Computational deconvolution of gene expression in leukemic cell hierarchies

Leukemias and some solid tumors are organized as cell hierarchies, sustained by cancer stem cells. We developed a computational method to study gene expression cancer cell hierarchies. Unlike traditional approaches based on physical cell sorting, our method extracts cell type-specific gene expression signals from gene expression profiles of unsorted tumor cells by deconvolution. We apply our method in the context of acute myeloid leukemia, and recover markers for acute myeloid leukemia stem cells (AML-LSC).