Abstract:Stochastic models for chemical reaction networks are increasingly popular in systems and synthetic biology. These models formulate the reaction dynamics as Continuous-Time Markov Chains (CTMCs) whose propensities are parameterized by a vector $\theta$ and parameter sensitivities are introduced as derivatives of their expected outputs with respect to components of the parameter vector. Sensitivities characterise key properties of the output like robustness and are also at the heart of numerically efficient optimisation routines like Newton-type algorithms used in parameter inference and the design of of control mechanisms. Currently the only unbiased estimator for second-order sensitivities is based on the Girsanov transform and it often suffers from high estimator variance. We develop a novel estimator for second-order sensitivities by first rigorously deriving an integral representation of these sensitivities. We call the resulting method the Double Bernoulli Path Algorithm and illustrate its efficiency through numerical examples.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of unbiased estimation of second - order parameter sensitivities in Stochastic Reaction Networks (SRNs). Specifically, the authors focus on how to develop a new and more efficient algorithm to estimate these second - order sensitivities in order to overcome the high - variance problem existing in existing methods (such as methods based on Girsanov transformation). #### Main problem background 1. **Stochastic reaction network model**: - Chemical Reaction Networks (CRNs) are used to describe molecular processes within cells. These models are widely used in systems biology and synthetic biology. - When the number of certain molecular species is low, SRNs can better capture the discreteness and inherent randomness of the system. 2. **Parameter sensitivity**: - Parameter sensitivity reflects the degree of response of the system output to parameter changes and is an important tool for understanding system robustness and optimizing parameter inference. - First - order sensitivities have been widely studied, but the study of second - order sensitivities is relatively less, although they can provide more refined dependencies. 3. **Limitations of existing methods**: - The only currently unbiased estimation method is based on Girsanov transformation (GT), but it usually has a high variance and cannot be calculated especially when some parameters are zero. - Other methods (such as the finite difference method) are biased, and as the bias decreases, the variance will increase sharply. #### Main contributions of the paper 1. **Theoretical basis**: - Provides the conditions for the existence of second - order parameter sensitivities in SRNs and gives a new integral representation form. 2. **New algorithm development**: - Develops a new algorithm named "Double Bernoulli Path Algorithm (DBPA)" to generate unbiased estimates of second - order sensitivities. - Through numerical experiments, shows that DBPA has significant performance improvement compared with the existing GT method. 3. **Variance reduction techniques**: - Introduces Bernoulli random variables to modulate the computational cost while maintaining the unbiasedness of the estimator. - Uses coupling techniques (such as split coupling) to reduce the variance of the estimator. In summary, this paper solves the high - variance problem in the estimation of second - order parameter sensitivities in SRNs, proposes a new and more efficient unbiased estimation method, and verifies its superiority through theoretical analysis and numerical experiments.

Unbiased estimation of second-order parameter sensitivities for stochastic reaction networks

Unbiased estimation of parameter sensitivities for stochastic chemical reaction networks

Efficient Finite-Difference Estimation of Second-Order Parametric Sensitivities for Stochastic Discrete Biochemical Systems

Estimation of parameter sensitivities for stochastic reaction networks using tau-leap simulations

Efficiency of the Girsanov transformation approach for parametric sensitivity analysis of stochastic chemical kinetics

Scalable Parameter Estimation for Genome-Scale Biochemical Reaction Networks

Accelerated Sensitivity Analysis in High-Dimensional Stochastic Reaction Networks

Variance-based sensitivity analysis for stochastic chemical kinetics

Sensitivity analysis for multiscale stochastic reaction networks using hybrid approximations

A Novel Analytical Second‐order Sensitivity Calculation Approach Using the Finite Element Method for Chemical Engineering Problems

Inference for Stochastic Chemical Kinetics Using Moment Equations and System Size Expansion

Parameter Estimation for Dynamical Systems with Discrete Events and Logical Operations

Parametric Sensitivity Analysis for Biochemical Reaction Networks based on Pathwise Information Theory

Steady-state parameter sensitivity in stochastic modeling via trajectory reweighting

Bayesian Parameter Estimation for Biochemical Reaction Networks Using Region-Based Adaptive Parallel Tempering

Parametric Sensitivity Analysis for Models of Reaction Networks within Interacting Compartments

LNA++: Linear Noise Approximation with First and Second Order Sensitivities

Efficient Sensitivity Analysis for Parametric Robust Markov Chains

Efficient computation of adjoint sensitivities at steady-state in ODE models of biochemical reaction networks

Computing Sensitivities in Evolutionary Systems: A Real-Time Reduced Order Modeling Strategy

State and parameter estimation from exact partial state observation in stochastic reaction networks