DADA2 in Mothur?


Hi. I heard in a course I attended recently that now QiimeII is more powerful and more asked to be used when reviewers judge a manuscript, due to the implementation of DADA2 but not because of the dicotomy between OTU vs ASV but because of the algorithms implemented to filter and deal with sequences before clustering in ASV.
What is the opinion of mothur loving people about that? and would it be possible to include DADA2 algorithms inside Mothur as it was implemented in QiimeII?
Sorry I am not experienced but I am reluctant to accept “don’t use Mothur anymore”
Thank you, cheers.


LOL. Who said that? Did they show any actual data? Reviewers who trash manuscript for using mothur over QIIME or QIIME over mothur are lazy and don’t deserve to review manuscripts. They need to provide specific points for why one should be used over the other.

For instance, I would have serious problems with papers that use open or closed reference clustering in QIIME based on the series of papers we have published over the past few years.

I would also have problems with people using ASVs and rejecting OTUs out of hand. ASVs have a real risk of splitting 16S rRNA genes from the same genome into different ASVs. When you add that dada fits a model with hundreds of parameters and then applies a ridiculously low p-value threshold, you start to see that it has problems. The reality is that dada looks better than mothur’s pre.cluster because they remove all of the singletons. If you leave them in, the performances are about the same. Removing singletons will have a negative impact on the ability to calculate alpha and beta diversity metrics and estimate relative abundance.

You might also want to read a lengthy blog post I wrote on mothur and QIIIME.

Qiime2 outperforming Mothur?

Can I cite this forum post in my response to a reviewer about why I left in singletons when I performed my analysis?


Certainly! You can also feel free to plagiarize. I honestly don’t know why these reasons aren’t universally accepted.

1 Like

I 100% agree with Pat over here,
Recently I ran a large dataset about 532 Samples with DADA2 and guess what, ended with ~24000 ASV(aka OTU) even uclust gave 11000


One of my users just got a review saying that they need to rerun all their analyses with Deblur, that OTUs against a database is invalid (um mothur doesn’t do db based clustering). Then went on to say that they shouldn’t have rarefied. It was the strangest review I’ve seen. lack of understanding of tools while also demanding that they use very specific tools (I think all in phyloseq, maybe the reviewer took a phyloseq workshop and knows the one and only way to analyze sequences?)


Hi! I was told to learn Phyloseq package to analyse data and produce nice plots, is it not right? If we wanted to use it, do you know how could we produce the tree to input together with the otu table? and if that package needs a tree or it is only used if we wanted to compute unifrac distances but other measures of distance or even the statistical tests could be performed with mothur outputs? Or doing the sequence analysis with qiime is the only way for using phyloseq package in R?


If you learn R, you can do anything and not worry about phyloseq. I’m also not clear how anyone can produce a meaningful tree using MiSeq data.

If you’re looking for materials to help you learn R with standard packages, I’d encourage you to check out my minimalR tutorial.


phyloseq is sort of an R dialect. I learned R first so find phyloseq frustrating. phyloseq encourages bad graphs by making them easy to do-stacked bargraphs with tens or hundreds of categories? phyloseq would love to make that for you.

1 Like
closed #10

This topic was automatically closed 10 days after the last reply. New replies are no longer allowed.