Easy way to perform the linear-mixed-effects analysis on the important features from the Feature volatility analysis?

I just finished the Feature volatility analysis, and I’m trying to perform the Linear mixed effect models analysis on the important features from the Feature volatility analysis. However, the important features form the filtered-table could be as many as more than 10, making it fussy and repetitive when I use the –p-metric parameter in the qiime longitudinal linear-mixed-effects because only one feature could be analyzed at a time. I know a loop could solve this problem. But it’s still still a bit cumbersome as I need to export the feature IDs first and then import them to the qiime longitudinal linear-mixed-effects analysis. I’m wondering if I can import the filtered-table directly to the qiime longitudinal linear-mixed-effects analysis? Any help would be much appreciated!
Thank you,
Hongbin Liu

1 Like

Hi @11131,
Great questions. The feature volatility analysis is meant to be exploratory and allow you to identify and then visualize the temporal changes in abundance of potentially important features. So I would not run an LME on every single important feature, necessarily. I recommend:

  1. use the feature volatility visualization to generate specific hypotheses:
    1. identify features that fit some criteria of interest, e.g., relative abundance, % change, increase/decrease following some biologically defined hypothesis about changes in abundance such as following an intervention, maybe also taxonomic membership in biologically relevant groups.
    2. examine their volatility plot to determine if the temporal change fits the assumptions of LME or other appropriate tests (e.g., linearity)
  2. choose LME or another appropriate test to test those specific hypotheses

That sounds more or less what you are doing, but I just recommend carefully selecting specific organisms of interest and testing those instead of all important features.

Yes — see the help documentation for that method, a feature table is an optional input.

Good luck!

1 Like

Hi Nicholas,
Thanks for your prompt reply!
I know you chose Bifidobacterium and Faecalibacterium to do the analysis in the q2-longitudinal paper. Here another question is that in the paper you fit three separate linear models to examine Bifidobacterium abundance between 0 to 6 and 6 to 18 months of life and Faecalibacterium between 6 to 12 months of life, since the trajectories appear approximately linear within each of these biologically sensible developmental phases. How are the time points for the developmental “windows” determined? Is this determined by observing the temporal changes in abundance of Bifidobacterium, or the LME method you mentioned above to determine the exact time points for the genus? How?
In my study, I monitor the gut microbiota dynamics after dietary fiber intervention through longitudinal sampling to screen the microbes that response to the fiber and associate ecological alteration. I know it’s will be much better to choose important features to perform the LME analysis, but I still want to test all the features in the filtered-table in case I miss some important features that associated the intervention. I checked the Linear mixed effects modeling and found that a feature table was indeed an optional input. If I want to test all the features in the filtered-table, should I ignore the --p-metric parameter? And how should I set the output --o-visualization?
Thank you for your help!
Hongbin Liu

1 Like

Determined manually, based on the temporal abundance profile.

It would be better to use a method like aldex2 or ancom to test treatment effects on ALL features, as these are compositionally aware and designed to control for false positives when testing a large number of inputs.

you would need to set up a loop as you described previously… inputting a table still requires a single metric at a time.

Thank you for your reply! I’ll try the analysis with a loop.