Linking genotypes with multiple phenotypes in single-cell CRISPR screens

Abstract

CRISPR/Cas9 based functional screening coupled with single-cell RNA-seq ("single-cell CRISPR screening") unravels gene regulatory networks and enhancer-gene regulations in a large scale. We propose scMAGeCK, a computational framework to systematically identify genes and non-coding elements associated with multiple expression-based phenotypes in single-cell CRISPR screening. scMAGeCK identified genes and enhancers that modulate the expression of a known proliferation marker, MKI67 (Ki-67), a result that resembles the outcome of proliferation-linked CRISPR screening. We further performed single-cell CRISPR screening on mouse embryonic stem cells (mESC), and identified key genes associated with different pluripotency states. scMAGeCK enabled an unbiased construction of genotypephenotype network, where multiple phenotypes can be regulated by different gene perturbations. Finally, we studied key factors that improve the statistical power of single-cell CRISPR screens, including target gene expression and the number of guide RNAs (gRNAs) per cell. Collectively, scMAGeCK is a novel and effective computational tool to study genotypephenotype relationships at a single-cell level.

Conditional resampling improves calibration and sensitivity in single-cell CRISPR screen analysis

“…We also simulated negative control gRNA data using a logistic regression model with the same covariates as the gene expression model. We assessed the calibration of three methods across the four simulated datasets: SCEPTRE, improved negative binomial regression, and scMAGeCK-LR 10 , a recently-proposed, permutation-based nonparametric method. To assess the impact of model misspecification on SCEPTRE and the improved negative binomial method (on which SCEPTRE relies), we fixed the dispersion of the negative binomial method to 1 across all four simulated datasets.…”

“…We applied SCEPTRE, negative binomial regression, and scMAGeCK-LR 10 to the four problem settings, each with n sim = 500 repetitions. The negative binomial method, and in turn SCEPTRE, was based on the z statistic from the Hafemeister-inspired negative bi-nomial model (3) with α = 1. scMAGeCK-LR differs from SCEPTRE and the negative binomial method in that scMAGeCK-LR computes p -values for all enhancers simultaneously.…”

“…Similarly, Xie et al 8 found an excess of false positive hits in their rank-based Virtual FACS analysis. Finally, Yang et al 10 found that their permutation-based scMAGeCK-RRA method deems almost all gene-enhancer pairs significant in a reanalysis of the Gasperini et al data. These works propose ad hoc fixes to improve calibration, but these adjustments are not sufficient to ensure reliability of the results.…”