geNomad

geNomad performs scalable classification, annotation, and discovery of mobile genetic elements (MGEs) from large-scale sequencing datasets to enable accurate identification and functional characterization of plasmids and viruses.

Key Features:

• Scalable MGE discovery: Performs classification, annotation, and large-scale discovery of mobile genetic elements from metagenomic and genomic sequencing datasets.
• Deep learning–based discrimination: Integrates gene-content information with deep neural network representations to discriminate plasmid and viral sequences.
• Marker protein library: Uses a library of >200,000 marker protein profiles to assign putative functional annotations.
• Conditional random field (CRF) boundary detection: Applies a CRF model to identify viral sequence boundaries of integrated proviruses within host genomic context.
• Automated taxonomic and functional inference: Provides automated taxonomic assignment and protein function inference for detected sequences.
• Extreme-scale processing: Optimized for high-throughput metagenomic analysis and demonstrated on datasets totaling ~2.7 trillion base pairs of sequencing data.
• Benchmarked classification performance: Reported Matthews correlation coefficients (MCC) of ≈0.95 for viral classification and ≈0.78 for plasmid classification.

Scientific Applications:

• Viral genome characterization: High-throughput characterization of viral genomes from metagenomic and genomic datasets.
• Provirus detection: Identification and boundary delineation of integrated proviruses within host genomes.
• Plasmid versus virus discrimination: Distinguishing plasmid and viral sequences in complex microbial communities.
• Protein function annotation: Assigning putative functions to proteins using marker protein profiles.
• Integration into analysis pipelines: Application within metagenomic and genomic pipelines for reproducible MGE identification and annotation.

Methodology:

Integrates gene-content information with deep neural network representations; applies a conditional random field (CRF) model to detect viral sequence boundaries of integrated proviruses; uses a library of >200,000 marker protein profiles for putative functional annotation and performs automated taxonomic assignment and protein function inference.