Inferring parameters of gene regulatory networks via particle filtering

Xiaohu Shen, Haris Vikalo

Abstract

Gene regulatory networks are highly complex dynamical systems comprising biomolecular components which interact with each other and through those interactions determine gene expression levels, that is, determine the rate of gene transcription. In this paper, a particle filter with Markov Chain Monte Carlo move step is employed for the estimation of reaction rate constants in gene regulatory networks modeled by chemical Langevin equations. Simulation studies demonstrate that the proposed technique outperforms previously considered methods while being computationally more efficient. Dynamic behavior of gene regulatory networks averaged over a large number of cells can be modeled by ordinary differential equations. For this scenario, we compute an approximation to the Cramer-Rao lower bound on the mean-square error of estimating reaction rates and demonstrate that, when the number of unknown parameters is small, the proposed particle filter can be nearly optimal.

Introduction: The Challenge of Biological Circuitry

Gene Regulatory Networks (GRNs) serve as the biological "circuitry" of a cell. These networks, consisting of genes, proteins, and mRNA, interact dynamically to control essential functions such as growth, metabolism, and disease response. Understanding these interactions is a cornerstone of modern drug discovery and personalized medicine.

However, GRNs are inherently noisy and stochastic. Thermal fluctuations at the molecular level mean that even genetically identical cells can exhibit different behaviors. While researchers often understand the qualitative structure of these networks—which gene activates or represses another—the quantitative parameters, such as reaction rates, remain largely unknown. Estimating these "knobs" from experimental data is difficult because biological measurements are typically noisy, sparse, and computationally expensive to process using traditional exact methods like the Chemical Master Equation (CME).

A schematic of a gene regulatory network involving transcriptional activation and repression.

Methodology: Bridging Accuracy and Efficiency

To address these challenges, this research proposes a hybrid approach that balances computational tractability with physical accuracy.

1. The Chemical Langevin Equation (CLE)

Instead of the computationally prohibitive CME, the authors utilize the Chemical Langevin Equation. The CLE is a stochastic differential equation that approximates the network's behavior while still accounting for intrinsic molecular noise. This makes the model more amenable to simulation and parameter estimation in complex systems.

2. The PF-MCMC Algorithm

The core of the proposed solution is a Particle Filter (PF) enhanced with a Markov Chain Monte Carlo (MCMC) move step.

  • Particle Filtering: This technique represents the probability distribution of unknown parameters using a set of discrete "particles" that evolve over time as new data arrives.
  • MCMC Move Step: A common failure in standard particle filters is "sample impoverishment," where particles collapse into a single value, losing the diversity needed to explore the parameter space. By adding an MCMC move step (using the Metropolis-Hastings algorithm), the filter "nudges" particles, ensuring they continue to explore the space effectively and converge on the true values.

3. Data Augmentation

Biological data is often sampled at wide intervals. To bridge these gaps, the authors use data augmentation, treating the state of the system between observations as "missing values." This allows the discretized model to better emulate the continuous diffusion process of the underlying biology.

Performance and Accuracy

The proposed PF-MCMC method was validated against simulated datasets, including a viral infection network and a prokaryotic regulation model. The results consistently showed that the hybrid filter provides more precise estimates than traditional MCMC methods.

Comparison of Mean Square Error (MSE) for parameter $c_5$ between the proposed Particle Filter (PF) and standard MCMC.

Key performance metrics include:

  • Speed: In a viral infection simulation, the PF-MCMC algorithm reached convergence in approximately 1,030 seconds, a significant improvement over the 5,500 seconds required by standard MCMC.
  • Precision: The algorithm demonstrated high robustness, accurately converging on correct reaction rates even when initial guesses were off by several orders of magnitude.
  • Theoretical Optimality: By comparing results to the Cramer-Rao Lower Bound (CRLB)—the theoretical limit for estimator accuracy—the authors found that for smaller networks, the particle filter is nearly optimal.

Conclusion and Future Directions

By combining the sequential processing power of particle filters with the exploratory capabilities of MCMC, this research provides a powerful tool for tuning the mathematical models of life. While the method is highly efficient for small-to-medium networks, its performance relative to the CRLB begins to degrade as the number of unknown parameters increases—a phenomenon known as the "curse of dimensionality."

Future work will focus on refining proposal distributions to maintain optimality in higher-dimensional spaces and developing methods to infer the network structure itself alongside its parameters.

Summary updated

Abstract

Gene regulatory networks are highly complex dynamical systems of biomolecular components - genes, mRNA, proteins. These components interact with each other and through those interactions determine gene expression levels, i.e., determine the rate of gene transcription to mRNA and, consequently, the rate of mRNA translation to proteins. In this paper, a particle filter with MCMC move step is employed for the estimation of parameters in a gene regulatory network modeled by a chemical Langevin equation. Simulations demonstrate that the proposed technique outperforms previously considered methods.
BibTeX
@inproceedings{shen2010inferring,
  author = {Xiaohu Shen and Haris Vikalo},
  title = {{Inferring parameters of gene regulatory networks via particle filtering}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2010},
  doi = {10.1109/icassp.2010.5495614},
}
Figure