VHI-Pred: A Multi-Feature-Based Tool for Predicting Human-Virus Protein-Protein Interactions.

Area of Science:

• Virology
• Computational Biology
• Bioinformatics

Background:

• Viral diseases represent a major public health concern, necessitating efficient methods to understand host-pathogen protein-protein interactions for therapeutic strategies.
• Traditional methods for studying these interactions are resource-intensive and slow, especially considering the rapid mutation rates of viruses.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting protein interactions between viral pathogens and human hosts.
• To identify key factors influencing these host-pathogen interactions.
• To enhance the efficiency of analyzing viral-human protein interactions for drug discovery.

Main Methods:

• Construction of prediction models using Random Forest (RF), XGBoost (XGB), and Artificial Neural Networks (ANN).
• Utilized features including physicochemical properties, motifs, and amino acid sequences.
• Performance evaluation using accuracy, precision, sensitivity, specificity, and K-fold cross-validation. Integrated dimensionality reduction and clustering for model optimization.

Main Results:

• The initial RF, XGB, and ANN models achieved accuracies of 87%, 86%, and 86%, respectively.
• Integration of dimensionality reduction and clustering techniques improved the RF model's accuracy to 90%.
• Demonstrated significant enhancement in the efficiency of analyzing viral-human host interactions compared to traditional methods.

Conclusions:

• Machine learning, particularly the optimized RF model, offers a highly efficient approach to predict viral-human protein interactions.
• This computational strategy significantly accelerates the understanding of host-pathogen dynamics, aiding therapeutic development.
• The developed models and insights provide valuable resources for future research in virology and drug discovery, with results available via a web application.