Table 3.

Results on NCBI catalogs

Pipeline	$mi{n}_{\mathit{lin}}$	$ma{x}_{\mathit{lin}}$	MALVIRUS		bcftools		lofreq
			Precision	Recall	Precision	Recall	Precision	Recall
A	20	50	.951	.919	.909	.932	.778	.856
B	20	50	.972	.955	.921	.909	.788	.823
A	50	100	.951	.924	.907	.931	.757	.856
B	50	100	.967	.959	.916	.909	.763	.821
A	100	100	.952	.924	.908	.932	.738	.857
B	100	100	.968	.962	.916	.909	.744	.822

For each catalog, we report the precision and recall achieved by MALVIRUS, BCFtools, and lofreq in calling the variations available in the catalog. The results are shown in terms of average over all the 41 considered samples. We highlighted in bold the best results. We considered 6 different catalogs built using pipeline A or B on the set of assemblies retrieved from NCBI, prefiltered using ${\tau }_N=1\%$ and then subsampled using different combinations of parameters $mi{n}_{\mathit{lin}}$ and $ma{x}_{\mathit{lin}}$