Statistical learning in preclinical drug proarrhythmic assessment.

Area of Science:

• Pharmacology and Toxicology
• Computational Biology
• Biostatistics

Background:

• Torsades de pointes (TdP) is a life-threatening arrhythmia linked to drug toxicity.
• Preclinical studies investigate drug-induced TdP, but robust predictive frameworks are needed.
• Accurate risk prediction is essential for drug safety assessment.

Purpose of the Study:

• To develop and validate a statistical learning framework for predicting drug-induced TdP risk.
• To classify drugs into low-, intermediate-, and high-risk categories using preclinical data.
• To provide interpretable and accurate tools for drug safety evaluation.

Main Methods:

• Ordinal logistic regression and ordinal random forest models were employed.
• Datasets from two experimental protocols were utilized for model training and testing.
• Leave-one-drug-out cross-validation, stratified bootstrap, and permutation importance assessed model performance and interpretability.

Main Results:

• The proposed models accurately predicted drug-induced TdP risk categories.
• Identified outlier drugs aligned with existing literature, demonstrating model validity.
• The methods provided accurate and interpretable risk assessments.

Conclusions:

• Ordinal logistic regression and random forest models offer a reliable approach for predicting drug-induced TdP risk.
• These models can serve as valuable supplemental evidence in drug safety assessments.
• The developed framework enhances the understanding and prediction of cardiac arrhythmia risks associated with drug candidates.