Every cell in the body carries essentially the same genome, yet liver cells behave nothing like neurons, and even genetically identical cells growing side by side can differ dramatically in the molecules they produce. A major source of this individuality lies in the peculiar rhythm of gene expression itself. Rather than manufacturing messenger RNA in a smooth, continuous stream, genes tend to flicker on and off, releasing mRNA in short, intense episodes known as transcriptional bursts. These bursts are inherently stochastic, and their statistics—the frequency with which a gene fires and the amount of RNA produced per burst—shape how noisy or stable a gene’s output is across a population of cells. For decades, scientists have studied bursting one gene at a time, painstakingly fitting mathematical models to data from individual loci. What has remained stubbornly out of reach is a genome-wide account of how the web of regulatory interactions between genes—the architecture of transcription factors acting on their targets—collectively governs these burst dynamics.
That gap has now been addressed by a team of researchers led by Jiajun Zhang of Sun Yat-sen University, together with colleagues at the University of California Irvine and Guangdong University of Technology. Writing in Molecular Systems Biology, the team introduces BurstLink, a statistical-mechanistic framework designed to simultaneously infer both the gene–gene regulatory interactions and the transcriptional bursting kinetics of thousands of genes directly from single-cell RNA sequencing data. The work, published as an open-access article, tackles a deceptively hard problem: single-cell data are static snapshots, yet bursting is fundamentally dynamic. The researchers reasoned that a cell population, sampled at a single instant, still encodes information about the underlying dynamical process, because the observed distribution of mRNA counts reflects the history of switching between active and inactive gene states.
At the heart of BurstLink lies an interpretable probabilistic model. Each gene is described by a Poisson–Beta distribution, the well-known steady-state solution of the classic telegraph model of bursting, in which a promoter stochastically switches between an inactive and an active state. The parameters of this distribution correspond directly to biological quantities: the activation and deactivation rates of the gene and its transcription rate. From these, the framework derives burst frequency—the number of bursting episodes per unit time—and burst size, the mean number of mRNA molecules produced per burst, along with a measure of gene-expression variability called the squared coefficient of variation. To capture relationships between genes, the authors embed these marginal distributions within a bivariate Poisson–Beta model based on the Sarmanov–Lee construction, a copula-like approach in which a single coupling parameter determines whether two genes are positively regulated, negatively regulated, or independent of one another.
A crucial concern for any such framework is whether a tractable statistical model faithfully represents the messier, more realistic dynamical system it claims to approximate. To address this, the team built an explicit stochastic model of two mutually interacting genes based on the genetic toggle switch circuit, complete with Hill-function regulation, protein-mediated feedback, and promoter switching. They simulated this dynamical system exhaustively using the Gillespie stochastic simulation algorithm and then asked whether BurstLink, fed only with the resulting synthetic count data, could recover the underlying behavior. The answer was emphatically yes. Across co-expression landscapes ranging from unimodal to quadruple-modal distributions, the generalized Kolmogorov–Smirnov tests showed close agreement between the statistical model and the dynamical ground truth, and the inferred statistical parameters correlated strongly with the true switching and synthesis rates, with Pearson correlations approaching 0.99.
Scalability was the second major hurdle. Inferring parameters for every pair of candidate genes in a genome is a daunting optimization problem, because each gene’s parameters influence the likelihood of many edges simultaneously. The researchers solved this by reformulating the inference as a distributed optimization amenable to the alternating direction method of multipliers, or ADMM, a technique that decouples the problem into edge-wise updates that can be computed in parallel, together with node-level consensus variables that keep gene-specific parameters consistent across the network. Numerical evaluation of the Poisson–Beta likelihood, which would otherwise be prohibitively slow, is accelerated using Gauss–Jacobi quadrature. The result is a pipeline that remains computationally feasible at genome scale, implemented in a user-friendly Python package with documentation available online.
Validation on synthetic data confirmed that the method recovers regulation types, regulation strengths, and burst kinetics accurately under positive, negative, and absent regulation, and that its performance degrades gracefully as cell numbers drop or dropout rates rise. The framework’s information-theoretic measure of regulatory strength, called reweighted mutual information, proved notably more sensitive than conventional correlation coefficients or normalized mutual information, which frequently failed to distinguish regulated from unregulated gene pairs in the same synthetic benchmarks. When benchmarked against established gene regulatory network inference methods, including PIDC, GENIE3, GRNBoost2, and SCENIC, BurstLink matched state-of-the-art performance on network reconstruction while offering something none of the competitors provide: a joint estimate of bursting kinetics and regulatory dynamics within a single mechanistic framework.
Applied to single-cell data from mouse embryonic fibroblasts, with chromatin accessibility data used to pre-screen plausible transcription factor–target pairs, BurstLink yielded a genome-wide regulatory network spanning 4,173 genes with valid inferred burst parameters. The analysis surfaced several striking regularities. Target genes, which sit downstream in the regulatory hierarchy, exhibited significantly higher burst frequency and greater expression variability than the transcription factors regulating them—a genome-wide signature of noise propagation, in which fluctuations in a regulator ripple into its targets. The team also found that the strength of transcription factor binding, captured by the equilibrium binding constant inferred from the switching rates, shapes bursting in a characteristic way: stronger binding affinity was associated with lower burst frequency but larger burst size, meaning that tightly bound factors appear to prolong individual burst episodes rather than trigger them more often.
Perhaps the most consequential finding concerns how the sign of regulation alters burst dynamics. Across both low- and high-expression gene groups, positive regulation was consistently associated with higher burst frequency, constrained burst size, and elevated gene-expression variability, while negative regulation produced the opposite pattern. These effects were confirmed at two levels: macroscopically, using Bayesian ridge regression to predict burst kinetics as the proportion of positive or negative regulatory loops in the network was systematically varied, and microscopically, by comparing genes according to their net regulatory input computed under a mean-field approximation. Applied to mouse embryonic stem cells, the framework found that burst frequencies and sizes correlate positively between the two cell types for shared genes, and that most genes retain the same regulatory loop type across systems, hinting at a conserved genome-wide regulatory grammar.
The framework also proved capable of revealing how perturbations remodel the regulatory landscape. When the researchers applied BurstLink to mouse embryonic stem cells treated with the DNA-damaging agent IdU and compared the results with vehicle-treated controls, they found that the drug reduced burst frequency, increased burst size, and raised expression variability across hundreds of genes without shifting average expression levels—consistent with earlier experimental reports that DNA damage modulates transcriptional noise. Differences in the regulating activity of transcription factors and the regulated status of their targets, quantified through network in- and out-degrees, pointed to altered DNA-templated transcriptional programs, suggesting that the drug reshapes cell fate decisions partly by reorganizing the burst architecture of the regulatory network itself.
The authors are candid about the limitations of their approach. The framework operates on mature mRNA counts, whereas regulation is ultimately exerted at the protein level, and the lag introduced by slower protein turnover attenuates inferred coupling strengths, although simulations indicate the qualitative regulatory type and direction are still recovered. Directional inference from snapshot data yields putative, rather than interventional, causality, and extending the bivariate model to genuinely multivariate network-wide inference remains an open theoretical challenge. Still, by uniting mechanistic interpretability with genome-scale tractability, BurstLink offers biologists a new lens on the stochastic engine of gene expression. As single-cell multi-omics technologies mature—adding nascent RNA measurements, chromatin conformation capture, and protein-level readouts—frameworks of this kind are poised to turn the flickering of individual genes into a coherent picture of how regulatory networks write the fates of cells.
The conceptual foundation for this work traces back to the telegraph model, first formulated in the 1990s to describe a promoter stochastically toggling between inactive and active states. Its steady-state solution, a Poisson–Beta distribution, has since become the workhorse for estimating burst kinetics from single-cell data, most prominently in genome-wide analyses showing that burst frequency and burst size vary systematically across mammalian genes. Earlier efforts also revealed that auto-regulatory feedback distinctly modulates the two kinetic parameters, foreshadowing the idea that network context matters. What distinguishes BurstLink is the extension of this single-gene machinery to pairs of genes, so that regulatory direction and sign emerge from the same likelihood that governs bursting.
The biological findings align with a broader literature on noise propagation in genetic circuits. Theoretical and experimental studies have long shown that stochastic fluctuations in a transcription factor’s abundance can be transmitted to its targets, amplifying heterogeneity through regulatory loops. The observation that downstream targets display higher burst frequency and variability than their upstream regulators provides a genome-wide, quantitative confirmation of this principle. Similarly, the inverse relationship between binding affinity and burst frequency echoes mechanistic expectations from promoter kinetics, where stable factor occupancy tends to sustain longer active episodes rather than trigger switching more often.
Practically, the framework’s explicit handling of technical noise and cell-size variation addresses well-known confounders in single-cell RNA sequencing, where differences in sequencing depth and molecule capture can otherwise masquerade as biological variability. By jointly modeling these artifacts, BurstLink reduces the risk that inferred regulatory edges reflect measurement artifacts rather than genuine coupling. The open-source implementation, together with its compatibility with chromatin accessibility pre-screening, should make the approach accessible to laboratories seeking to move beyond correlation-based network reconstruction toward mechanistic, burst-aware models of gene regulation.
Subject of Research: Genome-wide inference of gene–gene regulatory interactions and transcriptional bursting kinetics from single-cell RNA sequencing data
Article Title: Deciphering global transcriptional dynamics coordinated by gene-gene regulatory interactions using single-cell data
Article References: Zhou, L., Luo, S., Huang, Z., Zhang, Z., Wang, Z., & Zhang, J. (2026). Deciphering global transcriptional dynamics coordinated by gene-gene regulatory interactions using single-cell data. Molecular Systems Biology. https://doi.org/10.1038/s44320-026-00235-4
Image Credits: AI Generated
DOI: 10.1038/s44320-026-00235-4
Keywords: transcriptional bursting, gene regulatory networks, single-cell RNA sequencing, BurstLink, burst frequency, burst size, gene expression variability, mouse embryonic fibroblasts, statistical mechanistic modeling, transcription factors, noise propagation, DNA damage response
Cite Scienmag News
APA MLA Chicago
Juliet Wilcox. (September 11, 2026). New single-cell framework reveals how gene networks orchestrate transcriptional bursts. Scienmag. https://scienmag.com/new-single-cell-framework-reveals-how-gene-networks-orchestrate-transcriptional-bursts/
Juliet Wilcox. “New single-cell framework reveals how gene networks orchestrate transcriptional bursts.” Scienmag, 11 September 2026, https://scienmag.com/new-single-cell-framework-reveals-how-gene-networks-orchestrate-transcriptional-bursts/. Accessed 11 September 2026.
Juliet Wilcox. “New single-cell framework reveals how gene networks orchestrate transcriptional bursts.” Scienmag. September 11, 2026. https://scienmag.com/new-single-cell-framework-reveals-how-gene-networks-orchestrate-transcriptional-bursts/
Copy citation Download RIS
Tags: burst frequencyburst sizeBurstLinkcell-to-cell gene expression differencesDNA damage responsegene expression variabilitygene network architecturegene regulatory networksgenome-wide transcription regulationmolecular mechanisms of gene expression noisemouse embryonic fibroblastsnoise propagationsingle-cell gene expression analysisSingle-Cell RNA Sequencingstatistical mechanistic modelingstochastic gene expression modelingtranscription factor influence on burstingtranscription factorstranscriptional burst dynamicstranscriptional burstingtranscriptional regulation in different cell types
