Two Factor Multi-Level sPLS-DA Analysis


I am trying to conduct a multi-level sPLS-DA analysis to look at effects of group and time. I was looking on this site to try and go through a tutorial but I don’t think all of it there for multi-level? I am interested in determining the effect of time (2 time points, pre/post) on 2 groups. This is similar to what was done in one of your publications with the HIV data. Picture attached. Is there a site that can guide me through this process?

Thank you!

hi @ravibot,

The data we used are here: ?vac18.simulated then we give an example (not run) on a two factor analysis in ?splsda.


Thank you so much! I found it! :relaxed:

Hi Dr. Le Cao,

I wanted to follow up on my previous question. Using multi-level how is it possible to look at the specific effects of time? Or is it only possible to take into account the effect of time in multi-level?

I have two time-points, and at each time point, a subject can either have the condition in question or not. So at baseline, I have two groups (control vs. condition) as well as follow-up (control vs. condition). But it is possible that the subject may not longer have the condition at follow up, or may develop that condition. I am interested about the effect of time as well the effect of group here.

I also have the same, one 'omics datatype at each time point.

Thank you!

hi @ravibot,

In your study context, you may have two choices:

  • a one level multilevel analysis, where the effect of group would be taken into account in the Y = group in PLS-DA, whereas the effect of time is taken into account in the multilevel = subjectID. Here the aim is primarily to discriminate the two groups while assuming individuals are measured at two different time points (a bit like what you would do with a paired t-test)
  • a two level analysis group x time where you are interested in disentangling both effects.

For the latter, it would be difficult to assess both effects numerically. The sample plot will give you an indication of what is going on but that is about it. You may have to look in more depth in the variables that are selected and assess using univariate tests whether they have both a time and group effect. In that case you would use a multilevel analysis as an explorative / illustration tool and the univariate tests to assess statistical significance.