Thank you very much for this fantastic mixOmics package. I have a new data set on phenotypic/ethological data from animals. The part of the measured variables are zero inflated count data. I think that this is not a good prerequisite for running a PCA. Do you have any suggestion?
hi @AEggert
Maybe have a look at the CLR transformation (with a 1 offset) as here: mixMC Preprocessing | mixOmics
We also remove variables with too many zeroes beforehand.
A PCA would not work well if you see that the screeplot (the barplot with the amount of explained variance) does not decrease (or sometimes increases!).
Kim-Anh