DeepPBI-KG: a deep learning method for the prediction of phage-bacteria interactions based on key genes

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Phage therapy is gaining traction due to rising antimicrobial resistance.
• Existing computational methods for phage-bacteria interaction prediction are limited.
• There's a need for efficient methods to screen phages for therapeutic use, especially at the intraspecific level.

Purpose of the Study:

• To develop a novel computational approach for predicting phage-bacteria interactions.
• To create a prediction tool, DeepPBI-KG, that utilizes key genes and proteins.
• To enable accurate intraspecific prediction for phage screening.

Main Methods:

• Feature engineering using key phage and bacteria genes/proteins.
• K-means sampling for high-quality negative sample selection.
• Development of DeepPBI-KG, a deep neural network-based prediction tool.

Main Results:

• DeepPBI-KG achieved high prediction accuracy (AUC 0.89, AUPRC 0.92) on an independent test set.
• Validation confirmed the influence of key genes and proteins on phage-bacteria interactions.
• Intraspecific prediction experiments with Klebsiella pneumoniae showed promising applicability.

Conclusions:

• The proposed feature engineering and prediction approach enhance robustness and generalizability.
• DeepPBI-KG offers a reliable and efficient method for rapid phage screening.
• This work provides new insights for advancing phage therapy development.