Coordinated gene expression variability encodes the regulatory state of cells

Abstract

Gene expression is inherently stochastic, leading to substantial cell-to-cell variability in mRNA and protein abundances. Variability in the expression of individual genes has been associated both with impaired signal processing and with facilitation of stress responses and differentiation. Here, combining machine learning, theory, and analysis of scRNA-seq data across various organisms and tissues, we show that variability in gene expression can be coordinated cell-wide. We define a statistical score that quantifies this coordination in single-cell data and demonstrate that distinct coordination patterns reflect the regulatory state of cells. We further develop a physics-informed machine-learning framework that identifies and predicts such variability patterns. Coordinated gene-expression variability emerges as a hallmark of stem and progenitor cells and distinguishes intrinsic stochasticity from cell-population heterogeneity. Together, our results establish the structure of gene expression variability as a cellular-scale signature of cell identity and regulatory organization.