SEVA model integrates protein language models with structural and evolutionary features to predict virulence factors and antibiotic resistance genes (microbiomejournal.biomedcentral.com)

• SEVA integrates pLMs, structural features, and evolutionary information to predict VFs and ARGs from genome data.
• Achieved 97.13% accuracy, outperforming seven existing tools including Diamond and FoldSeek.
• Model is open source at GitHub; trained on >20,000 genes from five reference databases.

"Researchers developed SEVA, a model that combines protein language models (pLMs) with structural and evolutionary features to predict virulence factors (VFs) and antibiotic resistance genes (ARGs) from genome sequencing data. Using over 20,000 genes and five reference databases, SEVA achieved 97.13% accuracy, outperforming tools like Diamond, VRprofile, and FoldSeek. The model reduces false negatives by capturing conserved protein structures and site mutations. SEVA is open source and available on GitHub, providing a reliable tool for concurrent VF and ARG identification in pathogen surveillance."