Genome-wide discovery of hidden genes mediating known drug-disease association using KDDANet

Abstract

Inferring novel therapeutic indications of known drugs provides an effective method for fast-speed and low-risk drug development and disease treatment. Various computational tools have been developed to accurately predict potential associations between drugs and diseases. Nevertheless, no method has been designed to unveil pharmacological (toxicological) gene interactions underlying Known Drug-Disease Associations (KDDAs). System-level interpretation and elucidation of molecular mechanism underlying KDDAs remains a main challenge. Here, for the first time, we presented a novel and general computational framework, called KDDANet, for systematic uncovering hidden gene interactions underlying KDDAs from the perspective of complex molecular network. KDDANet effectively implemented minimum cost optimization and graph clustering on a unified flow network model. The excellent performance and general applicability of KDDANet on uncovering hidden genes underlying KDDAs were globally demonstrated by enrichment analysis of known and novel KDDA genes against two well-curated databases across broad types of diseases. Case studies on several types of diseases further highlighted that the potential value of KDDANet on revealing hidden gene interactions underlying 2 KDDAs. Particularly, it's worth noting that KDDANet can reveal the shared gene interactions underlying multiple KDDAs. For facilitating biomedical researchers to explore KDDA molecular mechanisms and guiding drug repurposing, we provided an online web server, http://47.94.193.106/kdda/index, to browse, download and analyze the predicted KDDA gene interactions. Our software and usage instruction were freely available at https://github.com/huayu1111/KDDANet/.