High BUSCO duplication rate for low coverage HiFi data

[BitaoQiu]

Dear authors,

I am applying Hifiasm to assembly our Hifi read data (DNA from a single individual for each species). However, we notice that the duplication rate (based on BUSCO) is negatively correlated with estimated read coverage. And we further confirmed with previous published short-read assembly that the high duplication rate in our low-coverage assemblies is not species-specific (see attached). And I think the correlation is because low sequencing depth leads to difficulty in assembling heterozygous regions.

We are now using Purge_Dups to remove potential false duplication, but may I ask is there any other parameters that we could use in hifiasm, where one can account for the sequencing depth?

Species	Sex	Estimated coverage	N50	# contig	Estimated genome size	Contig length	Completeness	Singleton	Duplicated	Fragment	Missing
Odontotermes	F.	28	5800000	1291	1194220328	1449477267	99.4%	97.4%	2 %	0.4%	0.2%
Trinervitermes	U.	10	605000	6072	348834812	1726917765	98.1%	83 %	15.1%	1 %	0.9%
C. secundus	U	13	1143825	3705	1182129932	1296819224	99.5%	90.1%	9.4%	0.2%	0.3%
C. secundus(Short reads, reference)	Mix		1184893	55483	1182129932	1018932804	98.8%	97 %	1.8%	0.7%	0.5%
Macrotermes bellicosus PNP	M	12	564726	5671	1214356371	1411697024	99.3%	89 %	10.3%	0.4%	0.3%
Macrotermes bellicosus (HiFi, reference)		21	11 MB	428	1113805679	1341469195	99.7%	96.5%	3.2%	0	0.3%

[chhylp123]

For the coverage issue, I guess so. The built-in purge_dups of hifiasm could be tuned like: https://hifiasm.readthedocs.io/en/latest/faq.html#p-large, which may work better in some cases.

[BitaoQiu]

Thank you for the reply... However, after setting --purge-max = 20 (twice the homozygous read coverage based on K-mer) and -s = 0.3, the BUSCO duplication rate still remains very high (15%).

We also tested genome assembly with Hi Canu. By setting correctedErrorRate=0.105 (https://canu.readthedocs.io/en/latest/parameter-reference.html#correctederrorrate), we got a BUSCO duplication rate of 3.9%, which is more reasonable to us...

[chhylp123]

Are the assemblies of HiCanu more contiguous than those of hifiasm?

[BitaoQiu]

It seems so.. but of course, it's only for the species that with very low sequencing coverage ...

Species	Method	N50	# contig	Estimated genome size	Contig length	Completeness	Singleton	Duplicated	Fragment	Missing
Trinervitermes10x	Hifiasm	605 KB	6072	X	1726917765	98.1%	83 %	15.1%	1 %	0.9%
	Hifiasm + purge	716 KB	3610		1457166292	96.9%	94.8%	2.1%	1.5%	1.6%
	Canu	734 KB	10555		1593394689	98.6%	94.7%	3.9%	1.1%	0.3%

[chhylp123]

Thanks a lot. It is reasonable as we haven't optimized hifiasm for such low coverage of reads.