Mathematical modeling of semi-Markov biomolecular processes and the memory effect

Zihang Huang, Haowen Chen, Joyce Teng, Wenjie Cao, Zhihui Cai, Jiajun Zhang, Tianshou ZhouScientia 2025 · DOI

This paper, published in Scientia Sinica Mathematica, provides a rigorous mathematical bridge between complex, memory-dependent biological processes and the more manageable world of Markovian modeling. By introducing "effective transition rates," the authors show how "molecular memory"—the idea that a reaction's probability depends on how long a molecule has already existed—functionally mimics feedback loops and can even trigger biological oscillations.

TL;DR

The authors develop a mathematical framework to transform non-Markovian (semi-Markov) biomolecular reaction networks into equivalent Markovian models using "effective transition rates." This approach reveals that molecular memory acts as an implicit feedback mechanism, providing a powerful tool for inferring biological mechanisms from noisy experimental data like single-cell RNA sequencing.

What Problem Does This Paper Address?

Most classical models in computational biology, such as the Chemical Master Equation (CME), assume that biochemical reactions are "Markovian." This means the probability of a reaction occurring depends only on the current state of the system, implying that waiting times between reactions follow an exponential distribution.

However, real biological processes—like gene transcription or protein signaling—are often composed of many hidden sub-steps. When we simplify these into a single observable step, the waiting times become non-exponential, introducing "memory" into the system. Modeling these "semi-Markov" processes is notoriously difficult because the standard tools of stochastic calculus often break down. The authors tackle the challenge of creating a unified mathematical framework that can handle this memory effect without becoming computationally intractable.

Methods

The researchers utilize Renewal Theory to build a generalized modeling framework. Their key contributions include:

  • Effective Transition Rate ($K_i(x)$): The authors derive a novel function that incorporates the memory effects of all reactions in a network. By using this rate, they can transform a semi-Markov problem into a mathematically equivalent Markov problem.
  • Generalized Master Equations: They derive the "Main Probability Equation" in both the time domain and Laplace domain, providing a hierarchy of models:
  • Master Equation: For discrete, stochastic counts.
  • Fokker-Planck Equation: For continuous stochastic approximations.
  • Langevin Equation: For stochastic differential equations (SDEs).
  • Deterministic Equations: For macroscopic averages (ODEs).
  • Mapping Topology: They prove that the transformed Markovian network retains the same steady-state distribution as the original semi-Markovian network, allowing researchers to use simpler tools to study complex long-term behaviors.

Datasets

As a theoretical and methodological paper, it does not introduce a new experimental dataset. Instead, it validates the framework using:

  • Simulated Microscopic Systems: Specifically, a generalized birth-death process and a microscopic version of the Brusselator (a classic chemical oscillator model).
  • Epidemiological Frameworks: The authors demonstrate their method's versatility by applying it to the SIR (Susceptible-Infected-Removed) model, showing that classic epidemic models are specific cases of their semi-Markovian framework.

Key Insights and Ideas

  • Memory as Hidden Feedback: One of the most striking "aha moments" is the demonstration that molecular memory is functionally equivalent to feedback regulation. For example, a birth process with memory can behave exactly like a system with negative feedback, even if no physical feedback loop exists.
  • Memory-Induced Oscillations: The authors show that memory can induce Hopf bifurcations. In the Brusselator example, a system that is stable in a Markovian setting can be driven into sustained oscillations simply by increasing the "memory index" (the non-exponentiality) of the reactions.
  • Noise-Induced Shifts: The paper proves that molecular noise doesn't just add "fuzziness"; it can actually shift the equilibrium of a system. Theorem 3.1 provides a way to calculate how noise intensity ($\epsilon$) alters the deterministic steady state.

Key Results

  • Analytical Transformation: The paper provides an explicit formula (Eq. 2.17) to calculate the effective transition rate for any waiting-time distribution.
  • Stability Analysis: They provide a first-order approximation (Theorem 3.2) for how noise and memory affect the eigenvalues of a system, allowing researchers to predict when a biological system will become unstable or start oscillating.
  • Statistical Inference Framework: The authors outline a Maximum Likelihood Estimation (MLE) framework (Section 4) that uses their models to infer "memory indices" from single-cell RNA-seq data, providing a path for experimentalists to "see" hidden multi-step processes.

Limitations and Future Work

  • Transient Dynamics: While the authors prove that the steady-state distributions are identical between the original and transformed models, they acknowledge that the transient (short-term) behavior might still differ. Further research is needed to quantify these differences.
  • Computational Scaling: Calculating the "theoretical distribution" for very high-dimensional systems remains a significant computational hurdle for real-world data inference.
  • Passive vs. Active CTRW: The paper focuses on "active" Continuous-Time Random Walks (where waiting times reset after any reaction). The "passive" case, where memory persists across different types of reactions, remains an open and more complex challenge in high dimensions.

Summary updated

Abstract

生物分子系统本质上是生化反应网络,其状态的时间演化能够看成是网格上的连续时间随机行走(CTRW). 假如系统状态之间的等待时间都服从指数分布,那么相应的生化反应网络是马氏的,但假如某些状态之间的等待时间服从非指数分布,则相应的生化反应网络是半马氏的. 马氏生化反应网络的建模问题已经得到解决,但半马氏生化反应网络的数学建模是一项挑战. 本文应用随机过程理论(实际是应用更新理论),建立了一大类半马氏生化反应网络的数学模型. 更为重要的是,通过对这类网络中的每个反应引入有效转移率函数,成功地把半马氏问题转化成马氏问题,这给理论分析带来了极大方便. 两个简单例子的分析表明分子记忆的效果等价于反馈调控的引入、分子记忆能够诱导振荡. 本文的数学模型为基于实验数据的推断打下了坚实基础,且建模方法能够应用于许多其它科学领域.
BibTeX
@inproceedings{huang2025mathematical,
  author = {Zihang Huang and Haowen Chen and Joyce Teng and Wenjie Cao and Zhihui Cai and Jiajun Zhang and Tianshou Zhou},
  title = {{Mathematical modeling of semi-Markov biomolecular processes and the memory effect}},
  booktitle = {Scientia},
  year = {2025},
  doi = {10.1360/ssm-2024-0261},
}
Figure