Nabo - a framework to define leukemia-initiating cells and differentiation in single-cell RNA-sequencing data

Single-cell transcriptomics facilitates innovative approaches to define and identify cell types within tissues and cell populations. An emerging interest in the cancer field is to assess the heterogeneity of transformed cells, including the identification of tumor-initiating cells based on similarities to their normal counterparts. However, such cell mapping is often confounded by the large effects on total gene expression programs introduced by strong perturbations such as an oncogenic event. Here, we present Nabo, a novel computational method that allows mapping of cells from one population to the most similar cells in a reference population, independently of confounding changes to gene expression programs initiated by perturbation. We validated this method on multiple datasets from different sources and platforms and show that Nabo achieves higher rates of accuracy than conventional classification methods. Nabo is available as an integrated toolkit for preprocessing, cell mapping, differential gene expression identification, and visualization of single-cell RNA-Seq data. For exploratory studies, Nabo includes methods to help evaluate the reliability of cell mapping results. We applied Nabo on droplet-based single-cell RNA-Seq data of healthy and oncogene-induced (MLL-ENL) hematopoietic progenitor cells (GMLPs) differentiating in vitro. Despite a substantial cellular heterogeneity resulting from differentiation of GMLPs and the large transcriptional effects induced by the fusion oncogene, Nabo could pinpoint the specific cell stage where differentiation arrest occurs, which included an immunophenotypic definition of the tumor-initiating population. Thus, Nabo allows for relevant comparison between target and control cells, without being confounded by differences in population heterogeneity.