Predicting protein-protein interactions between human and hepatitis C virus via an ensemble learning method

Area of Science:

• Computational biology
• Bioinformatics
• Virology

Background:

• Hepatitis C virus (HCV) chronically infects 170 million globally, causing over 350,000 deaths annually.
• HCV disrupts host cell functions by altering protein-protein interactions (PPIs).
• Predicting host-virus PPIs is crucial but challenging due to experimental limitations and a scarcity of computational models for inter-species interactions.

Purpose of the Study:

• To develop a computational method for predicting protein-protein interactions (PPIs) between human and hepatitis C virus (HCV) proteins.
• To address the gap in inter-species PPI prediction models, particularly in the context of host-pathogen relationships.

Main Methods:

• An ensemble learning approach was designed, utilizing Random Forest, Naïve Bayes, Support Vector Machine, and Multilayer Perceptron as base classifiers.
• A Multilayer Perceptron served as a meta-learner to integrate predictions from base classifiers.
• Protein feature vectors were generated using six descriptors: amino acid composition (ACC), pseudo amino acid composition (PAC), evolutionary information, network centrality, tissue information, and post-translational modification information.

Main Results:

• The ensemble model achieved 83% accuracy and 94% specificity in 10-fold cross-validation on a benchmark dataset.
• An independent test set yielded 84% accuracy and 92% specificity.
• The proposed method demonstrated superior performance compared to existing approaches for host-pathogen PPI prediction.

Conclusions:

• The developed ensemble learning method effectively predicts human-HCV protein-protein interactions.
• This computational approach offers a valuable tool for understanding host-pathogen dynamics and can be applied to other viral infections.
• The findings highlight the potential of ensemble methods in advancing the prediction of inter-species PPIs.