Markovian-Projection-in-filtering-for-SRNs

This repository contains an implementation of the Markov projection methods for the filtering problem for Stochastic Reaction Networks (SRNs). Two methods for dimensionality reduction are developed: standard Markovian Projection (MP) and Filtered Markov Projection (FMP).

A detailed description of the methods is provided in the paper:

Hammouda, C.B., Chupin, M., Münker, S. et al. Filtered Markovian projection: dimensionality reduction in filtering for stochastic reaction networks. Stat Comput 36, 189 (2026). https://doi.org/10.1007/s11222-026-10939-0

Running simulations

Use the main_bistable_network.m or main_linear_cascade.m files to run the simulations.

Code structure

The main functions and classes are listed below

• @SRN class stores the parameters of the stochastic process and provides functions to sample the trajectories using SSA and Tau-leap methods

• @PF class implements Particle Filter. Used at the first step of projection methods.

• @FFSP class implements the Filtered Finite State Projection method. Used to obtain the reference solution and at the second step of projection methods.

• markovian_projection_extrap function implements the standard MP and returns SRN of lower dimensionality

• generate_observations function generates input data for the filtering problem

Potential improvements

• Add parallel implementation of the FFSP and PF (see @FFSP/A.m and @PF/evolve.m)
• Use more efficient extrapolation methods in extrapolate_a_bar_