VirDetect-AI: a residual and convolutional neural network-based metagenomic tool for eukaryotic viral protein...

Area of Science:

• Virology
• Bioinformatics
• Computational Biology

Background:

• Metagenomic data analysis presents challenges for virus identification due to viral genome diversity and evolution.
• Current AI methods for virus detection are often binary classifiers, lacking specificity and relying on nucleotide sequences.

Purpose of the Study:

• To introduce VirDetect-AI, a novel tool for identifying eukaryotic viruses in metagenomic datasets.
• To overcome limitations of existing AI approaches by enhancing specificity and utilizing amino acid genomic data.

Main Methods:

• VirDetect-AI employs a hybrid model combining convolutional neural networks (CNNs) and residual neural networks (ResNets).
• The model analyzes complex amino acid genomic data to extract hierarchical features and detailed patterns.

Main Results:

• VirDetect-AI achieved high performance across all metrics: 0.97 sensitivity, 0.98 precision, and 0.98 F1-score.
• The tool accurately classifies metagenomic sequences into 980 distinct viral protein classes.

Conclusions:

• VirDetect-AI significantly advances the identification of eukaryotic viruses within metagenomic data.
• The tool enhances understanding of viral ecology and facilitates the discovery of novel viruses and their characteristics.