Deep Neural Networks Predict Inhibitors of Schistosoma Mansoni Thioredoxin Glutathione Reductase (SmTGR)

Over 200 million people worldwide are affected annually by schistosomiasis with debilitating socio-economic effects. Praziquantel remains the main chemotherapy against this neglected tropical disease but there are reports of drug resistance. Therefore, necessitating the need to identify potential biotherapeutic molecules. The study describes the first deep learning (DL)-based computational models for predicting inhibitors of Schistosoma mansoni Thioredoxin glutathione reductase (SmTGR), which is an essential protein for the survival of the helminths in the host. The state-of-the-art performance of DL in similar applications makes it ideal to deploy on bioactive datasets of the SmTGR drug target. Cost-sensitive deep neural network classifiers were trained using the binary classification approach. Based on the area under curve (AUC) of the receiver operating characteristic (ROC) Curve scores (86.3-86.5%), the five best models generated were able to classify inhibitors with high accuracy (85-90%). Additionally, the DL classifier outperformed random forest by far. This is a proof of concept that deep neural networks can efficiently and robustly classify active schistosomal molecules from inactive. The generated models could be used to screen large-scale compound libraries to prioritize potential inhibitors for experimental characterization. The study developed cost-sensitive deep neural network (DNN) classifiers for predicting anti-schistosomal molecules. This is a plausible proof of concept since the DNNs outperformed other models including random forest. The DNNs can be deployed to screen large-scale compound libraries to identify potential biotherapeutic entities against thioredoxin glutathione reductase.