The hERG potassium channel is a cardiac antitarget: drug-induced blockage can disrupt electrical repolarization, contribute to cardiotoxicity, and cause long QT syndrome or arrhythmia. Early screening therefore removes compounds with this liability while preserving uncertainty for experimental follow-up. section 3.1
A ligand-based screening workflow treats hERG blockage as a supervised classification problem:
- Predict class 1 for a hERG blocker and class 0 for a non-blocker. section 3.1
- Define blockers from experimental potency using IC50 < 40 μM, equivalent to pIC50 > 4.4. section 3.2.1
- Exclude pIC50 from features because it defines the label, requires experiments, and may contain missing or inconsistent measurements. section 3.2.1
- Compare machine-learning predictions with structural-alert filters and distinguish missed blockers from incorrectly rejected therapeutic compounds. section 3.1
Dataset and label
The hERG blockers dataset contains 587 records with SMILES, names, pIC50 values, class labels, and scaffold or random split assignments; 526 records have non-null pIC50 values. The class column supplies the training target, while split columns support alternative validation strategies. section 3.2.1
