CellRegMap – a statistical framework for mapping context-specific regulatory variants using scRNA-seq

Single-cell RNA sequencing (scRNA-seq) enables characterizing the cellular heterogeneity in human tissues. Recent technological advances have enabled the first population-scale scRNA-seq studies in hundreds of individuals, allowing to assay genetic effects with single-cell resolution. However, existing strategies to analyze these data remain based on principles established for the genetic analysis of bulk RNA-seq. In particular, current methods depend on a priori definitions of discrete cell types, and hence cannot assess allelic effects across subtle cell types and cell states. To address this, researchers at the European Bioinformatics Institute have developed the Cell Regulatory Map (CellRegMap), a statistical framework to test for and quantify genetic effects on gene expression in individual cells. CellRegMap provides a principled approach to identify and characterize genotype-context interactions of known eQTL variants using scRNA-seq data. This model-based approach resolves allelic effects across cellular contexts of different granularity, including genetic effects specific to cell subtypes and continuous cell transitions. The researchers validated CellRegMap using simulated data and apply it to previously identified eQTL from two recent studies of differentiating iPSCs, where they uncovered hundreds of eQTL displaying heterogeneity of genetic effects across cellular contexts. Finally, they identify fine-grained genetic regulation in neuronal subtypes for eQTL that are colocalized with human disease variants.

Overview of CellRegMap

A, B. Established workflows based on principal component analysis or factor analysis applied to scRNA-seq can be used to both estimate cellular manifolds (A) and to uncover individual factors that capture different cellular contexts (B). In addition to capturing major cell types, these factors can also explain subtle subtypes, as well as cell-type independent variation, such as the cell cycle and other cell-intrinsic factors. These cellular contexts can represent both discrete and continuous cell-state transitions, including cellular differentiation. C. Illustration of a genotype–context (GxC) interaction where genetic effects are modulated by a cellular differentiation context. Established analysis strategies (left) typically require discretization into discrete cell clusters (here low, mid, high), whereas CellRegMap enables assaying allelic effects as a function of the continuous differentiation context (right). Top panel: cellular manifold with color denoting allelic effects, either estimated in discrete cell populations (left) or in continuous fashion using CellRegMap (right). Middle panel: Alternative representation of allelic effects for different genotype groups, again either considering a discrete (left) or continuous modeling approach (right). Bottom panel: Encoding of discrete cell types (left) and continuous gradients using a cellular context covariance matrix in CellRegMap (right). D. The CellRegMap model can be cast as a linear mixed model, where single-cell gene expression values of a given gene are modeled as a function of a persistent genetic effect, GxC interactions, additive effects of cellular context, relatedness and residual noise. GxC interactions are modeled by treating allelic effect size estimates in individual cells (βGxC) as random variable with prior covariance Σ (C). E. CellRegMap allows to test for heterogeneous genetic effects across cells due to GxC at a given locus for a given gene (testing βGxC=0 vs. βGxC≠0). Color denotes the estimated GxC interaction component of genetic effects in individual cells (βGxC).