I see that both DADA2 and Deblur are denoising algorithms and could generate sort of 100% OTU tables. I was wondering are there preferences when picking one of these two methods? Like under which conditions, DADA2 is preferable, and vice versa.
Thank you,
Huang
Hi @hulfred,
I don’t believe an independent benchmark has been performed yet to help inform a decision here. The algorithms do have similarities, and key differences. And to be transparent, I am a developer with Deblur.
At a high level, DADA2 uses PHRED scores to inform an error model. Deblur deviates from this and instead uses a static error model. As a result, DADA2 is potentially better able to accommodate run specific errors. The benefit of a static error model though is that it potentially reduces study bias; the use of a static model was an important criteria for Qiita, the American Gut Project and the Earth Microbiome Project as these efforts are focused around meta-analysis.
@benjjneb, do you agree? I apologize in advance if I made a mischaracterization.
Best,
Daniel
Thanks for your explanation! I do appreciate it!
Hi @wasade
So do you mean if the data are from the same run, then DADA2 may be preferred and if the data are from different runs, then Deblur is better? Thanks.
Hi @Qianhao_Li,
I’m not aware of an independent benchmark which definitively explores that question. However, on the assumption that you are going to be comparing your results to other studies when writing up a manuscript, one strategy you may want to consider is using methods in common to what other similar studies have done in your environment / field / etc. For example, if I was performing a study in environment X, and researchers who study that environment typically use method Y, I’d likely first focus on that method as to have those results so they can be discussed if and as necessary, and only investigate other methods if method Y was ill-suited for the questions being asked. Does that make sense? I realize this is a bit abstract…
Best,
Daniel
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.