Merging outside QIIME2: results for DADA2 and deblur

[Nicholas_Bokulich]

@llenzi

I linked, with a final set of about 500 sequences per sample (merging percentage close to 0!). However if I merge the same set with PEAR I got a merge percentage around 97% (rising to almost 99% if not excluding sequences containing Ns).

This is indeed an issue with how you are performing your trimming. The low merge yield is a strong indicator.

PEAR is performing the merge prior to quality trimming, and then may be assigning an arbitrary Q value to the overlapping bases (I am not familiar with PEAR but other read joiners do this).

You have two routes:

1. if you want to use dada2, you will need to adjust your trimming parameters. See this post, which is reporting a very similar problem. If that does not help you solve the problem on your own, please report back with the same data (quality score plots, exact commands used) and answers to questions in that thread, and we can help solve this.
2. If you want to use deblur, you can join reads prior to denoising (either using PEAR and then importing, or using q2-vsearch to perform similar read joining from within QIIME2). This may be the easier approach, since you already have joined reads and would not need to fuss over trimming parameters.

@llenzi

My question to you is, what is your experience on merging the 2x300bp within DADA2? I’m trying to understand if is me, still not quite right on how to set the trimming lengths on DADA2.

This is probably an issue with your trimming parameters, as V3-V4 should be joinable with dada2, provided read quality is good enough to support overlap. We get lots of questions about trimming lengths on this forum, and it all depends on the primers, expected amplicon length, and read quality. See above for how to diagnose this or get support. You certainly should not be losing close to 100% of your reads.

@llenzi

Pairing within DADA2, gives quite skewed results compared with the expected.

That is expected, given that you are losing close to 100% of reads and so this is skewing the read profile.

@llenzi

To show the difference, is there a way to pass you the picture in private, so you can see what I mean, please?

Yes, you can send a direct message to forum moderators (e.g., me). But I understand what you mean — and linked to a very similar post above — so you do not need to send. I am in full agreement: the dada2 results need to be re-done because your first analysis had inadequate trimming parameters.

@llenzi

In your latest tutorial on how import pre merged reads, you suggest NOT using DADA2. Still there is a single sequence mode. Can you explain further why DADA2 is not suited for that, please?

Joining reads disrupts the Q score profile, which dada2 uses to predict read errors. Many read joiners assign arbitrary Q scores to overlapping bases, which confuses dada2. So you need to use dada2 on the unjoined paired reads and dada2 will join these reads after denoising forward/reverse reads separately. You must adjust trimming parameters to make sure there is at least 20 bp of overlap between forward and reverse reads.

I hope that helps!