Novel Two-Step Classifier for Torsades de Pointes Risk Stratification from Direct Features.

Area of Science:

• Pharmacology
• Computational Biology
• Cardiology

Background:

• Pre-clinical Torsades de Pointes (TdP) risk assessment initially focused on hERG channel block.
• Improved TdP risk prediction requires considering non-hERG ion channel interactions.
• Current multi-channel classifiers use direct or derived features, with derived features showing inconsistent accuracy gains.

Purpose of the Study:

• To develop a novel two-step TdP risk classification method, MCB@EAD.
• To evaluate the utility of direct versus derived features in multi-channel TdP risk assessment.
• To investigate the role of non-hERG channels in TdP risk at critical hERG block concentrations.

Main Methods:

• Proposed a two-step classification: compounds with insufficient hERG block were deemed non-torsadogenic.
• Second step: assessed non-hERG channel influence on TdP risk using direct or derived features at EAD-inducing hERG block levels.
• Utilized computational cardiac cell models to simulate drug-induced channel block and EADs.

Main Results:

• MCB@EAD demonstrated comparable or superior TdP risk classification accuracy using direct features compared to published methods.
• TdP risk strongly correlated with the propensity to generate Early After Depolarizations (EADs) in computational models.
• Derived features from biophysical models did not enhance predictive capability for TdP risk assessment.

Conclusions:

• The MCB@EAD method offers an effective approach for TdP risk classification.
• Direct features are valuable for TdP risk prediction, particularly when considering EAD generation.
• Complex derived features from biophysical models do not consistently improve TdP risk prediction accuracy over simpler direct features.