Predicting Drug-Drug Interactions with Machine-Learning and ATC-SMILES Combined Representation

Abstract

Polypharmacy is a potential strategy for managing such intricate disorders, encompassing conditions like cancer, diabetes, and age-related issues in older individuals. Nonetheless, when a medication is combined with one or more drugs that either enhance, diminish, or counteract its intended effects, it can lead to undesired adverse reactions. In severe cases, these interactions can cause serious morbidity and increased mortality rates globally. In this study, we collected a Drug-Drug interaction dataset from the DrugBank database. Various chemical features were then extracted from the Simplified Molecular-Input Line-Entry System of interacting drug pairs. Our emphasis was on representing the Molecular Access System fingerprints of these drug pairs. Molecular Access System fingerprints signify the presence or absence of specific substructures in the molecule and were generated using the RDKit Open-Source Cheminformatics Software. Furthermore, we incorporated Anatomical Therapeutic Chemical classifications into our analysis. Finally, we employed various machine learning algorithms, using K-Nearest Neighbor, Random Forest, Logistic Regression, Support Vector Machine, and XGBoost for learning the extracted features and predict large-scale Drug-Drug interactions among various drug pairs. Among these models, the XGBoost model exhibited superior performance across most measurement metrics.