Related Experiment Video
Updated: Jun 7, 2026
08:31
Isolation and Genome Analysis of Single Virions using 'Single Virus Genomics'
Published on: May 26, 2013
10.9K
Benchmarking informatics approaches for virus discovery: caution is needed when combining in silico identification
Bridget Hegarty1, James Riddell V2, Eric Bastien3
1Department of Civil and Environmental Engineering, Case Western Reserve University, Cleveland, Ohio, USA.
Msystems
|February 20, 2024
Summary
Combining viral identification tools requires caution. Benchmarking showed that specific combinations, particularly those including VirSorter2, optimize viral recovery from metagenomes, while excessive combinations do not improve accuracy.
Area of Science:
- Virology
- Bioinformatics
- Metagenomics
- Microbial Ecology
Background:
- Accurate viral identification from metagenomic data is crucial for understanding ecological impacts.
- Researchers often combine multiple bioinformatics tools to maximize viral recovery, but the efficacy of this approach is unvalidated.
- Existing viral identification tools present challenges in optimizing viral recovery for specific research needs.
Purpose of the Study:
- To benchmark combinations of six widely used viral identification informatics tools (rulesets).
- To assess the impact of tool combinations on viral recovery and accuracy in mock and environmental metagenomes.
- To provide guidance on optimal strategies for *in silico* viral identification.
Main Methods:
- Benchmarking of six viral identification tools (VirSorter, VirSorter2, VIBRANT, DeepVirFinder, CheckV, Kaiju) and their combinations (rulesets).
- Testing rulesets against mock metagenomes with diverse sequence types and aquatic metagenomes.
- Evaluation of tool performance based on accuracy metrics like the Matthews Correlation Coefficient (MCC).
Main Results:
- Six rulesets achieved equivalent accuracy (MCC = 0.77), with each including VirSorter2 and five using a 'tuning removal' rule.
- Combining multiple tools did not consistently lead to optimal performance; specific 2-4 tool combinations were most effective.
- The accuracy plateau (MCC 0.77) may be partly due to inaccuracies in reference sequence databases.
- The best-performing ruleset identified significantly more viral sequences in virus-enriched aquatic metagenomes (44%-46%) than in cellular metagenomes (7%-19%).
Conclusions:
- Combining viral identification tools should be done cautiously; not all combinations improve performance.
- The VirSorter2 ruleset combined with an empirically derived 'tuning removal' rule is recommended for robust viral identification.
- Future improvements in viral identification algorithms should be coupled with careful curation of reference sequence databases.
