Prediction of Microbe–Disease Associations by Graph Regularized Non-Negative Matrix Factorization

Journal of computational biology: A journal of computational molecular cell biology

Shu-LinWang^2,1¹  YueLiu^1² 

[1] Address correspondence to: Prof. Shu-Lin Wang, College of Computer Science and Electronic Engineering, Hunan University, Changsha, Hunan 410082, China^2;College of Computer Science and Electronic Engineering, Hunan University, Changsha, Hunan 410082, China^1

关键词: microbe–disease similarity;    non-negative matrix factorization;    Gaussian kernel similarity.;   

DOI  :  10.1089/cmb.2018.0072

学科分类：生物科学（综合）

来源: Mary Ann Liebert, Inc. Publishers

【 摘 要 】

More and more evidence shows that microbes play crucial roles in human health and disease. The exploration of the relationship between microbes and diseases will help people to better understand the underlying pathogenesis and have important implications for disease diagnosis and prevention. However, the known associations between microbes and diseases are very less. We proposed a new method called non-negative matrix factorization microbe–disease associations (NMFMDA), which used Gaussian interaction profile kernel similarity measure, to calculate microbial similarity and disease similarity, and applied a logistic function to regulate disease similarity. And, based on the known microbe–disease associations, a graph-regularized non-negative matrix factorization model was utilized to simultaneously identify potential microbe–disease associations. Moreover, fivefold cross-validation was utilized to evaluate the performance of our method. It reached the reliable area under receiver operating characteristic curve (AUC) of 0.8891, higher than other state-of-the-art methods. Finally, the case studies on three complex human diseases (i.e., asthma, inflammatory bowel disease, and colon cancer) demonstrated the good performance of our method. In summary, our method can be considered as an effective computational model for predicting potential disease–microbe associations.

【 授权许可】

Unknown