Accelerating reaction network identification via frequency-domain analysis

Abstract

The identification of intracellular reaction networks from single-cell data lies at the heart of many biological studies, as it can unravel key mechanisms in living organisms and provide insights for their rational engineering. However, the stochastic and nonlinear nature of these reacting systems poses significant challenges for this identification problem, making the state-of-the-art methods (e.g., particle-filtering-based approaches) computationally demanding. This paper reports a noteworthy numerical observation that for certain systems, the Fourier transform can render the likelihood of the measurement data approximately Gaussian, therefore resulting in a substantially accelerated Bayesian inference algorithm. A gene transcription example is presented to illustrate this finding and demonstrate the efficiency of {the proposed Fourier-based method}.