Fig. 2: Comparison of Vclust with other tools on various datasets.
From: Ultrafast and accurate sequence alignment and clustering of viral genomes
a, Difference between predicted and expected tANI values for 10,000 bacteriophage genome pairs with simulated mutation events. b, Correlations with VIRIDIC tANI values for 22,607 complete bacteriophage genome pairs. c, Wall time and peak memory usage for processing 4,244 bacteriophage genomes (32 threads). Vclust and VIRIDIC include clustering, while FastANI and skani only calculate ANI. d, Venn diagrams comparing numbers of contig pairs meeting MIUViG thresholds (ANI ≥ 95% and AF ≥ 85%) predicted by BLASTn (purple) and other tools (red). The boxen plot shows the error distribution of predicted ANI and AF values relative to corresponding BLASTn-based reference values for 4,361,743 contig pairs meeting MiUVUG thresholds. The center line denotes the median, while each box level from the median contains half of the remaining observations. e, Wall time and peak memory usage for calculating ANI and AF among 15,677,623 IMG/VR contigs (64 threads). BLASTn values were estimated from a random sample of 1,000 query contigs. Vclust was tested in its default setting and with a 0.2 fraction of k-mers used at the ‘prefilter’ step. f, Wall time and peak memory usage of Vclust’s clustering algorithms for grouping IMG/VR contigs into vOTUs.
