Comparison with others

Application notes

• all sizes in MiB
• spvcf was run in lossless mode (-n).
• sav is 2.0.1. It was run in default mode and --phasing none. The "sprs" mode is with --sparse-fields set for all fields.

Dataset 1 (Uk Biobank)

• generated by Graphtyper
• internal to deCODE
• 20KB region on chr1
• ~200.000 samples
• FORMAT string: GT:AD:MD:DP:GQ:PL

	Original (.vcf.gz)	Original (bcf)	Savvy	Savvy sprs	spVCF.gz	Shark	popVCF	DELTA (.vcf.gz)	DELTA (.bcf)
original	14,001	11,788	10,657	10,640	10,433	6,082	5,618	5,062	4,312
binned	10,607	8,901	8,675	8,663	5,333	4,332	2,800	2,984	2,554

The "binned" version is a file where the values are bucketed to increase compressibility.

Dataset 2 (Uk Biobank / WES)

• internal to deCODE
• ~200.000 samples
• FORMAT string: GT:DP:AD:GQ:PL:RNC

	Original (.vcf.gz)	Original (bcf)	Savvy	Savvy sprs	spVCF.gz	Shark	popVCF	DELTA (.vcf.gz)	DELTA (.bcf)
original	176	191	193	153	59	97	26	47	53

• The BCF of the original is larger than the VCF.GZ of the original. This is an indicator that DELTA might not be the best compression.
• It is still better than all other tools except POPVCF.