Accelerating Pharmaceutical Innovation Through AI: A Systematic Review of Machine Learning and Deep Learning Technologies for Drug-Drug Interaction Prediction

Abstract

Drug–Drug Interactions (DDIs) pose a significant challenge to patient safety and therapeutic efficacy. Traditional experimental methods for detecting DDIs are labor-intensive, costly, and cannot scale to the vast number of potential drug combinations. To address these challenges, Artificial Intelligence (AI) approaches—particularly Machine Learning (ML) and Deep Learning (DL)—have emerged as effective computational tools for predicting DDIs and supporting clinical decision-making. This study systematically reviews the application of ML and DL methods to DDI prediction based on the available English-language literature, evaluating their contributions, methodological strengths, limitations, and future opportunities. Following PRISMA guidelines, a Systematic Literature Review (SLR) of 75 primary studies was conducted, analyzing commonly used datasets, model architectures, algorithms, and evaluation metrics, with an emphasis on clinical relevance and potential to prevent adverse drug events. Similarity-based, network-based, and deep learning approaches dominate the field. While DL models often achieve higher predictive accuracy, they face challenges such as overfitting and limited interpretability, reducing their clinical trustworthiness and adoption in real-world settings. AI-driven methods have significantly advanced DDI prediction and demonstrate strong potential to enhance drug safety monitoring and clinical decision support. However, critical gaps remain, particularly in applying advanced AI paradigms such as transfer learning and multi-agent systems. Future research should prioritize larger, more balanced datasets and improved interpretability to ensure reliable, clinically actionable outcomes and foster wider adoption in healthcare practice.

Ähnliche Arbeiten