MHIPM: Accurate Prediction of Microbe-Host Interactions Using Multiview Features from a Heterogeneous Microbial Network
Jie Pan,Guangming Zhang,Yong Yang,Wenli Yang,Ning Mao,Zhuhong You,Jie Feng,Shiwei Wang,Yanmei Sun
DOI: https://doi.org/10.1021/acs.jcim.4c01296
2024-10-14
Abstract:Current studies have demonstrated that microbe-host interactions (MHIs) play important roles in human public health. Therefore, identifying the interactions between microbes and hosts is beneficial to understanding the role of the microbiome and their underlying mechanisms. However, traditional wet-lab experimental approaches are insufficient for large-scale exploration of candidate microbes, as they are costly, laborious, and time-consuming. Thus, it is critical to prioritize microbe-interacting hosts by computational approaches for further biological experimental validation. In this work, we proposed a novel deep learning-based method called MHIPM, to predict MHIs by utilizing multisource biological information. Specifically, we first constructed a heterogeneous microbial network that consisted of human proteins, viruses, bacteriophages (phages), and pathogenic bacteria. Next, we used one of the largest protein language models, ESM-2, and a document embedding model, doc2vec, combined with a self-attention mechanism to extract the interview features from protein sequences. Then, an inductive learning-based model, GraphSAGE, was used to capture the intraview features from the heterogeneous network. Experimental results on three prediction tasks indicated that the MHIPM model consistently achieved better performance than seven baseline algorithms and its four variants. In addition, case studies and molecular docking experiments for two human proteins further confirmed the effectiveness of our model. In conclusion, MHIPM is an efficient and robust method in predicting MHIs and provides plausible candidate microbes for biological experiments. MHIPM is available at https://github.com/JIENWU/MHIPM.