Research Progress in Network Medicine and Artificial Intelligence-Based Prediction of Drug Combination Synergy
TL;DR
Current studies remain constrained by data sparsity, negative-sample bias, insufficient model interpretability and limited prospective validation, so future work should strengthen standardized data integration, mechanism-constrained modeling and joint dose-timing optimization to improve interpretability, generalizability and clinical usability.
Abstract
Drug combination therapy can improve the treatment of complex diseases through multi-target intervention, but the number of candidate combinations expands rapidly with the size of the drug space, making purely high-throughput screening insufficient for research and clinical needs. Network medicine provides biological priors for assessing the rationale of combinations by integrating protein-protein interaction networks, disease modules and drug-target topology, whereas artificial intelligence can learn nonlinear features from heterogeneous data such as chemical structures, targets, omics profiles and cellular phenotypes. This review summarizes recent progress in drug combination synergy prediction based on network medicine and artificial intelligence. It compares network topology methods, conventional machine learning, graph neural networks and multimodal fusion models in terms of principles, applicable scenarios and limitations, and discusses their translational value in cancer, hypertension and other complex diseases. Current studies remain constrained by data sparsity, negative-sample bias, insufficient model interpretability and limited prospective validation. Future work should strengthen standardized data integration, mechanism-constrained modeling and joint dose-timing optimization to improve interpretability, generalizability and clinical usability.