megSAP benchmarks for germline pipelines

All benchmarks are perfomed with the megSAP release 2026_06.
As reference genome GRCh38 with decoy chromosomes, without ALT chromosomes and with masked false duplications was used.

Small variants benchmark

All small variant benchmarks are done on the GIAB reference sample NA12878/HG001 using the gold-standard variant list v4.2.1.
The analyses were performed with the short-read and long-read single sample pipelines.

Sensitivity, positive predictive value (PPV) and genotyping accuracy were measured using our validation tool.

The following data was used for the benchmark:

Type	DNA Fragmentation	Kit	Sequencer	Mean depth	Mean insert size
short-read WES	Covaris	Twist custom exome kit (Core, RefSeq, Mito and custom content)	NovaSeq 6000 - 2x105bp PE	101.1	304.4
short-read WGS	Covaris	Illumina TruSeq DNA PCR-Free	NovaSeq 6000 - 2x159bp PE	42.6	377.3
long-read WGS	-	Oxford Nanopore Tech. Ligation Sequencing Kit V14e (SQK-LSK114)	PromethION P24	44.3	-

The benchmarks were performed on the GIAB high-confidence region with at least 15x coverage:

Test	%ROI in high conf region and covered 15x	SNV			InDel
		sensitivity	PPV	genotyping	sensitivity	PPV	genotyping
short-read WES - bwa-mem2, DeepVariant 1.9	88.29%	0.9949	0.9848	0.9994	0.9727	0.9714	0.9986
short-read WES - DRAGEN 4.4	88.29%	0.9906	0.9968	0.9991	0.9835	0.9849	0.9944
short-read WGS - bwa-mem2, DeepVariant 1.9	81.09%	0.9963	0.9992	0.9998	0.9914	0.9971	0.9991
short-read WGS - DRAGEN 4.4	81.09%	0.9979	0.9983	0.9998	0.9972	0.9966	0.9994
long-read WGS (HAC6.0, minimap2, clair3v1.2)	81.34%	0.9989	0.9987	0.9999	0.8764	0.9456	0.9859
long-read WGS (SUP5.2, minimap2, clair3v1.2)	81.34%	0.9997	0.9959	0.9998	0.9042	0.9420	0.9843

Small variants benchmark - coding region

To allow a comparison of WES, WGS and lrGS independent of the coverage, we also perfomed a benchmark without depth cutoff on the coding region of all protein-coding genes padded by two bases to include the consensus splice sites.

Test	SNV			InDel
	sensitivity	PPV	genotyping	sensitivity	PPV	genotyping
short-read WES - bwa-mem2, DeepVariant 1.9	0.9634	0.9828	0.9991	0.8793	0.9725	0.9978
short-read WES - DRAGEN 4.4	0.9584	0.9971	0.9990	0.8755	0.9807	0.9934
short-read WGS - bwa-mem2, DeepVariant 1.9	0.9930	0.9875	0.9989	0.9770	0.9884	0.9980
short-read WGS - DRAGEN 4.4	0.9938	0.9972	0.9993	0.9847	0.9772	0.9981
long-read WGS (HAC6.0, minimap2, clair3v1.2)	0.9989	0.9984	0.9999	0.9674	0.9825	0.9980
long-read WGS (SUP5.2, minimap2, clair3v1.2)	0.9998	0.9928	0.9999	0.9655	0.9618	0.9980

CMRG benchmark

For genome sequening, we also performed the CMRG benchmark based on the NA24385/HG002 sample.

Type	DNA Fragmentation	Kit	Sequencer	Mean depth	Mean insert size
short-read WGS	Covaris	Illumina TruSeq DNA PCR-Free	NovaSeq 6000 - 2x159bp PE	42.4	348.3
long-read WGS (ONT)	-	Oxford Nanopore Tech. Ligation Sequencing Kit V14e (SQK-LSK114)	PromethION P24	44.3	-
long-read WGS (PacBio)	-	LongRead Sequencing with PacBio (LR-PB-highcov)	Revio	38.0	-

All benchmarks were performed on GIAB high-confidence regions with at least 15x coverage.

Test	SNV			InDel
	sensitivity	PPV	genotyping	sensitivity	PPV	genotyping
short-read WGS - bwa-mem2, DeepVariant 1.9	0.9803	0.9969	0.9981	0.9333	0.9526	0.9957
short-read WGS - DRAGEN 4.4	0.9791	0.9962	0.9979	0.9467	0.9301	0.9955
ONT long-read WGS (HAC6.0, minimap2, clair3v1.2)	0.9877	0.9603	0.9989	0.7582	0.8195	0.9735
ONT long-read WGS (SUP5.2, minimap2, clair3v1.2)	0.9905	0.9417	0.9980	0.7993	0.8058	0.9712
PacBio long-read WGS (minimap2, DeepVariant 1.9)	0.9972	0.9941	0.9992	0.9732	0.9769	0.9958

Structural variant calling benchmarks

All structural variant benchmarks are done on the GIAB reference sample NA24385/HG002 using the draft SV benchmark v1.1.
The analyses were performed with the short-read and long-read single sample pipelines.

samples: see CMRG benchmark

Sensitivity and positive predictive value (PPV) were measured using Hap-Eval.

Test	coverage	sensitivity	PPV
short-read WGS - Manta 1.6.0	44.55	0.3649	0.9656
short-read WGS - DRAGEN 4.4	44.22	0.6453	0.9474
ONT long-read WGS (HAC, minimap2) - Sniffles 2.7	42.32	0.8986	0.9636
ONT long-read WGS (SUP5.2, minimap2) - Sniffles 2.7	42.36	0.9027	0.9648
PacBio long-read WGS (minimap2)- Sniffles 2.7	38.0	0.9104	0.9729