Estimating the bacterial genome size using Kmer frequency

[Krish_143]

Hi,

How to estimate the bacterial genome size (GC rich) when there was no close reference genome ?

At first i tried jellyfish and generated the histogram plots (for all the avail kmers) and here the exact peak (what i guess) were identifed and calculated, but i am only getting less than half (too less ) off the genome size when compared to generated assemly produce from soapdenovo2.

And then i tried kmergenie (for all different kmers) same i am not getting proper estimation..

* Illumina hiseq : Paired end data : Read length 100bps ;
* GC perecent : 63 % ; (Read_1)
* Duplicates in fastq : ->48% (Read_1)
* Read_1 :10128605 (data from FastQC)


Any Suggestions could be really greatfull..

Thank you very much..

[krobison]

#distinct kmers / 2 should be the genome size with a few important caveats

1) Including erroneous kmers will inflate the count, so typically would count only those kmers with a count of >=2

2) Repeat regions will be collapsed

3) regions that just don't show up will be missed, again underestimating. With high G+C genome, there may be regions simply missing from Illumina or with very low coverage.

Ray produces the kmer statistics in a way that is easy to parse & generate these estimates.

Assemblies are often a bit too large due to missed overlaps. If you convert these histograms to genome size estimates, how big a range is covered?

Even without a reference, the taxonomy of the bug may suggest a range -- though you could well have something outside that range.

[Krish_143]

Hi krobison,

when i estimted the genome size using kmer information (histogram, kmer Peaks)
ESti_Gsize: 2.8mb (at Kmer 31)
Assembled Gsize using SoapDenovo : 5.7mb (Draft)

I will check with the Ray and very thanks krobison for the quick response.

[rchikhi]

I sometimes observe that SOAPdenovo contigs (not scaffolds) tend to assemble more than the genome size. Did you run a Velvet assembly, and if so, what was the assembly size?