DeepGVS: a bimodal deep learning framework integrating coding-sequence and protein-structural representations for virulence factor prediction.

DeepGVS achieves 87.50% accuracy in bacterial virulence factor prediction by integrating coding sequences and protein structures, outperforming existing models by up to 6.30 percentage points.

• Why it matters: Accurate identification of virulence factors is crucial for understanding bacterial pathogenicity and developing antimicrobial strategies, but current methods overlook key structural and genetic information.
• What they did: DeepGVS combines multi-scale CDS-derived features with protein sequence and 3D structural data using a bimodal deep learning approach, including graph attention networks, Bi-Mamba architecture, and a neural additive model.
• The result: The model's bimodal integration significantly improves prediction accuracy, highlighting the importance of combining genetic and structural data, though taxonomic factors may influence performance.