bio-crispr-screens-copy-number-correction

First run a diagnostic computing Spearman correlation between gene-level LFC and copy number. If bias is detected, apply CRISPRcleanR for unsupervised pre-hoc correction without a CN profile, or Chronos with alternate_CN when you have 3 or more cell lines with matched CN profiles.

Use CRISPRcleanR when you lack a matched copy-number profile or have fewer than 3 cell lines, since it works unsupervised from genomic position. Use Chronos for panels of 3 or more cell lines with CN profiles, especially DepMap-scale longitudinal data, where it jointly models population dynamics and CN effects.

alternate_CN fits the CN-to-gene-effect curve across a cell line panel, so it raises a RuntimeError below 3 lines. For one or two lines, correct the counts pre-hoc with CRISPRcleanR instead; Chronos training itself still works on a single line for gene-effect scores.

This is the classic copy-number artifact: Cas9 cuts at all amplified copies simultaneously, triggering DNA-damage response and G2 arrest that depletes sgRNAs regardless of true gene essentiality. Apply CN correction before hit calling and verify the post-correction Spearman rho is near zero.

Yes, switching from Cas9 knockout to CRISPRi or CRISPRa bypasses the artifact entirely because catalytically dead Cas9 creates no double-strand breaks. Base editing and prime editing also reduce the artifact, though CRISPRi knockdown is less complete than knockout.

Focal amplicons often contain only 3-10 genes, leaving the Mann-Whitney test underpowered even for real residual bias. Below 8 amplified genes, check the raw amplified-vs-diploid LFC gap and treat anything below -0.5 as suspicious regardless of p-value.