Prediction of human-virus protein-protein interaction using heterogeneous siamese neural network

Abstract

Viral infections represent pathological conditions arising from the intrusion of viruses into host cells and their replication. The onset of infection is intricately tied to the interplay between viral and host cell proteins. Thus, elucidating these protein-protein interactions assumes a pivotal role in the encompassing prevention, treatment, and control of viral infections. Given traditional laboratory experimentation’s prohibitively high costs and time-intensive nature, researchers have increasingly turned to computational approaches for predicting human-virus protein-protein interactions. Despite the performance of these computational approaches, a challenge persists in the need for an effective protein representation that adequately captures their structural intricacies. In this paper, we present PBS, a novel model for the prediction of protein-protein interactions between viruses and humans. PBS leverages the transformers to effectively represent proteins. The model unified the latent space for human and virus proteins through the implementation of heterogeneous siamese neural networks.The model achieves an accuracy score of 81.41%, an area under the ROC curve score of 87.35%, an area under the precision-recall curve score of 87.78%, an F1 score of 81.58%, and a precision score of 80.84%. These metrics collectively underscore the satisfactory performance of the PBS model. Furthermore, we assess the model’s predictive capabilities in discerning interactions between proteins associated with the H1N1 influenza virus and human proteins.

Zahra Ghorbanali

Assistant professor

Fatemeh Zare

Associate Professor

My research interests include bioinformatics, computational biology and artificial intelligence.