MHIPM: Accurate Prediction of Microbe-Host Interactions Using Multiview Features from a Heterogeneous Microbial...

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• Microbe-host interactions (MHIs) are crucial for human health.
• Wet-lab methods for identifying MHIs are slow and expensive.
• Computational approaches are needed to prioritize candidate microbe-host interactions.

Purpose of the Study:

• To develop a novel deep learning method, MHIPM, for predicting MHIs.
• To utilize multisource biological information for MHI prediction.
• To provide an efficient computational tool for identifying potential microbe-host relationships.

Main Methods:

• Constructed a heterogeneous microbial network including human proteins, viruses, phages, and bacteria.
• Employed ESM-2 and doc2vec with self-attention for feature extraction from protein sequences.
• Utilized GraphSAGE for capturing network features within the heterogeneous network.

Main Results:

• MHIPM demonstrated superior performance compared to seven baseline algorithms and four variants across three prediction tasks.
• Case studies and molecular docking experiments validated the model's effectiveness.
• The model successfully identified plausible candidate microbes for biological experiments.

Conclusions:

• MHIPM is an efficient and robust method for predicting microbe-host interactions.
• The tool aids in prioritizing candidate microbes for experimental validation.
• MHIPM contributes to a better understanding of the microbiome's role in human health.