Integrating multi-modal deep learning on knowledge graph for the discovery of synergistic drug combinations against infectious diseases

• Published in: Cell reports physical science Vol. 4; no. 8; p. 101520
• Main Authors: Ye, Qing, Xu, Ruolan, Li, Dan, Kang, Yu, Deng, Yafeng, Zhu, Feng, Chen, Jiming, He, Shibo, Hsieh, Chang-Yu, Hou, Tingjun
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 16.08.2023
• Subjects: drug combinations; infectious diseases; knowledge graph; virtual screening; virtual screening; knowledge graph; infectious diseases; drug combinations
• ISSN: 2666-3864; 2666-3864
• DOI: 10.1016/j.xcrp.2023.101520

The threat to global health posed by unpredictable infections and increasing antimicrobial resistance necessitates the urgent development of drug combination therapies (DCBs) for infectious diseases. Substantial efforts have been devoted to perfecting predictions for DCBs, but data scarcity and poor model interpretability continue to present significant barriers to the development of novel DCBs. To address these issues, here we propose a framework for predicting DCBs by combining knowledge graph representation learning and the technique of community discovery for complex networks. Within this framework, we demonstrate that multi-modal information and multiple types of DCBs could significantly facilitate the predictive performance and improve hit rates in realistic virtual screening scenarios. The high hit rate of 85% for experimental validation strongly supports the proposal that our approach could effectively harness useful information hidden in highly complex biological networks and accelerate in silico discovery of pairwise DCBs for infectious diseases and beyond. [Display omitted] •A computational framework is proposed for accelerating drug combination discovery•Multi-modal information enhances the prediction of synergistic drug combinations•Community discovery techniques could improve hit rates in virtual screening•The proposed rigorous protocol enables analysis of underlying mechanisms Ye et al. propose a framework that integrates knowledge graph representation learning and the technique of community discovery to predict drug combination therapies for infectious diseases. By incorporating multi-modal information, the framework enhances predictive performance and hit rates in virtual screening scenarios.