Bug with PLS-DA multilevel: example code not working

Support
1

I am trying to follow your example from Case Study of Multilevel sPLS-DA with Vac18 dataset Multilevel Vac18 Case Study | mixOmics. I am running the code given in the example, but I get this error in R once I run the splsda():

undergo sPLS-DA after parameter tuning

final.splsda.multilevel.vac18 ← splsda(X, Y, ncomp = optimal.ncomp,

  •                                     keepX = optimal.keepX,

  •                                     multilevel = design)

Error in rowSums(Y) : ‘x’ must be an array of at least two dimensions.

dim(X)
[1] 42 1000

Why am I getting this error using your example code and how do I fix it?

2

hi @erinc,

There has just been an update in Bioconductor this week and I think this has broken a few things :sob:. I’ll see if I can find a quick fix.

Kim-Anh

3

hi @erinc,

Here is a quick fix, but it comes with limitations. You can use the withinVariation() function externally to PLS-DA to extract the X data matrix that accounts for a multilevel design. The downside is that the tuning / perf functions for a PLSDA multilevel won’t work with that trick (as we remove unique individuals during the cross-validation process in that case). I’m not sure when we will be able to resolve that bug for time constraint reasons!

Here is the code:

data(vac18) # extract the vac18 data
# using withinVariation, externally
design <- data.frame(sample = vac18$sample) # set the multilevel design
X.multilevel <- withinVariation(vac18$genes, design = design)
dim(X.multilevel)
Y <- vac18$stimulation
optimal.ncomp = 4
optimal.keepX = c(20,70,50,80)
# here, no need to specify 'design' as the multilevel decomposition has already happened in withinVariation 
final.splsda.multilevel.vac18 <- splsda(X.multilevel, Y, ncomp = optimal.ncomp,
                                        keepX = optimal.keepX)

Explanation of the bug (for future developers): What is happening with this bug is that somehow the Y matrix during the multilevel decomposition is not a matrix anymore.

Kim-Anh

4

Thank you for the code. It is now working! Does mixOmics allow for covariates in the splsda() function? I have searched your website but is says no results were found.

5

hi @erinc,

We are thinking about adding covariates in PLS but we haven’t done anything yet.

Kim-Anh

7

Hello @kimanh.lecao ! I am also interested in adding an aspect of my metadata as a covariate into my PLS2 model (similar structure as liver toxicity study). Is there an update for this yet in mixOmics? If not, @erinc how did you include covariates in your analysis?

8

@hardytan,

No update on our end, it’s a hard problem and we are applying first for funding to develop the method! See some previous posts about this, we advise to correct for the covariates effects first (under the keywords ‘remove batch effects’ ‘remove covariate effects’.

Kim-Anh