Does too much cell coverage and read depth confound screen results?

[Nicki Leung]

Hi all,

I have been doing genome-scale positive selection screens with the Brie library and everything looked good. I then made a custom library for validation screen with much less sgRNAs (~4800 vs. 78000 in Brie) and repeat the screening under the exact same condition, without scaling down the cell overage and NGS read depth. This time I had quite a few problem:

1) Very low LFC (median/mean) => In the genome-scale screen I often see LFC of >5 or even >10 for strongest hits, but the validation library all hits had a LFC of <1

2) Perfectly matched guides ~50-60% => In the genome-scale screen the perfectly matched guides were always >80%

3) Very low Gini index => mageck count returned a gini index of 0.01 for the input and 0.05 for the positively selected population. I usually got >0.3 in the genome-scale screen

4) Reads dominated by a few guides => In the positively selected population (~5% of all), I still managed to detect the presence of all 4800 guides but there are 2 guides having >50-fold read counts than all other guides. Interestingly, these 2 guides are targeting the same gene but I don’t think it has any physiological relevance to the selection criteria.

I am thinking whether these problems are due to having a much reduced library size but we didn’t scale down accordingly. Any thoughts? Thanks in advance!