Abstract
Pooled CRISPR screens with single-cell RNA sequencing readout (Perturb-seq) have emerged as a key technique to determine the functionality of a gene by directly perturbing the DNA of the gene. One of the most intriguing recent problems is quantifying the similarity between CRISPR perturbations, for example, whether they upregulate the same set of downstream genes. In this context, genetic convergence refers to the phenomenon where CRISPR disruptions of different genes lead to a similar downstream outcome. Existing methods are mostly heuristic. We present XConTest, a two-step, cross-validated procedure for assessing the genetic convergence problem. The test statistics calculated from that procedure are approximately standard normal when the two perturbations have an orthogonal influence on the cell expression profile. We apply XConTest to two studies: an investigation of the common impact of a suite of autism genes, and a large-scale study of genes associated with immune response to determine sets of genes with common functionality.