PHIStruct

PHIStruct predicts phage-host interactions by using structure-aware protein embeddings of receptor-binding proteins (RBPs) to determine host specificity among ESKAPEE bacterial genera.


Key Features:

  • Structure-aware embeddings: Utilizes structure-aware embeddings generated by SaProt to incorporate structural information and capture host-specificity signals not evident from sequence alone.
  • RBP focus: Targets receptor-binding proteins (RBPs) of bacteriophages as primary determinants of host specificity.
  • Multilayer perceptron (MLP): Processes SaProt embeddings with an MLP architecture to predict interactions between phages and hosts, focusing on the ESKAPEE genera.
  • Performance metrics: Maintains a high and stable F1 score across confidence thresholds and sequence similarity settings, balancing precision and recall.
  • Low sequence similarity advantage: Improves class-averaged F1 scores compared to other machine learning tools and BLASTp for phages with low sequence similarity to known phages.
  • Robustness across confidence thresholds: Shows a 7%–9% improvement over structure-non-informed machine learning tools and a 5%–6% increase over BLASTp at high-confidence thresholds (>50%).

Scientific Applications:

  • Predicting Host Specificity: Accurately identifies potential hosts among the ESKAPEE genera by analyzing phage RBPs.
  • Enhancing Phage Therapy Research: Facilitates identification of suitable phages for targeting specific bacterial pathogens.
  • Broadening Computational Biology Scope: Integrates structural data into predictive models to advance computational methods in bioinformatics and molecular biology.

Methodology:

Generates structure-aware protein embeddings using SaProt and inputs them into a multilayer perceptron (MLP) to predict phage-host interactions from receptor-binding proteins, focusing on ESKAPEE bacterial genera.

Topics

Details

License:
MIT
Maturity:
Emerging
Cost:
Free of charge
Tool Type:
command-line tool
Operating Systems:
Mac, Linux, Windows
Programming Languages:
Python
Added:
1/14/2025
Last Updated:
1/14/2025

Operations

Data Inputs & Outputs

Prediction and recognition

Publications

Gonzales MEM, Ureta JC, Shrestha AMS. PHIStruct: improving phage–host interaction prediction at low sequence similarity settings using structure-aware protein embeddings. Bioinformatics. 2024;41(1). doi:10.1093/bioinformatics/btaf016. PMID:39804673. PMCID:PMC11783280.

Documentation

Installation instructions', 'Quick start guide', 'Citation instructions', 'Command-line options
https://github.com/bioinfodlsu/PHIStruct

Downloads

Links