Drug-induced torsadogenicity prediction model: An explainable machine learning-driven quantitative structure-toxicity...

Area of Science:

• Computational chemistry
• Pharmacology
• Toxicology

Background:

• Drug-induced Torsade de Pointes (TdP) is a dangerous heart rhythm caused by drug cardiotoxicity.
• Accurate prediction of TdP risk is crucial for drug safety but remains challenging.
• Current methods often require extensive testing, including animal models.

Purpose of the Study:

• To develop a novel quantitative structure-toxicity relationship (QSTR) model for predicting torsadogenic cardiotoxicity.
• To estimate the risk of TdP based on the molecular structure of pharmaceutical compounds.
• To create an in silico tool for early drug safety screening, reducing reliance on animal testing.

Main Methods:

• Developed a QSTR prediction model using an in silico approach based on the 4R rule.
• Employed machine learning algorithms: Support Vector Machine (SVM), Random Forest (RF), XGBoost, and CatBoost.
• Implemented a two-step feature selection process and utilized SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• The developed QSTR model successfully predicted the torsadogenic risks of various pharmaceutical compounds.
• The model demonstrated enhanced predictive accuracy through rigorous feature selection.
• SHAP analysis provided insights into the prediction of torsadogenic risk, particularly for the RF model.

Conclusions:

• The study presents a robust QSTR model for early screening of torsadogenic potential in drug candidates.
• This in silico approach offers a time- and cost-effective alternative to traditional animal testing.
• The model aims to improve drug safety evaluation and mitigate the risks of drug-induced TdP.