A web-based artificial intelligence system for label-free virus classification and detection of cytopathic effects

Area of Science:

• Virology
• Artificial Intelligence
• Cell Biology

Background:

• Traditional viral identification methods are labor-intensive and require specialized expertise.
• Cytopathic effects (CPEs) are visual indicators of viral infection in cell cultures.
• There is a need for automated, efficient tools for viral diagnostics.

Purpose of the Study:

• To develop and validate an AI-powered automated system, AIRVIC, for detecting and classifying label-free cytopathic effects (CPEs) induced by various viruses.
• To assess AIRVIC's performance in identifying specific viral strains (SARS-CoV-2, BAdV-1, PIV3, BoAHV-1, BoGHV-4) in different cell lines (Vero, MDBK).
• To establish AIRVIC as a novel tool for virological diagnostics and antiviral research.

Main Methods:

• Development of AIRVIC using convolutional neural networks, with ResNet50 architecture.
• Training the AI model on a dataset of 40,369 microscopy images of infected cell cultures.
• Testing AIRVIC's accuracy in CPE detection and virus classification across different viral strains and cell lines.

Main Results:

• AIRVIC achieved 100% accuracy in detecting a specific strain of Bovine alphaherpesvirus 4 (BoGHV-4) in MDBK cells.
• The system demonstrated varying accuracy for different virus-cell line combinations, with the lowest at 87.99% for BoGHV-4 in Vero cells.
• Multi-class accuracy for bovine virus classification reached 87.61% in MDBK cells, decreasing to 63.44% without cell line specification.

Conclusions:

• AIRVIC represents the first AI application for distinguishing animal virus infections in cell culture, offering unbiased infectivity scoring.
• The AI system streamlines viral isolation and facilitates antiviral efficacy testing.
• AIRVIC is available as a web-based platform, enabling global access for researchers in viral diagnostics.