VIBRANT

VIBRANT (Virus Identification By iteRative ANnoTation) recovers and annotates microbial viral genomes from metagenomic and genomic assemblies to enable virome functional characterization and identification of lytic viruses and integrated proviruses.

Key Features:

• Hybrid Machine Learning Approach: Employs a hybrid machine learning and protein similarity approach independent of traditional sequence features to recover and annotate bacterial and archaeal viruses.
• Neural Networks and v-score Metric: Uses neural networks trained on protein signatures combined with a v-score metric to improve identification of lytic viral genomes and integrated proviruses across diverse viral families.
• Genome Quality Assessment: Determines genome quality and completeness for viral sequences to support downstream analyses.
• Auxiliary Metabolic Gene Highlighting: Identifies viral auxiliary metabolic genes and associated metabolic pathways to facilitate virome functional analyses.
• Superior Performance Metrics: Recovered an average of 94% of viruses in comparative analyses of 120,834 metagenomically derived viral sequences versus VirSorter (56%), VirFinder (48%), and MARVEL (71%).
• Provirus Identification: Extracts integrated provirus regions from host scaffolds with performance comparable to or better than PHASTER, Prophage Hunter, and VirSorter.

Scientific Applications:

• Crohn’s Disease virome analysis: Applied to Crohn’s Disease datasets to reveal increased prevalence of Enterobacteriales-like viruses and dysbiosis-associated viral proteins in diseased versus healthy individuals.
• Virome functional profiling in microbial communities: Enables exploration of viral contributions to microbial community metabolism by identifying auxiliary metabolic genes and pathways.

Methodology:

Implements a hybrid machine learning and protein similarity approach using neural networks trained on protein signatures and a v-score metric; trained and validated on reference virus datasets alongside microbiome and virome data.