I have 3 years of data; the first (2023) was sequenced using Illumina and the last two (2024 & 2025) were sequenced using AVITI. When processing them, everything runs smoothly, but when I try to combine them, they are incomparable. Here's a Bray-Curtis plot to show what I mean (left is 2023, right is 2024 & 2025):
My question is: Has anyone tried to combine Illumina and AVITI data before? Is there a step in the pipeline that should be changed for different sequencing platforms?
I know the read lengths are different so I denoised them separately then merged them after, but the problem remains. I've attached my code below. I am working in version 2026.1, installed through conda. Any input is appreciated. Thanks!
Thanks for bringing this question to the forums, and sharing your full pipeline.
I know exactly where to start!
This would prevent ANY two data sources from merging, even if they were both Illumina.
The amplicon sequenced must match exactly, or else the ASVs will be different and cannot be merged. (Still a good idea to denoise each run separately, because it lets DADA2 learn and correct the error profile of each run.)
What are the two lengths you sequenced?
If you used only the forward read, what trimming settings would theoretically lead to identical amplicons between Illumina and AVITI? (I ask because paired reads can range in length)
I believe the Illumina reads were 250bp and the AVITI reads were 300bp. So is the problem caused by trimming/truncating them to different lengths? That would definitely make sense!
I just noticed I did in fact trim them to different lengths. Would trimming AVITI reads down to 230 cause an issue though?
If the amplicon itself was 300 bp, then I would expect the output from 250 and 300 bp reads would be basically the same after merging with dada2 or vsearch. So maybe that's not the issue!
It really does have to match exactly, like a single extra basepair on the end would cause no overlap!
Let's look for clues...
Maybe the next step is the pull the most common ASV sequences from both cohorts and compare? See if you can see if one set is just a little longer or shorter than the other? We should find either 100% identical ASVs, or evidence of how they are different.
