I was wondering if the DepMap team can provide a version of CRISPR_gene_dependency.csv without copy number correction?
The RNAi co-dependencies (DEMETER2) match reasonably with our co-IP and proximity labeling data while the CRISPR data looks like…complete nonsense. Was wondering if the copy number correction by Chronos/CERES might’ve made things worse?
There is no integrated (model-level) uncorrected version of the data, but there unintegrated (screen-level) files at three different levels: naive (just averaging guides and replicates), uncorrected (Chronos with no CN correction) and regular gene effect (Chronos run on each of the Avana and KY libraries separately, with results concatenated). All these files start with “Screen”. You’ll need to use the CRISPRScreenMap and Model tables to map screen IDs to cell lines.
It is unlikely that the copy number correction is causing your problems. I would split apart the Avana and KY screens in ScreenGeneEffect first and see if your gene looks like what you expect in either of those individually - perhaps only one library has good reagents and the other one is washing them out in the combined data. You can do the same with NaiveGeneEffect to make sure it isn’t some issue with Chronos as well.
It’s also worth reading this preprint on the differences between RNAi and CRISPR: Partial gene suppression improves identification of cancer vulnerabilities when CRISPR-Cas9 knockout is pan-lethal | bioRxiv. Your gene might be a common essential when completely knocked out but selective under inhibition.
Feel free to contact us with whatever you discover! We’re always looking for specific examples to evaluate our data processing with.