Table 2.

Quality and completeness of HG002 diploid assemblies.

Asm	Contig NG50 (Mb)	Scaffold NG50 (Mb)	Hamming error	QV	#Errors (Chr X & Y)
Downsampled (35× HiFi / 60× ONT-UL)
Verkko	12.90		0.13%	52.18	10
Verkko + trio	80.77	102.55	0.13%	52.40	10
Verkko + Hi-C	58.24	82.42	0.16%	52.48	10
LJA	0.39		0.14%	55.75	7
Hifiasm (unitigs)	0.35		0.23%	61.05	2
Hifiasm + trio	64.50		0.06%	60.86	26
Hifiasm + Hi-C	66.34		0.79%	60.57	37
Full-coverage (>100× HiFi / 85× ONT-UL)
Verkko	17.35		0.05%	54.91	5
Verkko + trio	134.00	135.80	0.05%	55.77	5
Verkko + Hi-C	68.32	85.97	0.05%	55.57	5
HPRC curated	72.70	146.75	0.13%	61.35	26

We evaluated contig NG50 and phase accuracy for all assemblies using Merqury ⁵⁸. With a goal of phased assemblies, pseudo-haplotype outputs from Flye and Hifiasm were not considered. We evaluated hamming error rate (the fraction of non-dominant allele variants in a unitig) and Phred QV ⁵⁹ using Merqury. We measured errors versus the recently finished HG002 chromosomes X and Y ¹⁴ using QUAST. Results are shown for DeepConsensus HiFi; results from the original HiFi reads are in Supplementary Table 6. LJA and Hifiasm do not output scaffolds, so no scaffold NG50 values are reported. Verkko assemblies used 105× HiFi coverage (mean 17.5 kbp, Data availability) while the HPRC curated assembly used 130× HiFi coverage (mean 14.8 kbp) ¹¹. The best result for each coverage level and category is highlighted in bold.