I tried dada2 on my data. After that I tried also vsearch uchime-denovo as suggested in the tutorial https://docs.qiime2.org/2018.4/tutorials/chimera/. I also played a bit with the parameters --p-minh of vsearch uchime-denovo (I tried both default and 1.0), as suggested in another post saying that value 1.0 of --p-minh should avoid excluding false positives chimeras. I could see that vsearch could identify many chimeras in my data (with --p-minh 1.0, 1.218 features are marked as chimeric and 13.460 sequences are removed), even if I ran it after dada2 which is supposed to remove chimeras as well. Why is this happening? Should I be worried on dada2 results regarding chimeras and so apply as well vsearch uchime-denovo after dada2 to be sure that chimeras have been properly removed?
Thank a lot
I do not know enough about the algorithms used by dada2 and q2-vsearch for chimera filtering to say for sure, but I expect there are enough differences that the two would not necessarily all the same seqs as chimera.
I expect that dada2 may also catch chimeric seqs that q2-vsearch may not.
That is not to say one is "better" or "worse" than the other. I am not aware of a head-to-head benchmark on good test data, so I would not be so quick to say that more seqs identified as chimera is "better".
After all, some of these could be false positives.
I would not worry too much, per my comments above. If you are very concerned, I would recommend checking the literature to see if there are any gold-standard methods and tests sets that you could use to evaluate.
Let's see if others weigh in, particularly anyone with better insight on the differences between the chimera algorithms used by each plugin.
In general, no you shouldn't be worried. The dada2 chimera removal method is better for ASV data than uchime, which was designed with the assumption of OTUs (no very similar sequences in data) and evaluated on such data as well.
For a longer discussion you can read the uchime2 paper, where Edgar discusses some of the pitfalls of chimera removal especially w/ ASVs. (The "denoised de novo" or "DDN" mode of uchime2 described in that paper is very similar to the dada2 chimera removal method.)