MHMDA: Human Microbe-Disease Association Prediction by Matrix Completion and Multi-Source Information

Research output: Contribution to journalArticle


Microbes are vital in human health. It is helpful to promote diagnostic and treatment of human disease and drug development by identifying microbe-disease associations. However, knowledge in this area still needs to be further improved. In this paper, a new computational model using matrix completion to predict human microbe-disease associations (mHMDA, Fig. 1) is developed. First, we extract the disease feature by Gaussian kernel-based similarity and symptom-based similarity. Meanwhile, the microbe feature is computed by Gaussian kernel-based similarity. As treating potential association as the missing elements of a matrix, the matrix completion is adopted to get the potential microbe-disease associations. Leave-one-out cross-validation (LOOCV) is carried out which get the AUC (The area under ROC curve) of 0.928 showing the effectiveness of mHMDA. Furthermore, 5-fold CV get the AUCs of 0.8838 ± 0.0044 (mean ± standard deviation). Moreover, through the four case studies (asthma, inflammatory bowel disease (IBD), type 2 diabetes (T2D), and type 1 diabetes (T1D)), we find that nine, ten, nine, and eight of top-ten inferred microorganisms for the four diseases are previously verified by experiments. All these results indicate the effectiveness of mHMDA. mHMDA might be helpful to infer the disease-related microorganisms.


External organisations
  • Shandong University
Research areas and keywords

Subject classification (UKÄ) – MANDATORY

  • Biomedical Laboratory Science/Technology


  • matrix completion, microbe-disease association prediction, Microbial community
Original languageEnglish
Article number8768373
Pages (from-to)106687-106693
Number of pages7
JournalIEEE Access
Publication statusPublished - 2019 Jan 1
Publication categoryResearch