Determination of torsade-causing potential of drug candidates using one-class classification and ensemble modelling approaches.

Adverse drug reactions (ADRs) are a main problem faced by drug companies and regulatory authorities. Not only do they contribute heavily to late-phase failure of drug development and withdrawal of drugs from the market, they also pose significant health risks to patients. Rare and severe ADRs are even harder to detect, and sufficient attention has not been paid to them. Torsade de pointes (TdP), an atypical ventricular tachycardia which is potentially life-threatening, is one of them. The objective of this project is to develop a computational model to predict TdP-causing potential of drug candidates. A total of 260 marketed drugs were collected and screened for their potential to cause TdP. 103 drugs were classified as TdP+ and 157 were likely to be TdP-. One-class classification methods were used to construct multiple base models. A model dependent applicability domain estimation method was used to determine the applicability of the base models for future dataset. A final ensemble model was constructed based on selected base models and it had sensitivity and specificity value of 78.4% and 90% respectively when estimated using external cross validation method. The result suggests that the ensemble model developed in this study is potentially useful for facilitating the prediction of TdP in drug candidates. The ensemble model is made available via the free software, PaDEL-DDPredictor.