Post Tue Apr 28, 2020 9:17 pm

Quantile normalisation of batches


There is a need to render the batches of experiments more comparable in order to compile many experiments and to gain/maintain statistical power. As there are always unattended sources of variation, normalization between batches is needed. There are different ways to normalize batches. Two recent papers by Van Gassen S et al. and Schuyler RP et al. address this issue. Here I examine more closely the use of quantile that has been proposed in these two articles.

I bring here a better understanding of why the full range quantile normalization might turn out to be odd. Unless you inspect the quantile plots that are available in cytNorm, or you do a bi-parametric plot, you can not suspect that the normalization is either good or bad. The density plot of the marker being normalized will have the right shape, the one of the reference.

The HTML analysis could not be attached in this post, but is available on i-cyto@github

I hope that you will enjoy this personal view and that it will raises an interesting discussion.

Stay safe,