PLS with Ordinal Y

1

Dear mixOmics team,

I am analysing data that were obtained from groups of samples that have meaningful order (i.e. disease severity). I thought of PLS-DA, but then it would be one-vs-all comparisons, which I was not investigating for. Would it make sense to set my group label as numbers (i.e. 1, 2, 3, 4, and so on) in the Y matrix? What is the caveat that I have to keep in mind if I run the analysis in this manner?

Thank you for your time.

Best regards,
Mikhael

2

hi @mdmanurung,

PLS-DA does not really perform a one-vs-all, but tries separate all classes, sometimes simultaneously (same component), sometimes not.
If you label the response as ordinal, then the caveat is that it assumes an additive type model (i.e. category 4 = 2 x category 2), so depending on your labelling, it might be OK with a PLS regression.

Kim-Anh

1 Like