Number of variables in final sPLS-DA

I have performed and tuned a sPLS-DA, and the recommende number of components is 2.
I have also tested how many features to use, and for one component it is 180, for 1 to 2 it is 160.
When I selectVar(), the features are different, exept for 19 features which is in both feature groups.
I started out with about 1200 features and 60 subjects.

My question is: When the final sPLS-DA has been built, it seems like this is based on the 180+160.
Is this correct? I have set it to use 2 components as recommended.

Regards, Live

hi @Live,

The results from the tune.splsda() are recommendations. You then set up those parameters yourself in your final sPLS-DA model (as we show in our examples on our website). Just make sure at this stage that you have input the argument keepX = c(180, 160).

You should expect a low overlap of common features between the components, as they are supposed to be orthogonal. The overlap may still happen if the feature selection size is a bit large (as it is the case for you in your context).