Hi mixOmics team,
I have a few questions regarding the mixOmics package (which I have found to have a lot of great resources - thank you!). I am using the package for PLS-DA analysis.
- I have data from multispectral imaging (for each y-variable (or image), I have 20 corresponding x-variables, a reflectance value for each of 20 wavelengths). Is this data suitable to use for this package? It doesn’t seem to be the usual data type (i.e. genetics data, etc.), but so far I seem to be getting reasonable results with this package. I assumed it does not really matter what the data is from, but I would like to check.
- Scaling - as part of my pre-processing step, I used robust scaling (scaling the training set by median and IQR, then scaling the test set by the median and IQR of the training set), since at some wavelengths my data tends to be skewed and I did not get good results when I tried Z-score normalisation or min-max normalisation. Is it then reasonable to turn-off scaling when I run the PLS-DA model? I have noticed in some answers this has not been recommended, but I’m not sure if previous scaling was assumed or not.
Thank you in advance!