Deep-HPI-pred

Deep-HPI-pred predicts and classifies host-pathogen protein-protein interactions to support analysis of molecular mechanisms and identification of potential therapeutic targets in infectious diseases.

Key Features:

• Network-Driven Feature Learning: Integrates network topology into feature learning to improve HPI prediction accuracy.
• Multilayer Perceptron (MLP) Models: Uses Multilayer Perceptron models trained on comprehensive datasets such as citrus and CLas bacteria interactions for prediction and classification.
• Topological Feature Evaluation: Incorporates topological metrics including Eigenvector Centrality to enhance predictive performance.
• Validation and Performance: Models were validated on independent datasets and achieved a Matthews correlation coefficient (MCC) greater than 0.80 for HPI predictions.
• Gene Ontology (GO) Term Information: Associates Gene Ontology annotations with proteins in the interaction network to provide functional context.
• Versatility Across Systems: Demonstrated high accuracy across multiple host-pathogen systems, including plant-pathogens (98.4% and 97.9%), human-virus (94.3%), and animal-bacteria (96.6%).

Scientific Applications:

• Molecular mechanism discovery: Predicts HPIs to help decipher molecular mechanisms underlying host-pathogen interactions.
• Therapeutic target identification: Supports identification of potential protein targets for therapeutic intervention.
• Disease progression analysis: Enables analysis of interaction networks to study disease progression and host defense mechanisms at the molecular level.

Methodology:

Network-driven feature learning that integrates topological features (including Eigenvector Centrality) with Multilayer Perceptron (MLP) models trained on datasets such as citrus and CLas, validated on independent datasets with performance assessed by Matthews correlation coefficient (MCC), and mapping Gene Ontology (GO) terms to proteins.