Genome size confirmation through genome assembly

[bioman1]

Hi all,

I have estimated genome size through jellyfish software.

1. Is it possible to confirm genome size estimation through de novo genome assembly through velvet?.

2. Can I use the same kmer which used for genome size estimation for genome assembly?. For e.g. I have estimated genome size with kmer 17. Can I use the same kmer for genome assembly.

[kbradnam]

If your genome is very heterozygous, then there is a risk that a genome assembler will resolve heterozygous regions into two contigs. Thus, your assembly size may not be a good reflection of the actual genome size. In the Assemblathon 2 contest, I think there was one assembly that was more than twice the expected genome size.

[Brian Bushnell]

17 is probably too short for a good assembly, and possibly even for estimating genome size in an organism with significant repeat content. If your organism is not haploid, you may want to look at the kmer frequency histogram and estimate size manually, probably using a larger kmer of 25+. For a diploid, you would do this by ignoring the low-frequency error kmers, then multiplying the number of unique kmers under the first peak (the heterozygous peak) by 0.5, multiplying the number of unique kmers under the second (homozygous) peak by 1, multiplying the number of unique kmers under the third (2-copy repeat) peak by 2, etc.

[jebowers]

The problem with an assembly and genome size is that with assemblies with many contigs, the same sequence will be in the overlaps between contigs that have not been merged yet. This produces an assembly larger than the genome size. At the same time any repeat sequence will only be represented one time in the assembly, and that results in an assembly smaller than the genome size. Finally hetrozygosity and sequencing errors will result in an assembly larger than the genome size.

Scaffold assemblies can be additionally misleading as sequences represented by Ns could be counted several times.

It all depends on the genome, but as the contig sizes climb the assembly size will approach the genome size.


To estimate the genome size based on kmer frequency. Estimate the total kmers covered by the 1x peak and add half the number of kmers in the 1/2x peak (if heterozygous). Find the total count of times these kmers have been seen. This is the single copy fraction (and single copy genome size). Sum the total counts of all kmers observed, subtract out the counts of all Kmers <.5x coverage (errors).

Find the fraction of this number that corresponds to the single copy sequences (calculated before), and then divide by that fraction to get the total genome size.

This works better than estimating the size of 2nd, 3rd etc peaks, as they will often be very difficult to distinguish as the copy number increases, and with large genome plants or animals some repeats can be very high copy number.

Finally to get a very accurate estimate it may be necessary to remove kmers corresponding to the chloroplast or mitochondrial genomes as they can be a significant fraction (>10%) in some DNA preps for (for plants at least).