🧬 biology

Quantitative ergodicity for gene regulatory networks with transcriptional bursting

This paper establishes the existence and uniqueness of stationary distributions for stochastic gene regulatory networks with transcriptional bursting and provides explicit Wasserstein convergence bounds using coupling methods.

Original authors: Mathilde Gaillard, Ulysse Herbach

Published 2026-05-12

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Mathilde Gaillard, Ulysse Herbach

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine a cell as a bustling factory. Inside this factory, there are blueprints (mRNA) and workers (proteins) that keep the operation running. The paper you're looking at is a mathematical study of how these factories behave over a very long time, specifically when the production of blueprints happens in sudden, chaotic bursts rather than a smooth, steady stream.

Here is a breakdown of the paper's findings using simple analogies.

The Problem: The "Bursty" Factory

In many biological models, scientists assume that genes produce mRNA (the blueprint) at a steady, predictable rate, like a faucet dripping water. However, real life is messier. Genes often produce mRNA in sudden "bursts"—like a firehose turning on for a split second, flooding the factory with blueprints, then turning off completely.

The authors studied two versions of this "bursty factory":

The Simplified Factory: They looked at just the workers (proteins), assuming the blueprints appear and disappear so fast they don't need to be tracked individually.
The Complete Factory: They tracked both the blueprints (mRNA) and the workers (proteins) to see how the whole system interacts.

The Big Question: Does the Factory Ever Settle Down?

The main question the paper asks is: If you start with two different factories (one with too many workers, one with too few), will they eventually become identical in their behavior?

In math terms, this is called ergodicity. It asks if the system forgets its starting conditions and settles into a stable, predictable pattern over time.

The Solution: The "Shadow Runner" (Coupling)

To prove that these factories settle down, the authors used a clever mathematical trick called coupling.

Imagine you have two runners on a track:

Runner A represents Factory 1.
Runner B represents Factory 2.

Usually, they run at different speeds and take different paths. The authors created a special "Shadow Runner" (which they call a companion process) that runs alongside them.

Here is how the Shadow Runner works:

It acts like a measuring tape between Runner A and Runner B.
Every time there is a sudden burst of activity (a "jump"), the Shadow Runner checks the distance between the two factories.
If the factories are far apart, the Shadow Runner gets a little bigger. If they are close, it shrinks.
Crucially, the authors proved that this Shadow Runner will eventually stop jumping and shrink down to zero.

The Metaphor: Think of the Shadow Runner as a rubber band connecting the two factories. Even if the factories get pushed apart by a sudden burst of activity, the rubber band (the math) is strong enough to pull them back together. Eventually, the rubber band goes slack, meaning the two factories are behaving exactly the same way, regardless of how they started.

The Key Findings

The paper makes three major claims:

They Always Settle Down: No matter how many genes are interacting or how strong their interactions are, the system will eventually find a stable rhythm. You don't need to assume the interactions are weak for this to happen.
We Can Measure the Speed: The authors didn't just say "it happens"; they calculated exactly how fast it happens. They provided a formula that tells you how many years (or hours) it takes for the factories to synchronize.
- Analogy: It's like knowing exactly how long it takes for two different clocks to tick in unison after you shake them.
The "Burst" Doesn't Break the System: Even though the production happens in chaotic, random bursts, the system is robust. It doesn't spiral out of control; it naturally self-corrects.

Why This Matters (According to the Paper)

The authors note that previous math models required very strict rules (like "interactions must be weak") to prove the system would settle down. This paper removes those restrictions.

They show that even with strong interactions—like a "toggle switch" where two genes fight each other to turn the other off—the system still finds a stable state. In fact, they show that if you force the interactions to be too weak (to satisfy old math rules), you actually lose the interesting biological behavior (like the toggle switch flipping back and forth). Their new math allows for the strong, complex interactions that real biology actually uses.

Summary

In short, this paper proves that even in a chaotic, bursty biological system, order eventually emerges. By using a mathematical "Shadow Runner" to track the distance between two different starting points, the authors showed that the system always converges to a single, stable behavior, and they gave us the exact formula to calculate how quickly that happens.

Technical Summary: Quantitative Ergodicity for Gene Regulatory Networks with Transcriptional Bursting

Problem Statement
The paper addresses the long-term behavior of stochastic gene regulatory networks (GRNs) modeled by Piecewise-Deterministic Markov Processes (PDMPs). Specifically, it focuses on two models describing transcriptional bursting, a biological phenomenon where mRNA is produced in random, instantaneous bursts.

Model 1 (Reduced): A protein-only model where bursts are applied directly to protein levels, valid when mRNA degradation is much faster than protein degradation.
Model 2 (Complete): A coupled mRNA-protein model where bursts occur at the mRNA level, influencing protein dynamics.

While the existence and uniqueness of stationary distributions for such systems have been established in prior literature (e.g., Costa and Dufour, 2008; Benaïm et al., 2015), these results often lack quantitative rates of convergence to equilibrium. Existing methods for obtaining explicit convergence rates in Wasserstein distance are typically restricted to specific subclasses of PDMPs, such as switched dynamical systems with bounded state spaces or models with constant jump rates. The authors aim to extend these results to multivariate PDMPs with unbounded trajectories and non-constant, state-dependent jump rates, without requiring restrictive assumptions on the strength of gene interactions.

Methodology
The authors employ coupling methods to derive explicit upper bounds for the convergence to equilibrium in terms of the Wasserstein-1 distance ( $W_1$ ). The core of their methodology involves the construction of a companion process, a one-dimensional Markov process that bounds the distance between two trajectories of the original system.

Key methodological steps include:

Coupling Construction: Two copies of the system (representing two cells with potentially different initial states) are coupled to evolve together. The coupling is designed to maximize the probability of simultaneous jumps (synchronization) while maintaining the correct marginal distributions.
Companion Process ( $U(t)$ ): Since the difference between the two trajectories is not itself a Markov process, a scalar process $U(t)$ is introduced. This process bounds the distance between the two systems ( $\|X_1(t) - X_2(t)\| \leq U(t)$ ).
Analysis of the Companion Process: The authors analyze the dynamics of $U(t)$ , which is driven by a deterministic flow (degradation) and jumps (representing the failure of synchronization). They prove that $U(t)$ almost surely undergoes a finite number of jumps before stabilizing.
Stochastic Bounds: The proof relies on constructing explicit stochastic bounds for the number of jumps and the waiting times between jumps. Unlike previous works on switched systems (e.g., Benaïm et al., 2012) which assumed bounded state spaces, this work handles unbounded trajectories and does not require the jump rate to be bounded away from zero.
Lipschitz Assumptions: The jump rate function $\alpha$ is assumed to be bounded and Lipschitz continuous, but the authors explicitly note that their ergodicity results hold regardless of the magnitude of the Lipschitz constant (interaction strength), provided the degradation rates satisfy specific biological constraints.

Key Contributions and Results

Existence and Uniqueness: The paper proves the existence and uniqueness of a stationary distribution for both Model 1 and Model 2 for an arbitrary number of interacting genes and arbitrary interaction strengths.
Explicit Convergence Rates: The primary contribution is the derivation of fully explicit exponential convergence rates in the Wasserstein distance.
- For Model 1, the convergence rate $\gamma$ is given by $\gamma = \frac{p^* \tau d_1}{p^* \tau + d_1}$ , where $p^*$ is a probability related to the initial distance and interaction strength, $\tau$ is the minimum of the burst frequency and degradation rate, and $d_1$ is the minimum protein degradation rate.
- Similar explicit bounds are provided for Model 2, accounting for the additional mRNA layer.
Relaxation of Constraints: The results extend previous theorems (specifically Chen and Jia, 2019) by removing the requirement that the interaction strength (Lipschitz constant $\ell$ ) must be small relative to the degradation rate. The paper demonstrates that ergodicity holds even for strong interactions, such as those found in bistable toggle switches, which were previously excluded by stricter conditions.
Handling Unboundedness: The methodology successfully adapts coupling techniques to PDMPs with unbounded trajectories driven by a single flow, a setting where standard bounded-domain techniques fail.

Significance and Claims
The authors claim that their work provides a rigorous mathematical foundation for the stability of stochastic gene regulatory networks under biologically realistic conditions (transcriptional bursting and variable interaction strengths).

Biological Relevance: The results confirm that even in regimes where strong interactions lead to complex dynamics like bistability (e.g., toggle switches), the system remains ergodic and converges to a unique stationary distribution. This contrasts with earlier theoretical restrictions that would have precluded such biological phenomena from being analyzed under ergodic frameworks.
Methodological Extension: The paper presents an extension of the coupling technique developed by Benaïm et al. (2012) for switched dynamical systems to a broader class of "bursty" PDMPs. The authors note that while the proof scheme is similar, the mathematical technicalities regarding unbounded trajectories and the control of waiting times are distinct and non-trivial.
Future Perspectives: The authors suggest that the companion process technique could be extended to models with more chemical species (e.g., pre-mature mRNA) and potentially to multidimensional companion processes to better capture network topology, though they acknowledge this increases mathematical difficulty. They also note that while the derived rates are explicit, their biological relevance for simulation purposes may be limited compared to the qualitative assurance of ergodicity.

In summary, the paper establishes that two standard models of gene expression with bursting are exponentially ergodic under minimal assumptions, providing explicit convergence rates that depend on the system's degradation rates, interaction strengths, and initial conditions.