DeepHVI: A multimodal deep learning framework for predicting human-virus protein-protein interactions using protein language models

Understanding human-virus protein-protein interactions is critical for studying molecular mechanisms driving viral infection, immune evasion, and propagation, thereby informing strategies for public health. Here, we introduce a novel multimodal deep learning framework that integrates high-confidence experimental datasets to systematically predict putative interactions between human and viral proteins. Our approach incorporates two complementary tasks: binary classification for interaction prediction and conditional sequence generation to identify interacting protein partners. By leveraging protein language models and multimodal fusion, the framework demonstrates improved accuracy in identifying biologically relevant interactions. For empirical validation, we applied this method to predict severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)-human interactions, identifying candidate proteins absent from training data, several of which were corroborated by independent studies. These predictions offer critical insights into potential therapeutic targets, facilitating the design of antiviral drugs and vaccines. By enabling rapid, cost-effective discovery pipelines, our study contributes to pandemic preparedness and public health interventions, underscoring its value in combating emerging infectious diseases.