The thing is to count the residues in the TM-helices. You can then take the counts in the helices, divided by the counts in the whole dataset, and take the logarithm. That will give you twenty preference parameters.