Current leukemia therapies target cancer cells with specific phenotypes or genotypes, but this assumes that either genomic mutations or immunophenotypes alone serve as faithful proxies for treatment response1. Moreover, the heterogeneity inherent to all cancers, including leukemias, makes direct mapping of genotype-phenotype relationships challenging2, 3. Here, we present a method to genotype and phenotype single cells at high throughput, allowing direct characterization of proteogenomic states on tens of thousands of cancer cells rapidly and cost efficiently. Using the approach, we analyze the disease of three leukemia patients over multiple treatment timepoints and recurrences. We observe complex genotype-phenotype dynamics and extensive decoupling of the relationships over disease progression and response to therapy, illustrating the subtlety of the disease process and the inability to use genotypes as direct proxies for phenotypes. Our technology has enabled the first rigorous test of the prevailing paradigm that treatment of a disease phenotype is equivalent to treatment of its underlying genotype. More broadly, our results highlight the power of single-cell multiomic measurements to resolve complex biology in heterogeneous populations, and illustrate how this information can be used to inform treatment. We thus expect that our methodology will find broad application to study proteogenomic tumor landscapes across cancers and will support the next generation of immunotherapy.