Choice of count normalisation

[ilevantis]

Hi and thank you for producing this package and the associated paper. I am interested in your idea of using the Beta-Binominal distribution for modelling CRISPR-screen count data and your comparisons with the MAGeCK package. I was wondering why the choice was made in CB2 to only normalise by total library size rather than by a normalisation method that attempts to handle the imbalance caused by highly abundant sequences.

As I understand it, MAGeCK uses a median ratio normalisation (function is very similar to that carried out by DESeq2) for counts before evaluating fold changes, does it make sense to also use a median ratio normalisation before running the measure_sgrna_stats function from CB2 or shoud we stick to using only the total library size normalisation (implented as get_CPM in CB2)?