A Deep Learning-Based Method for Identification of Bacteriophage-Host Interaction.

Area of Science:

• Bioinformatics
• Microbiology
• Computational Biology

Background:

• Multi-drug resistance (MDR) poses a significant global health threat, necessitating innovative treatments.
• Phage therapy offers a promising alternative, but requires precise matching of therapeutic phages to target bacteria.
• Deep learning excels at identifying complex patterns for accurate biological predictions.

Purpose of the Study:

• To develop PredPHI, a deep learning tool for predicting phage-host interactions using sequence data.
• To enhance the accuracy and robustness of phage-host interaction prediction.

Main Methods:

• Collected over 3000 phage-host pairs and protein sequences from public databases (PhagesDB, GenBank).
• Extracted sequence-based features and selected high-quality negative samples using K-Means clustering for a balanced training set.
• Developed a predictive model using a deep convolutional neural network.

Main Results:

• PredPHI achieved an 81% area under the receiver operating characteristic curve on the test set.
• The K-Means clustering method for negative sample selection proved significantly more robust than random selection.
• Demonstrated the efficacy of deep learning in predicting phage-host interactions.

Conclusions:

• PredPHI is an accurate and useful tool for identifying phage-host interactions from sequence data.
• The study highlights the potential of computational approaches in advancing phage therapy.
• Accurate phage-host matching is key to the successful application of phage therapy against MDR pathogens.