Using same samples as baseline for two groups (longitudinal analyses)

Dear QIIME developers,

I have samples collected from 3 timepoints as shown below:

  • Timepoint 1 - Collected @ 4 weeks after subjects had consumed same baseline diet
  • Timepoint 2 - Collected @ 12 weeks after subjects were split into two groups and consumed DIET 1 or DIET 2
  • Timepoint 3 - Collected @ 24 weeks after subjects were split into two groups and consumed DIET 1 or DIET 2

My problem is that my 4 week samples act as timepoint 1 (baseline) samples for both diets 1 and 2. I am struggling to figure out how to tell qiime in the metadata file to use week 4 samples as timepoint 1 for both diet 1 and diet 2 so I can compare longitudinal changes between diets. The only way I can think is to duplicate the samples prior to running DADA2 and I don’t think that is the correct way.

This is preventing me from running longitudinal tests to compare the diet groups, as I cannot use timepoint 1 as a baseline for BOTH diet 1 and 2.

Is there any way to do this?

Thank you.

Hi @JenKelly,

One path of least resistance would be to use qiime feature-table filter to split your data (Im not sure if it does metadata, check the documentation) into a datasets where you've got T1 and T2, and then T1 and T3, and then run your commands.

I think I would also probably run the T2 vs T3, particularly if you have a group on the same diet. (I came up with three possible ways this experiment might be designed... wasnt sure which and that influences the way Id recommend handling this.)

If you have multiple timepoints on the same diet, they might act as a measure of change on a stable diet, and give you a comparison group to be able to comment on the magnitude of your change from Baseline to T2 or T3.

If you've got the crossover design, you might be at risk of getting a bias due to the first diet. (The microbiome is complex and fun in that way. There's a really nice paper by Sonnenberg et al about microbiome recovery after diet.... Science, I think.) In which case, you might see more change with one diet to another than you did the inverse, or from baseline to a diet.

I still think you will get more powered from a paired sample design than you will from a cross sectional approach: the human microbiome is pretty specialised, and its hard to make someone stop looking like themselves with any intervention Ive seen except FMT (Extreme diet and medication included.) So, I think I'd focus there. I might also play with the multivariate models like gneiss that let you handle longitudinal data.


1 Like

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.