🔬 physics

Asymptotics of Protein Number Distribution in Stochastic Gene Expression Models under Burst Approximation

This paper systematically analyzes stochastic gene expression models under the burst approximation by deriving analytical time-dependent solutions for surrogate models with multiple gene states, establishing theoretical properties of the resulting protein distributions, and developing efficient computational algorithms while quantifying the approximation error relative to full models.

Original authors: Yuntao Lu, Yunxin Zhang

Published 2026-05-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Yuntao Lu, Yunxin Zhang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a cell as a busy, chaotic factory. Inside this factory, there are blueprints (genes) that tell the workers how to build products (proteins). But this factory isn't run by a strict, predictable manager; it's run by a chaotic system where things happen randomly. Sometimes the blueprints are hidden, sometimes they are out in the open, and when they are out, the workers don't just build one product at a time—they often go on a "spree," building a whole batch of products in a single burst before stopping.

This paper is a mathematical study of how to predict the number of products (proteins) in this factory, specifically focusing on these "bursts."

Here is a breakdown of what the authors did, using simple analogies:

1. The Problem: The Factory is Too Complicated to Count

In the real world, the factory process has two main steps:

Transcription: The blueprint is copied into a temporary note (mRNA).
Translation: The note is used to build the product (protein).

The authors explain that trying to track every single note and every single product in real-time is like trying to count every grain of sand on a beach while a storm is blowing. It's mathematically too messy to solve exactly for most situations.

2. The Shortcut: The "Burst" Approximation

Scientists have a clever trick called the "burst approximation." Since the temporary notes (mRNA) disappear much faster than the products (proteins), they decided to skip tracking the notes entirely.

Instead, they imagine that when the blueprint is active, it doesn't just make one product; it instantly "bursts" out a random number of products all at once, like a popcorn machine popping kernels in a single second. This simplifies the math significantly, turning a two-step process into a one-step "burst" process.

3. What the Authors Did: Building a Better Map

While others have used this "burst" shortcut before, they mostly just used it to get rough numbers. This paper goes much deeper. The authors built a new, more flexible mathematical model that allows for:

Multiple Switches: The blueprint can be in many different states (like "off," "on," "flickering," or "super-active"), not just on or off.
Any Burst Size: They didn't assume the bursts are always the same size. They allowed for any kind of random distribution (e.g., sometimes 1 product, sometimes 10, sometimes 100).

They derived a master formula (a "time-dependent solution") that tells you exactly what the probability of having a certain number of products is at any specific moment in time.

4. The Big Discoveries

Using this new formula, they found some surprising things about the "shape" of the protein distribution:

The "Geometric" Rule: They found that if the size of the bursts follows a specific pattern (called a geometric distribution, which is common in real biology), the number of proteins in the factory tends to follow a predictable pattern called a Negative Binomial Distribution. Think of this as a specific "fingerprint" that the protein counts leave behind.
Light vs. Heavy Tails: They discovered that if the bursts aren't too huge (specifically, if the probability of a huge burst drops off quickly), the chance of having an enormous number of proteins is very low (a "light tail"). However, if the bursts can get very large, the chance of having a massive number of proteins stays higher (a "heavy tail"). They used a method called the "Chernoff bound" (a statistical flashlight) to estimate exactly how likely these massive spikes are.
The "Recurrence" Puzzle: They tried to use a standard math trick (binomial moments) to reconstruct the full picture of protein numbers from the average data. They found that this trick works perfectly when bursts are small, but if bursts get too large, the math breaks down and the picture becomes blurry.

5. Checking the Shortcut: Is the "Burst" Model Accurate?

The most critical part of the paper is asking: "If we skip the mRNA notes and just use the burst shortcut, how much are we messing up the answer?"

They compared their simplified "burst" model against the full, complicated "two-step" model.

The Good News: The average number of proteins is exactly the same in both models. If you just want to know the average, the shortcut is perfect.
The Bad News: If you look at the variability (how much the numbers jump around) or higher-level statistics, the shortcut introduces errors.
The Verdict: The error gets smaller as the temporary notes (mRNA) degrade faster. However, the authors proved that just having fast-degrading notes isn't always enough to guarantee the shortcut is perfect for all details. They provided a mathematical "error bar" to tell scientists exactly how much trust they can put in the shortcut.

Summary

Think of this paper as a team of cartographers who found a way to draw a map of a chaotic city (the cell) by ignoring the side streets (mRNA) and only looking at the main highways (protein bursts).

They didn't just draw the map; they:

Proved exactly what the map looks like under different traffic conditions.
Showed that the map is accurate for finding the "average" location of a person, but might be slightly off if you are trying to predict rare, extreme traffic jams.
Gave other scientists a set of tools (algorithms) to quickly calculate these probabilities without needing a supercomputer.

The paper concludes that while the "burst" shortcut is a powerful tool, it is an approximation. It preserves the average but distorts the details of the chaos, and knowing exactly how it distorts them is crucial for accurate scientific modeling.

Technical Summary: Asymptotics of Protein Number Distribution in Stochastic Gene Expression Models under Burst Approximation

Problem Statement
Stochastic gene expression is inherently driven by random molecular events, leading to significant cell-to-cell variability in mRNA and protein abundance, particularly when copy numbers are low. While the Continuous-Time Markov Chain (CTMC) and the Chemical Master Equation (CME) provide a standard framework for modeling these systems, the full gene expression models (explicitly tracking mRNA and protein dynamics) are analytically intractable for protein copy number distributions, even in steady state. To address this, the "burst approximation" is widely used, eliminating the explicit mRNA intermediate and modeling protein production as instantaneous bursts. However, existing literature on burst approximation models has primarily focused on steady-state analytical results for specific cases (e.g., two-state genes) or numerical computations. There is a lack of systematic analysis regarding the general time-dependent dynamics, fine properties of the distribution (such as tail behavior), and the rigorous estimation of approximation errors relative to full models.

Methodology
The authors propose a generalized surrogate model under the burst approximation featuring $N$ distinct metastable gene states ( $S_i$ ) and arbitrary burst size distributions. The model is defined as a CTMC where:

Genes switch between states $S_i$ and $S_j$ with transition rates $a_{i,j}$ (no production) or $b^{[r]}_{i,j}$ (producing $r$ proteins).
Proteins degrade at a constant rate $\delta$ .
The dynamics are governed by a CME for the joint probability $P_{i,j}(m; t)$ of being in gene state $S_j$ with $m$ proteins.

The study employs the following analytical and computational techniques:

Generating Function Method: Used to derive an analytical, time-dependent solution to the CME. The solution is expressed as a series involving matrix exponentials and integrals over ordered time simplices.
Binomial Moment Method: The authors derive expressions for steady-state binomial moments ( $B_m$ ) by taking the temporal limit of the time-dependent solution. They establish a recurrence relation for these moments to facilitate computation.
Functional Analysis: Inequalities (Hölder, submultiplicativity of operator norms) are applied to the derived series to bound the probability mass function (PMF), moment generating functions, and tail probabilities.
Asymptotic Analysis: The validity of the burst approximation is assessed by comparing low-order moments of the burst model against the full gene expression model (which includes explicit mRNA dynamics), deriving upper bounds for the absolute differences.

Key Contributions and Results

Analytical Time-Dependent Solution:
The paper derives explicit analytical solutions (Equations 3 and 4) for the protein number distribution $P_{i,j}(m; t)$ for any number of gene states and arbitrary burst size distributions. While these series are complex for direct numerical evaluation, they serve as the foundation for further theoretical analysis.
Properties of the Steady-State Distribution:
- Boundedness and Tail Behavior: For geometrically distributed burst sizes (a common experimental observation), the authors prove that the steady-state PMF is bounded by a scaled negative binomial distribution.
- Light-Tailed Regime: If the geometric parameter $\lambda \in (0, 1/2)$ , the distribution is proven to be light-tailed (the moment generating function is finite in a neighborhood of zero). The tail probability is estimated using the Chernoff method, generalizing previous heavy-tailed laws.
- Recurrence Relations: A recurrence relation for binomial moments is derived (Equation 9), allowing for the recursive calculation of moments of arbitrary order.
Computational Algorithms:
The authors identify three specific scenarios where efficient algorithms can be developed to compute the steady-state distribution:
- Geometric Burst Sizes: Analytical expressions for the moment coefficients $C_r$ are derived, simplifying the recurrence.
- Poisson Burst Sizes: Analytical expressions for $C_r$ are derived using the properties of the Poisson distribution.
- Truncated Burst Sizes: For cases where large burst sizes are negligible, the series is truncated, allowing for finite-term computation.
  The paper notes that for geometric distributions with $\lambda \in [1/2, 1)$ , the series reconstruction of the PMF from binomial moments may diverge, highlighting a limitation of the moment method in certain parameter regimes.
Approximation Error Estimation:
By mapping the full gene expression model to the burst approximation, the authors derive identities relating the parameters of both models. They establish that while the first-order moment (mean protein count) is preserved exactly under the burst approximation, higher-order moments differ.
- An upper bound for the absolute difference between the second-order binomial moments of the full and burst models is derived (Equation 16).
- The error is shown to converge to zero at a rate of $O(u^{-2})$ as the mRNA degradation rate $u \to \infty$ , provided other parameters are fixed. This clarifies that the condition $u/\delta \gg 1$ alone is not sufficient to guarantee the validity of the approximation for higher-order statistics.

Significance and Claims
The paper claims to provide the first systematic analysis of the burst approximation that bridges time-dependent dynamics, steady-state distribution properties, and error quantification.

Theoretical: It establishes fine properties of the protein number distribution, specifically proving light-tailed behavior under geometric burst assumptions and providing rigorous bounds on tail probabilities.
Computational: It offers efficient algorithms for calculating distributions in three practically relevant cases (geometric, Poisson, and truncated bursts), overcoming the combinatorial bottlenecks of direct moment enumeration.
Validity: It rigorously quantifies the limitations of the burst approximation, demonstrating that while it preserves the mean, it alters higher-order moments, and the error is controllable only under specific asymptotic conditions regarding mRNA degradation rates.

The authors conclude that while the burst approximation is a powerful simplification, its application requires careful consideration of the burst size distribution parameters and the specific moments of interest, as higher-order statistics are not automatically preserved. Future work suggested includes estimating mixing times, addressing the divergence of moment methods for heavy-tailed bursts, and extending models to non-Markovian processes.