Machine learning approach to predict protein-protein interactions between human and hepatitis E virus: revealing links to hepatocellular carcinoma

Abstract

Hepatitis E virus (HEV) remains a widespread yet underrecognized cause of acute and chronic liver disease, contributing to an estimated 44,000 to 70,000 deaths annually. With recurrent HEV outbreaks and limited treatment options, there is an urgent need for effective therapeutic development. Though, many aspects of the HEV life cycle, particularly the host-virus interactions that shape infection outcomes, remain poorly understood. Understanding virus-host protein-protein interactions (PPIs) is essential for targeted drug discovery, a task increasingly facilitated by advancements in machine learning. Here, we applied KNN, SVM, NV, and RF to predict novel HEV-human PPIs, offering critical insights into pathogen virulence strategies and potential therapeutic targets. Among 88 descriptors, the most effective features were GC content, Gene Ontology semantic similarity, Normalized frequency of beta-structure, and normalized frequency of alpha-helix and coil. Among the models, DT achieved the highest sensitivity (77%). while Logistic Regression (LR) had the highest specificity (52%) and the best accuracy of 0.61, showing robust prediction of positive and negative cases. Additionally, our proposed LR model has predicted novel potential targets in hepatitis E virus-human PPIs, which have been further validated through Gene Ontology enrichment analysis. Gene Ontology and disease enrichment analyses revealed HEVs impact on immune modulation, lipid metabolism (FASN, APOB, EPHX2), and oncogenic pathways (FN1, JUN, HRAS, TP53), supporting its potential role in liver pathology and hepatocellular carcinoma (HCC). These findings provide novel insights into HEV-host interactions, offering targets for future antiviral strategies.