Collective Gene Expression Fluctuations Encode the Regulatory State of Cells

Gene expression is inherently stochastic, but how fluctuations propagate at the macroscopic scale of gene networks and whether they play a functional role remains unclear. Here, we develop a theoretical and statistical learning framework that shows how collective gene expression fluctuations encode the regulatory state of cells. Specifically, we map fluctuations in gene expression to a disordered, non-equilibrium system and derive the macroscopic phase behaviour of these fluctuations. Applying our theory to single-cell RNA sequencing data, we show that cells operate close to a dynamical transition separating strongly correlated fluctuations, reminiscent of glassy dynamics, from weakly correlated, disordered regimes. Drawing on these experiments, we adapt Restricted Boltzmann Machines to infer cell states by the structure of their collective gene expression fluctuations. This shows that strongly correlated fluctuations characterize cells with the ability to differentiate into more specialised cell types. Our results point to the emergence of collective fluctuations as a layer of regulation in cells and as a hallmark of high regulatory flexibility. They also demonstrate how physical principles can yield functional insights into cell states from widely used single-cell data.