Cutadapt prior to dada2 denoise-paired

[benjjneb (Ben Callahan)]

@hsapers

I’m also curious about allowing for some variability in the read length. I’m assuming that even a single nucleotide difference between reads will lead to separate feature calling (which is why when separating samples from different runs the trim and trunc parameters must be the same if the resulting feature tables are to be merged). Does the dada2 error model generation take into account some read length variability? Should I just chop of the first 19(20) (515F, 926R) bases and not account for other variability here?

During denoising sequences that differ solely by variation in length will be grouped together. Pure length variation is not part of the error model.

However, quality control that introduces artefactual length variation (e.g. by trimming when quality drops below some threshold) does tend to reduce the sensitivity of denoising, and thus is not recommended. Instead truncating at a fixed length cutoff is typically the better approach.