PROJECT SUMMARY/ABSTRACT Cancer is a genetic disease, and the set of mutations in a tumor affects both its behavior and its response to therapies. Large sequencing initiatives have produced catalogs of gene variants arising in different cancers. Substantial challenges remain, however, in interpreting their effects. First, even when variants affect the same gene, their molecular phenotypes may be distinct. Second, many variants are common enough that they have been identified, but still sufficiently rare that no targeted studies have characterized them. Finally, cancer in general arises from cooperation among multiple mutations, so the function of a variant in one context—cell type, genetic background, or environment—may only partly inform its behavior in another. The sheer number of possible variants and contexts argues for taking a systematic approach to phenotyping. Here, we present BEAT-seq (Base Editing Allele Transcriptome sequencing), a flexible, scalable, and robust approach for engineering cancer-associated variants by CRISPR-mediated base editing and measuring the resulting effects on cellular phenotype by single-cell RNA sequencing. Robustness follows from our development of a sensor assay that can quantify the base editing efficiency of many sgRNAs in parallel, enabling us to identify those that reliably introduce cancer variants. We then exploit an improved Perturb-seq protocol, enabling us to introduce libraries of variants in pooled format and simultaneously capture both the sgRNAs, encoding the programmed edits, and single-cell transcriptomes, carrying their phenotypic consequences. In Aim 1, we credential BEAT-seq by generating validated sgRNAs targeting common cancer variants. We profile the effects of these variants across different epithelial cell types—pancreatic and lung—and across different genetic backgrounds to study the role of context. Finally, we explore whether BEAT-seq can assign function by constructing a library targeting ~500 somatic and germline variants of unknown significance identified through MSK-IMPACT sequencing. These tasks grow gradually in analytical complexity. In Aim 2, we establish rigorous statistical pipelines for the interpretation of single-cell functional genomics experiments. We show that the orthogonal characterization from the sensor assay enables a Bayesian approach to identify edited and unedited cells, addressing a central challenge that affects many single-cell screens. We then develop a data normalization procedure for representing perturbations’ effects in relative terms, enabling comparisons to be made across contexts. Finally, in Aim 3 we conduct in vivo BEAT-seq experiments profiling cells carrying dozens of p53 variants introduced by orthotopic transplantation into mouse pancreases. This work enables parallelized characterization of cancer variants on a scale not previously feasible. Our results will provide insight into how variants affect tumor phenotype in different contexts, illum...