Toxicity detection of small drug molecules of the mitochondrial membrane potential signalling pathway using bagging-based ensemble learning
Vishan Kumar Gupta
Abstract
This study is focused on the in-silico method QSAR for the detection of chemical and drug-induced toxicities of small drug molecules of Mitochondrial Membrane Potential (MMP). This prediction is based on the various physicochemical properties of MMP and its corresponding target class to reduce the animal testing, time, and cost associated with risk assessment. Here, is a total of 8070 drug molecules of MMP out of which 1260 drug molecules are toxic and the remaining 6810 are non-toxic. Pa-DEL descriptor software is used to extract features of MMP signalling pathway. Initially, the class imbalance issue is fixed then feature selection is performed using a random forest importance algorithm. A bagging-based ensemble model is proposed for toxicity prediction based on the voting of five base classifiers, and it is found that our proposed ensemble method achieved 97.62% accuracy. Finally, K-fold cross-validation is applied to check the consistency of the proposed model.