Estimation Of Framework And Parameters For The Model Of Stochastic Gene Transcription Based On Qualitative Features Of Data

Gene transcription is the core process of life.Since the end of last century,the rapid development of fluorescent protein technology and high-resolution mi-crotechnique has promoted a large number of gene expression experiments at the single-cell level.Massive data confirm that gene transcription is essentially a dis-continuous and random burst process,which leads people to rethink the essence of life.The problem discussed in this paper is driven by experimental data,and the framework of the model is screened based on the dynamic characteristics of transcriptional mean;Parameter estimation of the model is carried out based on transcriptional distribution morphology.The first chapter summarizes the research background of stochastic gene tran-scription,emphasizes different model frameworks and parameter estimation meth-ods,and points out some limitations of the current methods.In Chapter 2,We found that prior estimation of the framework can be facili-tated by the traditional dynamical data of m RNA average level M(t),presenting discriminated dynamical features.Rigorous theory regarding M(t)profiles allows to confidently rule out the frameworks that fail to capture M(t)features and to test potential competent frameworks by fitting M(t)data.We implemented this procedure for a large number of mouse fibroblast genes under tumor necrosis fac-tor induction and determined exactly the‘‘cross-talking n-state”framework;the cross-talk between the signaling and basal pathways is crucial to trigger the first peak of M(t),while the following damped gentle M(t)oscillation is regulated by the multi-step basal pathway.This framework can be utilized to fit sophisticated single-cell data and may facilitate a more accurate understanding of stochastic activation of mouse fibroblast genes.In Chapter 3,we focus on the parameter estimation of the two-state model.We find that the maximum likelihood method or the method based on the moment of m RNA distribution data to infer the gene parameters in the mathematical model can not well grasp the characteristics of transcription distribution morphology P₀plays an important role in distinguishing different distribution forms.Therefore,in this study,we emphasize the importance of P₀data,and a generalized moment based method is developed for more reliable parameter estimation in the classical two-state model.Our method is easy to follow and allows estimating the parame-ters of E.coli and mammalian genes under different conditions.Finally,it shows that our method performs better than the traditional moment based method,and our method is more likely to grasp the bimodal m RNA distribution data than the distribution morphology obtained by the maximum likelihood method.The last chapter looks forward to the future work.We will introduce nascent m RNA data and m RNA dynamic distribution data;Further optimize the screening method of model framework estimation to improve the reliability of parameter estimation.