Beta diversity longitudinal analysis (3 time points or more)

Hello,

I am trying to find an appropriate command for running beta diversity data (Bray-Curtis) over time. Currently, the command in the Qiime2 tutorial covers just two-time points. Would you please let me know if there is any command that can handle beta diversity data more than two-time points (three or more time-points)?

Thanks!

1 Like

Hello @Barandouzi,
I am not sure what tutorial you are referring to but beta-group-significance can be run on any number of "Groups"(In this case time points). However you might be able to use beta-correlation if your timepoint metadata is numeric.

Hope that helps!
:turtle:

Thanks for your answer. I looked at the link you sent, but there is no option that we can put time or timepoints as a variable in the commands. I am interested in knowing how beta diversity changes among groups (not paired) over time (e.g., T1, T2, T3).

Hey @Barandouzi,
I think I misunderstood your goal. So you have groups that you are tracking over time? I think that the longitudinal plugin is what you are looking for. The parkinson's mouse tutorial(specifically the longitudinal section) might be a good example to work off of if you want a tutorial: longitudinal — QIIME 2 2022.2.0 documentation

Also we have youtube videos walking through that tutorial if that is helpful. Here is a video for more general background on longitudinal analysis and a video walking through the tutorial.

Hope that answer is alittle bit more on target for what you need! Let me know
:turtle:

1 Like

Thanks for your reply. The tutorial that you cited is what I am looking for!
I have one more question. In the tutorial, the outcome variable is beta diversity. How can I put beta diversity as predictor or independent variable in the model (mixed effect model)?

Thanks!

On top of q2-longitudinal which is a fantastic tool that comes pre-packaged with the core QIIME 2 install, there's also an additional plugin you can try for longitudinal microbiome data that I personally find super useful: q2-gemelli.

Thanks for your answer. I had a quick look at the link that you provided. In their results (below figure), beta diversity has not been considered as an independent variable since we do not have a p-value for it in the table. If we want to have beta diversity as predictor/independent variable in the model, what we should do?

Thanks!

1 Like

Hi @Barandouzi,
Yes, I didn't mean to suggest that the plugin would be answering your question about placing beta diversity as an independent variable, only that it is a very useful tool for longitudinal microbiome data analysis. The answer to your question is a bit more complicated.

The first thing is, from an experimental design perspective, it would be pretty hard to manipulate "beta diversity" in a way that it would be a true independent variable, which is probably why in most tools the expectation is that you are trying to look at changes in beta diversity in response to some other signal. Can you describe your specific design a bit for us, would certainly help us guide to what you can maybe do.

Beta diversity describes the (dis)similarity between 2 communities, so, unlike alpha-diversity it doesn't have a "within" sample definition. One possible option would be to take the values of Axis 1 or Axis 2 from your PCA/PCoA plot and treat those as "independent variables" scores, then you could add those to your regular linear regression formula. The limitation is that these would be based on reduced data and may not necessarily catch the signal you may be interested in.

1 Like

Thanks for your answer. I have a longitudinal design with three time points (same subjects) and I would like to know the relationship of beta diversity (independent variable) and education (independent variable) with fatigue (dependent variable) over time (fatigue = beta diversity + education). As you mentioned, it is easy to run this model (mixed effect model) for alpha diversity but not for beta diversity. I have also the same issue for taxonomy analysis. I appreciate any potential option/solution for longitudinal analysis when microbiome data (diversity and taxonomy) are independent variables.

Thanks

Hi @Barandouzi,
Thanks for the details, sounds like a neat study!
So, this is my personal take on your design, perhaps others want to disagree.
In this context, I don't think you need to include beta diversity as a covariable, primarily because you can't prove that it is a true independent variable. Fatigue on the other hand you can certainly control in real life. Fatigue and microbiome are very likely related in a bidirectional way too. If I were you I would phrase your question as beta diversity ~ fatigue + education, and perhaps subject as a random effect? You still can isolate the association between microbiome and fatigue while accounting for education which would answer initial question I believe.

Thanks for your response. Yes, that would be another approach.

I have another question. If we want to select some bacteria from taxonomy and put in the model (e.g., Ruminococous), how we can get the value for that bacteria (frequency of the bacteria).

Thanks!

1 Like

Hi @Barandouzi,

This kind of really depends on how you've processed your data so far and what tool you want to use downstream. Are you planning on adding Ruminococous as a covariable with some other variables? Or are you interested in seeing what taxa in general are associated with your response variable? There are a ton of tools out there for doing this kind of differential abundance testing, several which have QIIME 2 plugins. I once put a bunch of these tools in a spreadsheet, this is by no means comprehensive but maybe of help.

One thing you should be aware of that with microbiome data it is not so straight forward to just grab the relative abundance value of one taxa and add it in your model. See discussion on this topic here, here, and here for example. You'll want to carefully pick an appropriate normalization step, or select an appropriate ratio of taxa.

Also, since you were asking specifically for data at the genus level. In QIIME 2, you'll first want to collapse your OTU/ASV table down to the genus level using the qiime taxa collapse action. This gives you a new feature-table that is now collapsed at the genus level. You can then use this table to extract the specific genus you are interested in. There are a lot of ways of extracting the specific genus, but that also depends on what plugin or platform you are planning on using. So if you can give us some more details about what you are trying to do we can help you get there. I personally would just export this table into R and do all my customization there but not sure what you will be doing specifically.

2 Likes

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