What would you recommend for Maaslin2 model validation?
Normally, one of the key steps is to plot the residuals against the fitted values and I guess one could do that for each of the ASVs, but Maaslin only returns the residuals and not the fitted values (which may have been transformed/normalized).
Hi @Ricardo_Ramiro - you can use the residuals to make a per-feature QQ plot as a visual inspection (see, e.g.
http://www.r-tutor.com/elementary-statistics/simple-linear-regression/normal-probability-plot-residuals). Another way would be to randomly shuffle the dataset (possibly many times) and construct a similar QQ plot based on the shuffled p-values (which should closely follow a uniform distribution under the null). Hope this helps.
Thanks a lot for your reply. That helps and could be easily done. I guess I was more wondering about plots of fitted values against residuals, to check for problems in variance. Is there a way to also extract the fitted values (post-transformation/normalization)?
Hi, @Ricardo_Ramiro- I just pushed an update in GitHub which ensures that you get the fitted values along with the residuals as output.
Hi, @himel.mallick - just a follow-up to the above thread. I’ve just re-installed the R package and still do not seem to get the fitted values as an output? Just wondering if there is anything else I need to do, or specify to get this output. Thanks very much!
Hi @CHall - if you install the latest version from GitHub, you should be able to see
fitted.rds in the output folder. Let me know if you still don’t see it after re-installing.