ML-AdVInfect: A Machine-Learning Based Adenoviral Infection Predictor

Area of Science:

• Virology
• Bioinformatics
• Machine Learning

Background:

• Adenoviruses (AdVs) cause diverse infections, impacting clinical relevance and therapeutic vector potential.
• Viral entry hinges on specific ligand-receptor interactions, crucial for predicting adenoviral infectivity.
• Existing data on virus-host interactions are fragmented, lacking integrated computational models for infection prediction.

Purpose of the Study:

• To develop a machine learning model for predicting adenoviral infection in hosts.
• To establish a computational approach for interpreting virus-host interaction data.
• To assess the potential for cross-species adenoviral transmission.

Main Methods:

• Constructed a comprehensive adenovirus-host interaction dataset using known AdV receptors and ligands.
• Utilized sequence data and publicly available virus-host protein-protein interaction (PPI) tools.
• Developed a Support Vector Machine (SVM) model with an RBF kernel for infection prediction.

Main Results:

• The ML-AdVInfect model achieved high performance metrics: 0.88 sensitivity, 0.83 specificity, and 0.86 AUC.
• Successfully predicted adenoviral infection capacity and potential cross-species host shifts.
• Demonstrated the model's effectiveness in screening viral infectivity.

Conclusions:

• ML-AdVInfect is the first computational tool to effectively predict adenoviral infection potential.
• The developed methodology can be adapted for predicting other viral infections.
• This approach aids in understanding and anticipating viral pathogenesis and host tropism.