I am jumping into the discussion here - working with Rune, not with the same samples, but with Illumina data being analyzed similarly through DADA2.
Concerning the discussion on trimming: Estaki´s information on trimming is a bit opposing to Callahan´s feedback here: https://github.com/benjjneb/dada2/issues/175 - taking into consideration that we do have paired-end seq.
The dada2 plugin in QIIME2 (simplified):
qiime dada2 denoise-paired --i-demultiplexed-seqs SampleData[PairedEndSequencesWithQuality] --p-trunc-len-f --p-trunc-len-r --p-trim-left-f --p-trim-left-r --output-dir
For our sequences --p-trim-left-f --p-trim-left-r are constant for all runs, it removes the primer regions. Callahan emphasize this as being important as a non-consistent trim-left option will give sequences of different lengths after merging sequences.
Meanwhile the --p-trunc-len-f --p-trunc-len-r will be adjusted after inspection of the quality plots, and therefore vary somewhat between runs. Again with the answer given by Callahan this has not been a concern to me so far! And I still do not understand why varying truncating length in order to keep Q-scores above a chosen limit is more troubling than varying Q-score levels which would be the consequence of setting a strict truncating length for all runs…? This being said - we are fortunate enough not to encounter the issue of having to truncate so early that overlap is a concern.
To summarize my understanding of things just to see if you need to correct me:
Pre-dada2 quality filtering as with trimmomatic can be troublesome because you interfere with the error model.
With paired-end sequences make sure trim-left option is kept consistent between all your runs such that sequences are of the same length.
Truncating length needs not be the exact same across runs, but the more similar the better - and do not ignore that you need a minimum overlapping!