Predictive Modeling of Host-Pathogen Interactions

Abstract

Host-pathogen interactions (HPIs) -- the molecular, cellular, and immunological interplay between pathogens and their hosts -- determine infection outcomes, pathogen virulence, host immune response, and therapeutic vulnerabilities. Predictive modelling of HPIs supports vaccine design, antimicrobial development, infection risk assessment, and pandemic preparedness, but faces substantial complexity: pathogens span viruses, bacteria, fungi, and parasites with diverse genomes and lifecycles; host responses vary with genetics, age, immune history, and microbiome; and interactions span multiple scales from protein-protein binding to population-level epidemiology. Traditional experimental characterisation of HPIs (two-hybrid screens, co-immunoprecipitation, in vivo challenge studies) is slow and expensive, covering only a fraction of possible pathogen-host combinations. We present the Host-Pathogen Interaction Prediction Framework (HPIPF), evaluating five computational approaches -- sequence-based protein-protein interaction prediction, structure-based docking and complex modelling, network-based interactome prediction, machine learning host immune response prediction, and integrated multi-scale HPI models -- across four infection modelling tasks (pathogen-host PPI prediction, vaccine antigen identification, drug target prioritisation, and infection outcome prediction). Our Host-Pathogen Interaction Score (HPIS) measures interaction prediction accuracy, antigen identification validity, therapeutic target quality, outcome prediction performance, and pathogen diversity coverage. Integrated multi-scale models achieve the highest HPIS (0.926) through combining molecular, cellular, and clinical data, while structure-based docking achieves the highest interaction accuracy (0.960) for well-characterised pathogens through AlphaFold-enabled complex prediction.