A Spectral Koopman Approximation Framework for Stochastic Reaction Networks

TL;DR

This paper introduces a spectral Koopman operator framework for analyzing stochastic reaction networks, enabling low-dimensional, data-driven predictions of system dynamics, sensitivities, and spectral properties across various biological and ecological systems.

Contribution

It develops a spectral approach to approximate the Koopman operator for SRNs, providing error estimates and efficient computation of spectral modes from data.

Findings

Successfully applied to biological systems including feedback controllers and oscillators.

Enables prediction of moments and event probabilities across initial states.

Provides tools for sensitivity analysis and frequency-domain characterization.

Abstract

Stochastic reaction networks (SRNs) are a general class of continuous-time Markov jump processes used to model a wide range of systems, including biochemical dynamics in single cells, ecological and epidemiological populations, and queueing or communication networks. Yet analyzing their dynamics remains challenging because these processes are high-dimensional and their transient behavior can vary substantially across different initial molecular or population states. Here we introduce a spectral framework for the stochastic Koopman operator that provides a tractable, low-dimensional representation of SRN dynamics over continuous time, together with computable error estimates. By exploiting the compactness of the Koopman operator, we recover dominant spectral modes directly from simulated or experimental data, enabling efficient prediction of moments, event probabilities, and other summary statistics across all initial states. We further derive continuous-time parameter sensitivities and cross-spectral densities, offering new tools for probing noise structure and frequency-domain behavior. We demonstrate the approach on biologically relevant systems, including synthetic intracellular feedback controllers, stochastic oscillators, and inference of initial-state distributions from high-temporal-resolution flow cytometry. Together, these results establish spectral Koopman analysis as a powerful and general framework for studying stochastic dynamical systems across the biological, ecological, and computational sciences.