Compilation and physicochemical classification analysis of a diverse hERG inhibition database

Area of Science:

• Computational Chemistry
• Drug Discovery
• Cardiovascular Pharmacology

Background:

• The hERG channel is crucial for cardiac repolarization.
• hERG inhibition by drug compounds can lead to potentially fatal cardiac arrhythmias (e.g., Torsades de Pointes).
• Predicting hERG inhibition early in drug discovery is vital for mitigating cardiotoxicity risks.

Purpose of the Study:

• To construct a large, diverse dataset of hERG inhibition data.
• To develop a predictive model for hERG inhibition using machine learning.
• To identify key physicochemical descriptors influencing hERG channel interactions.

Main Methods:

• Compiled a dataset of 6690 compounds from ChEMBL and literature.
• Assessed hERG activity using patch-clamp and competitive displacement assays.
• Applied gradient boosting machine classification using physicochemical and topological descriptors (log P, pKa, PSA, etc.).

Main Results:

• Achieved 75-80% accuracy in classifying compounds for hERG inhibition.
• Identified descriptor-response profiles consistent with known hERG binding.
• Weakly basic groups (pKa < 6) and ionization state significantly impact hERG inhibition.
• Lipophilicity (log P) influence varies by ionization state (bases > zwitterions > neutrals > acids).

Conclusions:

• The developed model accurately predicts hERG inhibition potential.
• The model provides insights into structure-activity relationships for hERG ligands.
• This predictive tool can guide drug discovery towards safer compounds with reduced cardiotoxicity risk.