GSPHI

GSPHI predicts phage-host interactions (PHIs) from DNA and protein sequence data to identify bacterial targets for phage therapy.

Key Features:

• Deep Learning Architecture: Predicts potential interactions between bacteriophages and bacterial hosts using DNA and protein sequence data.
• NLP Initialization: Initializes node representations for phages and bacterial hosts via a natural language processing algorithm applied to biological sequences.
• Graph Embedding (SDNE): Uses Structural Deep Network Embedding (SDNE) to extract local and global information from interaction networks.
• Deep Neural Network (DNN): Applies a deep neural network to detect phage-host interactions from extracted network features.

Scientific Applications:

• ESKAPE dataset benchmarking: Evaluated on the ESKAPE dataset under 5-fold cross-validation, reporting accuracy 86.65% and AUC 0.9208 and outperforming other existing methods.
• Phage therapy target identification: Identifies candidate sensitive bacteria for phage therapy by predicting PHIs.
• Cross-taxonomic applicability: Applicable to both Gram-positive and negative bacterial species.

Methodology:

Initial node representation via natural language processing (NLP); graph embedding using Structural Deep Network Embedding (SDNE) to capture local and global network information; feature-based interaction prediction using a deep neural network (DNN); evaluation with 5-fold cross-validation reporting accuracy and AUC.