CRISPR KO screens

Q&A
1

We did our first CRISPR Cas9 KO screen with 2 cancer cell lines of interest in the lab and wanted to compare our results to the ones available at DepMap.
I used the MAGeCK-VISPR tool to identify negatively and positively selected genes in our negative screens (looking for essential genes, i.e. negatively selected genes, but also genes which KO induces cell proliferation, for our cell lines).
I used the MLE approach implemented in MAGeCK-VISPR which estimates a beta score for each gene: a positive score indicates positive selection and a negative score indicates negative selection. A p-value and FDR are also provided in association with the beta score.

In DepMap release 24Q4, i can obtain:

  • CRISPRGeneEffect.csv: gene effect estimates for all models, integrated using Chronos. Copy number corrected, scaled, and screen quality corrected
  • CRISPRGeneDependency.csv: gene dependency probability estimates for all models in the integrated gene effect: which corresponds to the probability that a given gene effect represents a true effect, am I correct ?

Is that correct to compare the results of my screens to the gene effect estimates available in CRISPRGeneEffect.csv for our cell lines of interest ?
If i want to check the congruence of significant results between the screens, should i set a threshold on the p-values in CRISPRGeneDependency.csv ?
Is there any correction for multiple testing on the p-values ?

Many thanks for your feedback.
Best, Isabelle

2

I’ll see if I can find someone to comment on how comparable MAGeCK-VISPR’s beta score and Chronos’s Gene Effect are, but I can say that negative Chronos scores are also indicative of negative selection (which is to say, growth defect due to genetic loss).

Also, I want to point out that the “gene dependency probability estimates” that you find in CRISPRGeneDependency.csv are not really a p-value on the gene effect. The probability that is being computed is the probability that the observed gene effect for a given gene was sampled from the set of pan-essential genes as opposed to the non-essential genes.

It is actually computed from the gene effect and is not intended as a value used to determine significance but rather to allow people to binarize lines into “dependent” and “not dependent” while accounting for false discovery.

We offer both “Gene Effect” and “Probability of Dependency” so that people can choose which to use depending on their downstream analysis. If you are looking for a continuous value capturing the amount of killing due to knockout, for say a correlation analysis, “Gene Effect” is what you likely want.

However, if you want to define two populations to do something like a t-test comparing sensitive vs non-sensitive lines, we recommend thresholding “Probability of dependency” to define the two populations.

Thanks,
Phil