Paired-data more distant than different samples

[mw55309]

Hi

Following on from #32 where I (accidentally) sketched both reads of a pair, I get these curious results:

Mash sketch/dist sample against itself:

%> mash dist SRR1262629.mash.msh SRR1262629.mash.msh
SRR1262629_1.t.fastq.gz SRR1262629_1.t.fastq.gz 0 0 1000/1000
SRR1262629_2.t.fastq.gz SRR1262629_1.t.fastq.gz 0.0234513 0 440/1000
SRR1262629_1.t.fastq.gz SRR1262629_2.t.fastq.gz 0.0234513 0 440/1000
SRR1262629_2.t.fastq.gz SRR1262629_2.t.fastq.gz 0 0 1000/1000

Mash sketch/dist two samples against each other:

%> mash dist SRR1262629.mash.msh SRR1262625.mash.msh
SRR1262629_1.t.fastq.gz SRR1262625_1.t.fastq.gz 0.0166117 0 545/1000
SRR1262629_2.t.fastq.gz SRR1262625_1.t.fastq.gz 0.0189925 0 505/1000
SRR1262629_1.t.fastq.gz SRR1262625_2.t.fastq.gz 0.0174175 0 531/1000
SRR1262629_2.t.fastq.gz SRR1262625_2.t.fastq.gz 0.0212923 0 470/1000

So when looking at read 1 and read 2 from the same sample (SRR1262629) they only share 440/1000 hashes.

When comparing different samples (SRR1262629 vs SRR1262625) they actually share more hashes! (470 - 545)

Is this expected??

Cheers
Mick

[ondovb]

I think the issue is just that each pair of SRR1262629 doesn't have enough coverage to saturate the genome (If I'm looking at the right stats, each should be <2x). When compared against a set with more saturation, there would be more overlap, apparently enough to outweigh genomic variants.