Low sequence numbers after merging using DADA2 but reads are long enough

[Nicholas_Bokulich]

Hi @kfri,
I am not sure what is going wrong — with everything you describe it sounds like you should get sufficient overlap but you are still losing ~50% of your reads at merging. My guess is that there is more variation in the amplicon length than you may expect — I do not know enough about that primer pair, but you could use extract-reads to estimate amplicon size as described here:

Can we use extracted reference reads to calculate expected amplicon sizes?

Dear all, I’m interested in calculating the expected amplicon sizes of a primer set targetting V1-2 region of the 16S rRNA. Are the reference sequences extracted from the database by qiime feature-classifier extract-reads suitable for that purpose? The default minimum combined primer match identity threshold (0.8) is quite lenient compared to what’s usually used in the in silico evaluation of universal primer sets, which allows no or a single non-3′-mismatch. References to in silico evaluatio…

But that does not explain why usearch merge_pairs is giving you better merging; I suspect it could be because dada2 is trimming just a little bit prior to the merge, enough to miss the mark for half your sequences. You could use qiime vsearch join-pairs to join the reads first, then use q2-deblur for denoising. join-pairs would be similar to usearch fastq_mergepairs so we should see similar results — maybe give that a try and see if you still see sufficient merging?