Over the past decades, combination therapy has become a promising strategy to combat antimicrobial resistance. Compared with monotherapy, drug combinations can enhance efficacy and delay resistance. However, experimental identification of drug–drug interactions (DDIs) remains time-consuming and limits large-scale screening.
Recently, a review article titled “A comprehensive review of cluster methods for drug–drug interaction network” was published in Quantitative Biology. The study provides a systematic overview of clustering-based approaches for DDI network analysis, highlighting their potential in drug combination discovery and mechanism exploration.
Unlike previous reviews that mainly focus on supervised learning, this work emphasizes the role of unsupervised learning (clustering), which can uncover hidden patterns in large-scale datasets without requiring labeled data. This makes clustering particularly suitable for exploring complex DDI networks.
From a methodological perspective, clustering can be applied from two complementary directions. On the one hand, drugs can be grouped based on their intrinsic features (e.g., chemical structure and mechanisms of action), and interactions between clusters can then be analyzed to infer unknown mechanisms. On the other hand, clustering can be performed directly on DDI networks, allowing drug properties to be inferred from their interaction patterns.
Despite these advances, several challenges remain. Current studies on antibiotic–adjuvant interaction networks are still limited, integration of multi-modal data with missing information remains difficult, determining the optimal number of clusters is non-trivial, and higher-order drug combinations are insufficiently explored.
Future research may address these challenges through semi-supervised clustering, multi-modal learning, reinforcement learning-based model selection, and hypergraph-based modeling of complex drug interactions.
DOI
10.1002/qub2.70015