Table 1 Overview of tools to identify and predict phage sequences in microbial ecosystems
From: Gauge your phage: benchmarking of bacteriophage identification tools in metagenomic sequencing data
Software | Description | Reference |
|---|---|---|
DeepVirFinder | Predicts viral sequences via a k-mer-based deep learning method using convolutional neural networks (CNN). Based on VirFinder | [37] |
MARVEL | Machine learning tool for predicting phage sequences in metagenomic bins | [38] |
MetaPhinder | Integrates BLAST hits to multiple phage genomes in a database to identify phage sequences in assembled contigs | [39] |
viralVerify (metaviralSPAdes) | ViralVerify is a module of metaviralSPAdes which classifies contigs with a Naïve Bayes classifier based on Hidden Markov models protein hits | [40] |
PhaMers | Identifies phage sequences by a machine learning model based on k-mer frequencies | [41] |
PPR-Meta | Deep learning CNN approach to identify both phages and plasmids | [42] |
Seeker | Deep learning framework that uses long short-term memory model (LSTM) which does not depend on sequence motifs | [43] |
VIBRANT | Deep learning neural network based on protein signatures which also highlights auxiliary metabolic genes and pathways | [35] |
ViraMiner | Extension of DeepVirFinder that is trained to identify any virus that may colonise human samples | [44] |
VirFinder | K-mer-based machine learning method for identification of viral contigs | [45] |
virMine | Iterative pipeline that relies on the abundance of nonviral sequences in databases to strictly filter out unwanted contigs. Pipeline accepts both reads or assembled contigs | [46] |
VirMiner | Web-based pipeline that handles genome assembly, functional annotation using a variety of databases and identification of phage contigs via a random forest algorithm | [47] |
VirNet | Deep learning neural network using an attentional neural model trained on nucleotide viral fragments | [48] |
VIROME | Web-based pipeline that classifies viral sequences based on homology to databases and functionally annotates them. No local version | [34] |
VirSorter | Uses referenced-based and reference-free approaches in unison relying on probabilistic similarity models and referenced-based protein homology searches to increase novel virus detection | [28] |
VirSorter2 | Builds on VirSorter by applying machine learning to evaluate “viralness” using genomic features. Works with a wider variety of viral groups than its predecessor | [36] |
VirusSeeker | Made up of two BLAST-based pipelines — virome and discovery. Virome aligns reads to a curated database to identify viral sequences and compute their abundance in the sample. Discovery focuses on contig-based analysis to aid novel virus discovery | [49] |