RNN-VirSeeker

RNN-VirSeeker applies deep learning to identify short viral sequences in metagenomic datasets, enabling accurate detection and analysis of viruses in microbial communities.

Key Features:

• Deep Learning-Based Approach: Employs a deep learning methodology to improve detection of short viral sequences from metagenomic data.
• Training Data: Trained on 500 bp sequences sampled from Virus and Host RefSeq genomes.
• Performance Metrics: Reports AUROC of 0.9175, recall of 0.8640, and precision of 0.9211 for 500 bp sequences.
• Comparison with Existing Methods: Outperforms VirSorter, VirFinder, and DeepVirFinder and achieves higher AUPRC and AUROC on a CAMI dataset and a human gut metagenome.

Scientific Applications:

• Virology: Identification and characterization of viral sequences and the potential discovery of novel viruses from metagenomes.
• Microbial Ecology: Assessment of viral diversity and functional roles within microbial communities.
• Human Health: Profiling viruses in human microbiomes to support studies of virus–host interactions and tracking pathogen evolution.

Methodology:

Requires Python packages "sklearn", "numpy", and "matplotlib"; users provide query contigs in a ".csv" file with one contig per line; the input sequences are processed by the tool's deep learning model which assigns scores to each contig, where higher scores indicate greater likelihood of being viral.