Clustering and regression analysis using q2-gneiss

User Support
, ,
14

Hi @mortonjt,

Thank you for your previous replies.

Would you mind commenting on my previous question about whether using ols modelling is appropriate with longitudinal data.

Although I do want to add host_subject_id as a random effect, I did run a linear regression on the data just as an initial investigation. This allowed me to evaluate the effect of each covariate by looking at R2diff, and I found that as expected, timepoint and host_subject_id explained the most variance in the balances. Is this still a valid approach to understand the contribution of each individual covariate, or does this not make sense as it’s a longitudinal study with random effects?

Also, when I get the output of running gneiss lme-regression with the following

–p-formula diet+timepoint

–p-groups host_subject_id

I don't really understand how the continuous and categorical covariates put into "--p-formula" interact, and whether this formula will actually identify significant balances that are effected by diet over time, or just those effected by either diet1, diet2, or individual timepoints. For example when I look at the coefficient and p-value output csv files (from lme-regression) like you suggested, each covariate (diet2, timepoint2, timepoint3, timepoint4) ect.. has separate coefficient and p-values. Does that mean each covariate is being treated separately, rather than testing for any interactions?

One last thing! I came across a post talking about "Correct p-values" when looking at output of gneiss regression (Using ANCOM and gneiss with one categorical variable - #7 by mortonjt).

  • Careful - you’ll want to only consider those pvalues under Corrected Pvalues. And pvalues close to 0.05 will also warrant skepticism, so make sure to carefully sanity check the boxplots / scatter plots in the balance-taxonomy command.

I don't see a "Corrected Pvalues" column, therefore I just wanted to check that the pvalues I download from "Coeficcient pvalues" from the regression output are in fact already corrected?

Thanks again and sorry for my confusion!
Jen