Table 1.

Evaluation of minimap2 v2.22. Numbers in the bold fontface indicates the best performing tools

[Benchmark] Metric	v2.22	v2.18	Winno	lra
[sim-map] % mapped reads at Q10	97.9	97.6	99.0	97.3
[sim-map] err. rate at Q10 (phredQ)	52	52	38	24
[winno-cmp] rate of diff. (phredQ)	41	37	truth	18
[winno-cmp] CPU time (hour)	5.0	5.3	71.8	13.1
[winno-cmp] peak RAM (Gb)	17.1	14.4	9.6	12.4
[sim-sv] % false negative rate	0.5	2.0	0.5	1.4
[sim-sv] % false discovery rate	0.0	0.1	0.0	0.1
[real-sv-1k] % false negative rate	7.3	20.0	13.0	N/A
[real-sv-1k] % false discovery rate	2.7	2.4	2.7	N/A

Note: In [sim-map], 152 713 reads were simulated from the CHM13 telomere-to-telomere assembly v1.1 (AC: GCA_009914755.3) with pbsim2 (Ono et al., 2021): ‘pbsim2 –hmm_model R94.model –length-min 5000 –length-mean 20000 –accuracy-mean 0.95’. Alignments of mapping quality 10 or higher were evaluated by ‘paftools.js mapeval’. The mapping error rate is measured in the phred scale: if the error rate is e, $-10{\hspace{0.17em}\text{log}\hspace{0.17em}}_{10}e$ is reported in the table. In [winno-cmp], 1.39 million CHM13 HiFi reads from SRR11292121 were mapped against the same CHM13 assembly. 99.3% of them were mapped by Winnowmap2 at mapping quality 10 or higher and were taken as ground truth to evaluate minimap2 and lra with ‘paftools.js pafcmp’. [sim-sv] simulated 1000 50 bp to 1000 bp INDELs from chr8 in CHM13 using SURVIVOR (Jeffares et al., 2017) and simulated Nanopore reads at 30-fold coverage with the same pbsim2 command line. SVs were called with ‘sniffles -q 10’ (Sedlazeck et al., 2018) and compared with the simulated truth with ‘SURVIVOR eval call.vcf truth.bed 50’. In [real-sv-1k], small and long variants were called by dipcall-0.3 (Li et al., 2018) for HG002 assemblies (AC: GCA_018852605.1 and GCA_018852615.1) and compared to the GIAB truth (Zook et al., 2020) using ‘truvari -r 2000 -s 1000 -S 400 –multimatch –passonly’ which sets the minimum INDEL size to 1 kb in evaluation.