Integrative Analysis for Identifying Co-Modules of Microbe-Disease Data by Matrix Tri-Factorization With Phylogenetic Information.

Microbe-disease association relationship mining is drawing more and more attention due to its potential in capturing disease-related microbes. Hence, it is essential to develop new tools or algorithms to study the complex pathogenic mechanism of microbe-related diseases. However, previous research studies mainly focused on the paradigm of "one disease, one microbe," rarely investigated the cooperation and associations between microbes, diseases or microbe-disease co-modules from system level. In this study, we propose a novel two-level module identifying algorithm (MDNMF) based on nonnegative matrix tri-factorization which integrates two similarity matrices (disease and microbe similarity matrices) and one microbe-disease association matrix into the objective of MDNMF. MDNMF can identify the modules from different levels and reveal the connections between these modules. In order to improve the efficiency and effectiveness of MDNMF, we also introduce human symptoms-disease network and microbial phylogenetic distance into this model. Furthermore, we applied it to HMDAD dataset and compared it with two NMF-based methods to demonstrate its effectiveness. The experimental results show that MDNMF can obtain better performance in terms of enrichment index (EI) and the number of significantly enriched taxon sets. This demonstrates the potential of MDNMF in capturing microbial modules that have significantly biological function implications.