classification of non-overlapping reads treated with justConcatenate

[SoilRotifer (Mike Robeson)]

Hi @lukasbeule,

@lukasbeule

most previous studies that conducted sequencing from bulk soil using COI, the proportion of metazoan reads was often less than 5% (if more than 5% were metazoan reads

What COI primer sets have you tried? I find this surprising and has not been my experience.

• Leray et al. 2013
• Robeson et al. 2017
• O'Rourke et al. 2020

@lukasbeule

a high number of touchdown cycles was used during PCR, which is not what I want to do

This approach work quite well, especially if your touch down portion of the protocol is no more than ~10 cycles, give or take. I've had success using both Folmer and my own clade specific primers, and can also be combined with nested PCR too. See Robeson et al. 2011 for more details.

RE: 18S & Ns
I think it would be advisable to only use the forward reads. The addition of Ns may result in as bad, or worse, taxonomic specificity as compared to simply using the forward reads. 18S can vary quite a bit in length, even within the variable regions, and the addition of 10 bp can either be too little or too long for many clades. The impact of these Ns on classifiers can be wonky. I would think that classification approaches similar to vsearch/usearch might be less negatively impacted compared to sklearn. But I've not tested this. You could try making a mock community and test your classifactions using vsearch and sklearn. If you do, keep us posted!

-Cheers!
-Mike