what statisical test to use and qiime has it?

I’m studying if diet influences the oral microbiome and for that i have to compare a lot of variables (proteins, carbs, vitamins, minerals, etc) with the microbiome. Anyone knows if qiime2 has some statistical test that allows me to do that comparisson?

Thank you

Hello Ana,

qiime diversity beta-group-significance

There are some great examples inside the moving pictures tutorial, which you might have found. Do you think these tests are a good fit for your data?

It sounds like you might be interested in a more detailed discussion of these methods, as you have many variables to compare. The Parkinson’s Mouse Tutorial just dropped and it includes lots of metadata and shows how to work with it.


Thank you for your reply. Yes, i am following Parkinson’s Mouse Tutorial, however if i am not mistaken, we have to input one variable at the time in beta diversity, i am right? And i have about almost 50 variables.

1 Like

And what i want to analyse is not only diversity, it is for example if some species or genus are related for example with high sugar intake.

(I am sorry, i just remembered to write this after my last reply to you)

1 Like

Check out mmvec, which has a QIIME 2 plugin included in the standard install. 50 variables is a lot! This was written with microbiome-metabolome associations in mind but works for associations between any high-dimensional data it sounds like 50 nutritional variables is pretty similar to metabolome data anyway so this may be an appropriate test.

You could use qiime diversity bioenv to find the top variables associated with beta diversity, then test associations of those multiple variables with beta diversity using qiime diversity adonis and with alpha diversity using qiime longitudinal anova.

I hope that helps!

1 Like

Thank you, i am going to try to do that !!

Hello Ana,

Wow, 50 is a lot of internal variables!

One of the big challenges with working with so many variables is that many will covery with each other. So if unsaturated fat and saturated fat are correlated, then any change in microbiome will be associated with both these variables, and you can’t figure out which more is most important!

So in this context, a first goal could be to find the 10 most important / interesting variables, then zoom in on just those 10. Like this :point_down:

Let us know what you find!
(We should probably have a tutorial about working with 10+ metadata variables. I know this problem came up for the EMP so people here have encountered it before.)



Another common beta diversity approach right is to look at the Adonis statistic and rank covariates based on that. So, yes, univariate, but also, another way to rank both categorical and continuous variables if you chose to handle them. I think this - or something similar - showed up in Flemish Gut, and definitely in Ye et al. And then, adonis is multivariate (I think that’s in the PD mice example)… just make sure that you pay attention to the order of your variables in the model!

I’ve also done step wise alpha diversity testing where I determine which covariates are significantly associated and then do forward selection on multivariate models to figure out whats associated (I usually do mine outside of QIIME, TBH, because its faster).

But, I agree with @colinbrislawn that limiting your covariates would be a good idea; you’ll get lots of data that will co-vary and it wont matter so much in univariate models but it will make your multivariate life hell.

Finally, other thing you could maybe try is to do some sort of distance transform on your diet, and then see if there’s some kind of relationship based on big patterns.



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