I am trying find highly variable features (genes in my data) for downward consensus clustering analysis. I have used variation, median absolute deviation and coefficient of variation (CV) for finding top 1000 highly variable features (HVF).
I am getting expected clustering when I use CV for ranking than the other two measures. I wish to understand if CV is an appropriate measure for calculating HVF.
Also, the data is already normalized to log scale before I calculate means and variation. Shall I untransform it to find means and variation?