Identification And Function Analysis Of Gene Regulatory Network

Gene regulatory network(GRN)is used to investigate the interactions among deoxyribonucleic acid,ribonucleic acid,proteins and metabolites.In natural environment,the organisms with good adaptation are of more survival opportunity.Investigating GRN adaptation is of great biological significances.Identifying GRN from observed data is helpful to recognize the system.This paper dedicates to the research on GRN adaptation and GRN identifiability.The main work and conclusions of this paper are as follows:(1)Using the modified optimization algorithms,the identification efficiency(including topology and parameters)and the quality of solutions are improved for finding three-node GRN with robust adaptation.Several improved heuristic searching algorithms are proposed to solve the shortcomings of the Latin hypercube sampling(LHS)method,including low efficiency and poor quality solutions.1)The unknown topological three-node GRN is modeled by the general Michaelis-Mentexi rate equations.A multi-objective genetic algorithm(GA)is proposed to effectively identify robust adaptation GRN,including topology identification and patameters identification.The proposed multi-objective GA can find more feasible topologies and parameter sets with high efficiency.2)For three-node GRN of negative feedback loop with a buffering node(NFBLB)topology,an improved particle swarm optimization(PSO)algorithm is proposed.The proposed PSO algorithm can obtain high quality solutions,which make the NFBLB topological GRN be of better adaptation.Several optimization algorithms are compared.They can provide the guidance for optimization scheme selection,and also provide sufficient samples for the following chapter to analyze the relationship between the adaptation and GRN parameters.(2)The parameter distribution rules are analyzed for three-node GRN with robust adaptation.For three-node GRN with NFBLB and incoherent feedforward loop with a proportioner node(IFFLP)topology,the robust adaptation dynamics are analyzed.Firstly,two new adaptation indices,i.e.,peak time and settle down time,are proposed for the first time to give more accurate and comprehensive description of robust adaptation.Secondly,using the sensitivity and precision as optimization objectives,the peak time and settle down time as constraints,seven constraint multi-objective optimization algorithms are used to find large number of solutions.Thirdly,a fuzzy C mean algorithm is used to analyze these solutions to obtain the relationship between the model parameters and the desired adaptation.And the parameter motif can be found for satisfactory and better adaptation GRNs.Finally,by means of analyzing four adaptation indices for two feasible topologies,it is found that the robust adaptation depends more on the GRN topology than the model parameter set.These results are helpful to design GRN with robust adaptation.(3)The GRN identifiability based on observed data is proposed.The identifiability indices are defined and the guiding principles are provided to design observed data experiment.To show how the observed data influence the identification accuracy,three data collection schemes are designed for a five-node GRN,i.e.,the transient observed data from a given initial condition(scheme 1),steady-state observed data(scheme 2)and the dynamic observed data from impulse excitation response(scheme 3).The improved GA-PSO algorithm is used to identify GRN parameters.The simulation results show that the GRN is of identifiability using scheme,and it is not of identifiability using scheme 2.Meanwhile,the equilibrium point of the S1system model is used to explain the reason for identification failure due to using scheme 2.Although the GRN is of identifiability using scheme 1,the observed data in scheme 1 are obtained by the zero-input response from a specific initial condition.It is difficult to obtain this type of observed data(scheme 1)for biological system.Usually,the stable-state observed data can be obtained in routine observation.To address this issue,in scheme 3,an impulse excitation is used to obtain the dynamic observed data from impulse excitation response,which make the GRN be of identifiability.What kinds of input-response data can be used to identify GRN?This problem is how to excite a nonlinear system,which is still an open filed.In this paper,there is a good performance by means of using the observed data from impulse excitation response,which provides the guidance for biologists to design the experiment for data collection.We also test the identification performance using step excitation.The results show that the impulse excitation is better than step excitation.In this paper,for the three-node GRN,several heuristic searching algorithms are compared to identify the unknown and predefined topological GRN with robust adaptation.They are helpful to address the existing issue in LHS method,i.e.,low efficiency and poor quality solutions.Meanwhile,they provide different schemes for practical application according to their characteristics,and give a new way to identify robust adaptation GRN effectively.Two new adaptation time indices are proposed to give more comprehensive and accurate description of robust adaptation.The relationship between the parameters and the desired adaptation is analyzed,and the parameter motifs can be found.The dynamic characteristics of robust adaptation are analyzed for two feasible topological GRNs.These results are helpful to design robust adaptation GRNs.For a five-node S-system GRN,the validity of the observed data is proposed.We analyze that how the validity of observed data influences on estimation accuracy.The reason why the GRN is not of identifiability using stable-state observed data is explained.The data collection schemes are proposed to guarantee the identifiability of GRN.