深度学习在药物-药物相互作用预测中的研究进展OA

The risk of drug-drug interaction(DDI)associated with polypharmacy has become increasingly prominent,while traditional experimental approaches,though reliable,are limited by their high cost and lengthy procedures.The pres-ent study focuses on drug feature representation and categorises existing deep learning methods into five types:sequence-based,structured,graph-based,knowledge graph-based,and multimodal fusion approaches.A systematic analysis of their application characteristics,strengths,and limitations is conducted.The experimental findings,derived from the analysis of two benchmark datasets,demonstrate that graph-based and knowledge-graph methods achieve superior performance in complex prediction tasks.Moreover,the merits and constraints of DDI prediction models are analysed across three domains:namely,drug development,clinical application,and decision support.In conclusion,the primary challenges pertaining to data quality,model interpretability and clinical validation are outlined,and the subsequent research direc-tions are discussed.