keepX for sPLS-DA - small sample size


I have a proteomics dataset (1400 proteins) with 6 classes. I am trying to do sPLS-DA but I only have two samples per class. I’m trying to understand how to select an appropriate ncomp and keepX. I use to ‘loo’ method for perf() and tune() but choice.ncomp is null (looking at the plots I think 3) and choice.keepX suggested 1 or 2 features per component which seems too low. I don’t intend to use this predictively. Can anyone give suggestions?

Thank you!

hi @emmawh,
Given the low number of samples per group I would not bother :slight_smile: You saw that the tuning function is limited for such scenario.

Simply to decide how many components and keepX yourself. Inspect a few plotIndiv() graphs to see whether your groups are separated as you expect (it could also be the case that your samples are difficult to separate). I think you will need more than 10 proteins selected. If if appears from these plots that some groups are similar, you could consider merging them in the same group.