Abstract:
Gene expression is pivotal in numerous biological processes, making the comprehension and analysis of its modeling critically important. The gene regulatory network modeling process often involves stochastic simulation algorithms, which necessitate extensive random simulations and ensemble averaging to determine moment values. This results in a considerable computational burden and added intricacy. Traditional moment closure approximations, based on oversimplified distribution assumptions, fall short in capturing the intricate nature of real-world systems and fail to accurately represent the nuances of biochemical reaction models with extensive interactions. Such methods typically neglect the full spectrum of possibilities inherent in biochemical reactions, characterized by complex interplays among numerous components.To overcome these obstacles, this study exploits the exceptional capabilities of artificial neural networks for regression analysis and introduces a novel moment closure approximation for gene regulatory networks that harnesses these networks. This innovative method employs neural networks to infer low-order moments representations of higher-order moments, subsequently utilizing ordinary differential equation solvers to compute the predicted moment values. This approach effectively resolves the limitations of traditional moment closure approximations, which do not adequately leverage the intricate details present in biochemical reaction models.The research utilizes simulated datasets, meticulously validated for integrity and reliability. A comparative analysis of the moment values predicted by the neural network-based method against those derived from traditional approaches demonstrates a marked increase in precision with the neural network method. Furthermore, when assessing computational time across varying sample data, the neural network moment closure method is shown to outperform both traditional moment closure and stochastic simulation algorithms in terms of efficiency.To summarize, the enhanced precision and computational efficiency of the neural network moment closure method not only underscore its validity but also introduce an innovative tool and methodology for advancing gene regulatory network research.