Increase tolerable collision rate for low quality read libraries

[EmeraldSama94]

Hello @arangrhie! First of all, thank you for all your work and support!

I'm implementing meryl/merqury in a pipeline to evaluate a bunch of public genomic assemblies of neglected parasites. While some of them used newer sequencing technologies, a good portion relied on genomic reads with rather low quality, specially early Illumina and PacBio reads, and even some now discontinued technologies such as 454 and IonTorrent. As many scientists of our community still use these assemblies, it is a necessity to evaluate them as well. I know that you do not recommend running meryl/merqury for these low quality read libraries. However, I noticed that the tolerable collision rate parameter in the best k script is set to the same error rate generally expected from a Illumina sequencing, that is 0.001, equivalent to a Q30. So, I was wondering that, if this association is true, I could increase the tolerable collision rate for read libraries with lower quality and continue to run meryl/merqury without big issues. Does this make any sense or am I getting the meaning of the collision rate completely wrong?

Thank you in advance!

[arangrhie]

Hi @EmeraldSama94 ,

Your understanding is correct, you could increase the tolerable collision rate to match the sequencing platform. This may make the k size smaller. I wouldn't recommend going blow k=21 though, because the smaller the kmer size becomes, it looses power to detect errors, which ends up over-estimating the QV.

If 454 or IonTorrent are the only available platforms, then yes you'd need to adjust the tolerable collision rate and use a smaller k size. Still, I think it's important to communicate what sequencing platform was used for collecting the kmers, rational behind choosing a smaller k size (increasing the tolerable collision rate), and acknowledge the QV presented are potentially over-estimating the actual QV since the assembly error kmers that are also present in the reads are inherently not detectable.

Best,
Arang

[EmeraldSama94]

Hi Arang!

Thank you so much for your response! I'll make sure to properly communicate the smaller k-mer size and the potential over-estimated QV. :)

On another note, you mentioned that k-mers should not be smaller than 21; however, even with high-quality Illumina reads, the best k-mer size I get is either 17 or 18. Should I set k-mer size to 21 regardless of it? Could this smaller k-mer size be due to the size of the genomes I'm working with (around 30-40 Mbp)?

Best regards,
Samuel

[arangrhie]

Yes, consider the suggested size k as the 'minimum' guidance given the genome size. Something larger is considered more conservative.