Thank you @lizgehret for adding ancom-bc plugin for the last QIIME2 release! very useful addition.
One question though - is there a way to specify random/fixed effects in my model? Let's take a simple example of treatment and placebo groups who provided samples at baseline and after intervention.
In order to calculate differenital abundance change over time between groups you have to include participant identifier as a random effect. However, from the code it seems you can only call a formula without specify type of effect. Any way to do so/suggestions to overcome this issue? by the paper, it seems the method use linear regression framework, which should make it possible to my understanding.
Currently ancombc, implemented in Qiime2, do not support longitudinal design through random effects.
However, R package is updated. You can check Ancombc 2, which supports random and fixed effects.
Thanks, didn't know about the new R package.
I think it would be a very good addition to QIIME, especially as Maaslin which offers similar solution is not a QIIME plugin, and there is not a great way currently to deal with longitudinal data statistically speaking.
From a developer's perspective, wrapping someone else's code with QIIME2 is a complicated task. It is almost a requirement to have original developers involved. I am not aware of any efforts up until today.
Please, open a request in a relevant Forum (BioBakery or GitHub repo) if you think that's important, and let's make it happen by popular vote
For more complicated designs, you can use @gibsramen 's BIRDMAn for differential abundance tests, this includes longitudinal, LME, etc. There isn't a QIIME 2 plugin yet but I believe one will be in the future. The tool has a bit of learning curve and requires much more user involvement than something like ANCOM-BC, or Maaslin, but it is certainly much more powerful. The documentation is excellent with lots of great examples, might be worth a try.
Glad to hear you're using ANCOM-BC in q2-composition! Wrapping ANCOM-BC2 could be a good addition to q2-composition, but we will most likely wait until a paper has been published on it since it is still very new functionality (this was only added to the R package 3 months ago). Until then, either using the R package, or @Mehrbod_Estaki's suggestion of BIRDMAn would probably be your best bet.
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.