Computational correction of cell-specific gene-independent effects in CRISPR-Cas9 essentiality screens: REStricted CUbic SplinEs with Mixed Models (RESCUE-MM)

Abstract

Increased gene copy number has been associated with a greater antiproliferative response upon genome editing, regardless of the true essentiality profile of the targeted gene. Many methods have been developed to adjust for genomic copy number technical artifacts. Existing methods use a two-step correction by pre-processing the data prior to the final analysis. It has been shown that two-step corrections can produce unreliable results, due to the creation of a correlation structure in the corrected data. If this structure is unaccounted for, gene-essentiality levels can be inflated or underestimated, affecting the False Discovery Rate (FDR). We propose a one-step correction using restricted cubic splines (RCS) to be a simpler alternative which reduces the bias in downstream analyses. Moreover, most existing methods combine guide-level results to yield gene-level estimates which can misrepresent the true gene essentiality profile depending on the guide-averaging method. Our model-based approach (RESCUE-MM) for copy number correction provides a more flexible framework that allows for guide-level essentiality estimation while accommodating more complex designs with grouped data. We provide comparisons to existing copy number correction methods and suggest how to include copy number adjustment in a one-step correction fashion in multiple experimental designs.