Research On Some Computational Issues Of Biological Metabolic Network

The Human Genome Project launched in the 1990s and officially opened the era of global biological big data.With the completion of the Human Genome Project,genomic informatics was born to deal with the huge amount of data generated by it,and the metabolic network is closely related to genomic informatics,which has become a hot spot of research and application.This involves the calculations of correlation between enzyme reactions,the calculation algorithm of metabolic pathways,the simulation calculation of ¹³C labeling kinetics and the ¹³C metabolic flux calculation.The metabolic network is a complete set of metabolic and material processes that determine the cell physiological and biochemical properties.Metabolic networks perform the material and energy exchange required for biological survival,development and reproduction.These networks contain metabolic chemical reactions and regulatory interactions that guide these reactions.In this thesis,the relationship analysis and calculation between the evolutionary correlation and functional correlation of enzymatic reactions in the metabolic network,the calculation algorithm of metabolic pathways between two metabolites in the metabolic network,the stochastic simulation calculation of ¹³C labeling kinetics at the single-cell scale and the nonstationary ¹³C metabolic flux calculation were studied in depth.The main research contents and innovations are as follows.（1）A relationship calculation method for evolutionary correlation and functional correlation of enzymatic reactions in the metabolic networks was proposed.The evolution of species is inevitably accompanied by the evolution of metabolic networks to adapt to different environments.The metabolic networks of different species were collected from the KEGG（Kyoto Encyclopedia of Genes and Genomes）website,and some enzymatic reactions with the highest occurrence frequency in all species were found.The correlation coefficients of whether the enzymatic reactions appear in all species were calculated,and the corresponding evolutionary correlation connection networks were calculated according to different correlation coefficient thresholds.These studies show that,as the evolutionary correlation of enzymatic reactions increases,the weighted average of the mean functional concentration ratios of the enzymatic reactions also increases,indicating that the functional concentration ratio of enzymatic reactions has a certain correlation with the evolutionary correlation.The work in this part enhances our understanding of the characteristics and general rules of metabolic network evolution.（2）An algorithm for calculating the metabolic pathways that do not contain each other between any two metabolites in a metabolic network was proposed.Synthetic biology is a very important research field for medicine,pharmacy and agriculture.An important component of synthetic biology is the design of small molecular biosynthetic pathways,which involves the characterization,alteration and manipulation of biosynthesis pathways.However,it is not appropriate to use only a simple path to represent the biosynthetic pathway because there are multiple substrates or products in enzymatic reaction.B-hyperpath is the best approach for describing biosynthesis pathways.The definition and classification of hyperpath,the definition of B-hyperpath,and the reasons for using B-hyperpaths to represent biosynthesis pathways were introduced in this thesis.An algorithm for calculating B-hyperpaths between any two nodes in a directed hypergraph that does not contain each other was proposed,and the algorithm was tested in the Escherichia coli central metabolic network.Finally,the time complexity of the algorithm was analyzed.The algorithm can be used to find necessary metabolic pathways for synthesizing a particular compound,which is the basis for designing and modifying biosynthesis pathways.（3）Theτ-leap algorithm was introduced into the simulation caculation of ¹³C labeling kinetics at the single-cell scale was proposed.¹³C fluxomics is a new type of omics that can detect the overall flux in a metabolic network.At present,this method has been widely adopted in homogeneous cell systems and some heterogeneous cell systems and even in living tissues.However,unlike other omics methods,this method has not been implemented at the single-cell level.In this part of the research work,aτ-leap algorithm for stochastic simulation of chemical reaction was adopted to ¹³C labeling kinetics simulation of enzymatic reaction.The ¹³C labeling kinetics simulation of enzymatic reactions in single cell was completed,and the ¹³C labeling kinetics simulation of enzymatic reactions in multiple single cells was completed with the parameters from Escherichia coli cell.In a very short time interval,the numbers of isotopomer molecules and the proportions of isotopomer molecular numbers of metabolites were calculated.Finally,the intergroup variances for multiple cell groups were calculated and compared.The simulation results show that the numbers of isotopomer molecules and the proportions of isotopomer molecular numbers of the metabolites are heterogeneous at the single-cell level.This heterogeneity gradually decreases with the increase of the initial numbers of isotopomer molecules of the metabolites in the single-cell.In addition,the intergroup heterogeneity of the numbers of isotopomer molecules and the proportions of isotopomer molecular numbers shows a decreasing trend with the increase of the number of simulated cells within the group.These results demonstrate the necessity and possibility for ¹³C flux analysis at single-cell in future.（4）The neural network and intelligent algorithm were used to predict the nonstationary ¹³C metabolic flux by inputting the isotopomer abundances of metabolites was proposed.The ¹³C metabolic flux analysis is a powerful tool for synthetic biology.The flux estimation based on ¹³C labeling depends on iterative fitting and complex simulation that requires a certain level of expertise.In order to overcome this problem,BP（Back Propagation）neural network,a method of combining BP neural network with genetic algorithm,a method of combining BP neural network with particle swarm optimization algorithm,traditional recurrent neural network,long short-term memory neural network and convolutional neural network were respectively used to predict the nonstationary ¹³C metabolic flux of Escherichia coli central metabolic network by inputting the isotopomer abundances of metabolites in this thesis.The results show that the prediction effect of convolutional neural network is the best.In addition,the results of predicting the nonstationary ¹³C metabolic flux by convolutional neural network were compared with the results of optimizing the nonstationary ¹³C metabolic flux by genetic algorithm.It is concluded that the effect of predicting metabolic flux by convolution neural network is better than the effect of optimizing metabolic flux by genetic algorithm.This paves the way for the nonstationary ¹³C metabolic flux analysis in future.