Research work · 2026

Machine Learning in Drug–Drug Interactions

Rashi Jain, Rupal Kushwah, Dhairiya Agarwal, Prabha Garg

Machine Learning and Deep Learning in Drug Design

Abstract

Drug–drug interactions (DDIs) pose significant challenges to patient safety and effective medication administration. Machine learning (ML) has become a powerful approach for forecasting DDIs, leveraging data-driven approaches to mitigate associated risks. ML methodologies for DDI prediction include shallow learning techniques such as support vector machines (SVMs), decision trees (DTs), random forests (RF), and k-nearest neighbours (KNNs). Advanced deep learning (DL) models, such as graph neural networks (GNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), autoencoders, and transformers, demonstrate flexibility in capturing complex drug relationships. Additionally, knowledge graph (KG)-based frameworks, including KGNN, MUFFIN, and SumGNN, are increasingly utilized for their ability to integrate structured biomedical knowledge. Biochemical databases play a critical role in DDI prediction, providing data for model training and validation. Despite significant progress, challenges in model interpretability and inconsistent performance metrics remain barriers to clinical adoption. Addressing these limitations through improved data integration, innovative modelling approaches, and enhanced interpretability frameworks is essential. Future efforts focusing on integrating multi-source data and real-world evidence have the potential to revolutionize DDI prediction, enabling safer drug therapies and facilitating advancements in drug discovery and repurposing.

DOI: 10.1039/9781837070206-00359