TITLE: “Model driven design and analysis of quantitative phenotype screens”
ABSTRACT: Increasingly, CRISPR screens are coupled with flow cytometry (FACS) to sort cells and quantify the
impact of genetic perturbations on a continuous phenotype. While FACS provides much more
quantitative information than simple survival screens, it introduces a number of experimental and
statistical challenges. Furthermore, as experimentalists push these screens in limited primary cells
or in vivo, they lack principled guidelines on how experimental parameters including multiplicity of
infection, guide RNA coverage, or FACS bin cutoffs affect statistical power. We present models for
both experimental design and inference of gene regulation in these screens. We show that
commonly used parameters are far from optimal and screens can be performed with a 20 times
reduction in cells at comparable accuracy. Our inference procedure models biological replicates,
infers the latent protein distribution, and infers experimental parameters. Together these analyses
provide a holistic framework for designing and analyzing highly parallel FACS screens.
