Drug Toxicity Evaluation using Ordinal Logistic Regression with Multi-Features from ORd and To R – ORd In-silico Ventricular Cell Model

Abstract

Several drugs have been withdrawn from the market since their potential to cause Torsade de Pointes (TdP), a potentially fatal form of ventricular arrhythmia. To mitigate this risk, the Comprehensive in Vitro Proarrhythmia Assay (CiPA) proposes assessing the arrhythmogenic potential of drugs via in-silico simulations based on pharmacological data obtained in vitro. Various studies have utilized in-silico models with machine learning algorithms to classify TdP risks and yield promising results. In this study, we applied an ordinal logistic regression approach to assess TdP risk using 364 feature pairs derived from 14 features of the modified ORd and ToR-ORd models. This method allowed us to analyze drug-induced features and classify TdP risk levels. Ordinal logistic regression enabled us to explore complex relationships between these features and TdP risk levels. Notably, combining under the ToR-ORd model with under the ORd model achieved excellent performance, with Area Under the Curve (AUC values of 0.98 for high-risk and 0.92 for low-risk categories. These findings suggest that our approach can significantly enhance the understanding and assessment of TdP risk, contributing to developing safer drugs for clinical use.