Log-Transformation - Erro


I was trying to do Log-transformation followed the link http://mixomics.org/mixmc/mixmc-pre-processing/ for so to continue analysis the PCA.

But return the error :
Error in clr.transfo(X, offset = offset) : make sure you use pseudo counts before normalisation to avoid 0 values with log ratio transformation
I need to transform the log always? Or I can to do the others steps? The error can be the type of the transformation?
Until step previous for to check heterogenity library size it worked.

Thank you so much

hi @aoliveira,

The clr.transfo() performs a centered log ratio transformation, so there is indeed a log involved as part of the process (not negotiable with CLR). All you need to do, as described in the tutorial is to set an offset = 1 in that function to resolve the problem (that states you have zero values currently). It will add a count of one to ALL the values in X.


Hi @kimanh.lecao

Your tip helped me and worked.
I was still in doubt, does this mean that in my data it had “zero” and the offset = 1 adds this value so the matrix does not contain values less than 1?


Yes this is correct @aoliveira, consider this as a global shift of unit 1 of the entire data matrix.

Thank you for support!