Importance of Drug Features in Drug–Drug Interaction: A Comparative Study

A drug–drug interaction is an interplay of drugs where one or more drugs interfere with the activity of other drugs. Interaction between drugs may cause side effects that are unforeseen. They may also intervene and oppose each other’s action and nullify the medication. Many of such interactions are often negligible but some might be harmful if not discovered at the earliest. Features or characteristics of drugs do play a vital role in interaction. Our study aims to predict important features that are involved in the interaction of two drugs. Firstly, train multiple learning models on the dataset. These models are Random Forest Classifier, Extreme Gradient Boosting, Support Vector Machines, Autoencoders + Extreme Gradient Boosting, Convolutional Neural Network, and Capsule Networks. Then, compare the results of trained models and select the most appropriate performance yielding model. Then, ask the model to predict the interaction between two drugs and also plot the feature importance distribution of two interacting drugs. In the study, it was found that Random Forest Classifier did slightly better than Extreme Gradient Boosting (XGBoost) and outshined against all other models. Hence, Random Forest Classifier was used as The BackBone model to predict feature importance among two interacting drugs.