Prediction of Drug-Induced TdP Risks Using Machine Learning and Rabbit Ventricular Wedge Assay

TL;DR

This study develops an interpretable machine learning model using rabbit ventricular wedge assay data to accurately predict drug-induced torsades de pointes risks, aiding safer drug development.

Contribution

Introduces a multinomial logistic regression approach for three-class TdP risk prediction based on experimental data, with robust validation and interpretability features.

Findings

Model achieves accurate three-class TdP risk classification.

Bootstrap confidence intervals demonstrate prediction reliability.

Permutation importance highlights key predictors.

Abstract

Torsades de pointes (TdP) is an irregular heart rhythm as a side effect of drugs and may cause sudden cardiac death. A machine learning model that can accurately identify drug TdP risk is necessary. This study uses multinomial logistic regression models to predict three-class drug TdP risks based on datasets generated from rabbit ventricular wedge assay experiments. The training-test split and five-fold cross-validation provide unbiased measurements for prediction accuracy. We utilize bootstrap to construct a 95% confidence interval for prediction accuracy. The model interpretation is further demonstrated by permutation predictor importance. Our study offers an interpretable modeling method suitable for drug TdP risk prediction. Our method can be easily generalized to broader applications of drug side effect assessment.