GSPHI: A novel deep learning model for predicting phage-host interactions via multiple biological information.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Antibiotic resistance is a growing global health threat.
• Bacteriophage (phage) therapy is a promising alternative for treating bacterial infections.
• Identifying phage-host interactions (PHIs) is crucial for understanding bacterial responses and developing effective phage therapies.

Purpose of the Study:

• To develop a computational framework for predicting phage-host interactions (PHIs).
• To leverage DNA and protein sequence information for accurate PHI prediction.
• To provide an efficient and economical alternative to traditional wet-lab experiments for PHI identification.

Main Methods:

• Developed GSPHI, a deep learning framework utilizing natural language processing for node representation initialization.
• Employed Structural Deep Network Embedding (SDNE) to extract network information.
• Applied a deep neural network (DNN) for predicting phage-bacterium interactions.
• Validated the model on the ESKAPE dataset using 5-fold cross-validation.

Main Results:

• GSPHI achieved 86.65% prediction accuracy and an AUC of 0.9208 on the ESKAPE dataset.
• The model significantly outperformed existing methods in predicting PHIs.
• Case studies demonstrated GSPHI's competence in identifying interactions across Gram-positive and Gram-negative bacteria.

Conclusions:

• GSPHI is an effective deep learning tool for predicting phage-host interactions.
• The framework provides valuable candidate phage-sensitive bacteria for experimental validation.
• GSPHI offers a time- and cost-efficient approach to advance phage therapy research.