Title: 

Inference and modeling for single-cell CRISPR screens

Abstract:

CRISPR is a genome engineering technology that has enabled scientists to precisely manipulate and perturb human genomes. Single-cell CRISPR screens combine genome engineering and single-cell sequencing to survey the effects of DNA alterations in individual cells. These experiments have generated substantial interest in recent years, promising advances such as gene editing to accelerate medical research.  However, single-cell CRISPR screens pose considerable statistical challenges, currently limiting the power and interpretability of conclusions drawn. These statistical challenges include high dimensionality and non-Gaussianity of the data, which arise even for two-sample contrasts. We develop methods for performing valid statistical tests in the single-cell data settings. In addition to developing reliable pairwise gene comparisons, we also investigate high-dimensional tests that compare the whole transcription profile simultaneously. In the statistical literature, most effort has been spent on enhancing the power of tests, but we find interpretability is an important component, which is especially crucial because single-cell sequencing involves hundreds or thousands of genes. Our methods provide the users with rigorous results and also a sparse estimate of how each gene contributes to a positive rejection.