machine learning (ML) and deep learning (DL) models for predicting hERG channel blockers.
• To assess the impact of different molecular fingerprints on the performance of predictive models.

Main Methods:

• Collected a dataset of 5299 known hERG inhibitors with diverse chemical structures.
• Developed quantitative structure-activity relationship (QSAR) models using ML and DL algorithms with various integer and binary fingerprints.
• Validated model performance using training, test, and two external datasets.

Main Results:

• Models utilizing integer-type fingerprints demonstrated superior performance compared to those using no, converted binary, or original binary fingerprints.
• Integer-type fingerprints were found to be more suitable for ML algorithms, while binary fingerprints performed better with DL algorithms.
• The rational selection of molecular fingerprints significantly impacts the accuracy of hERG blocker prediction.

Conclusions:

• The study highlights the importance of selecting appropriate molecular fingerprints for building accurate predictive models of hERG channel blockers.
• Findings suggest that integer-type fingerprints combined with ML offer a robust approach for hERG activity prediction in early drug discovery.