A method for calculating non-steady-state flux changes in biological network metabolism

Abstract

The present invention discloses a method for calculating the non-steady-state flow change of biological network metabolism, comprising the following steps: loading a metabolic calculation model; inputting the total system reaction time T and the time step Δt into the metabolic calculation model, obtaining the change state of the mass isotope distribution of the intermediate metabolites, and obtaining the predicted state of the metabolic reaction flow; the change state of the mass isotope distribution of the intermediate metabolites is composed of the change value of the mass isotope distribution of the intermediate metabolites over time and the metabolic time point t; the predicted state is composed of the flow value of the metabolic reaction at the metabolic time point t and the metabolic time point t; the metabolic calculation model includes: an intermediate metabolite concentration change calculation unit, an intermediate metabolite mass isotope distribution change calculation unit, a sample processing unit and a neural network construction unit. According to the above technical scheme, the calculation and prediction of the non-steady-state flow change of the biological network metabolism in the metabolic process can be realized, and the prediction result can be made more accurate.

Description

Technical Field

[0001] The present invention relates to the field of neural network applications, and in particular to a method for calculating changes in non-steady-state flow rates of biological network metabolism. Background Art

[0002] With the development of biological technology applications, the demand for metabolic heterogeneity analysis of brain tissue, tumors, plants and other systems is becoming increasingly urgent. For example, in pathological research, the metabolic activities in different parts of tumor tissue often differ significantly. Mass spectrometry imaging combined with metabolic non-steady-state flux analysis can track the metabolic activities of different regions in brain tissue, summarize the metabolic characteristics of neurodegenerative diseases or brain damage, and further reveal the changes in metabolic flow in the tumor microenvironment, helping to optimize tumor diagnosis and treatment plans. On the other hand, the metabolic activities in different regions of plant tissue are closely related to environmental responses. The spatial metabolic flow method can be used to analyze the metabolic responses of plants under different conditions, providing a reference for studying metabolic interactions.

[0003] During the analysis of metabolic systems, metabolic flux cannot be measured directly and must be inferred from other observations. In simple cases, metabolic flux can be obtained from the equilibrium concentration of extracellular metabolites, but in order to accurately estimate the metabolic flux within cells, cellular metabolomics based on isotope tracing is required. 13 C metabolic flux analysis ( 13 Metabolic flux analysis (13C-MFA) is the main method of metabolomics and has been successfully applied in many fields. Currently, three main methods have been formed: steady-state flux analysis (steady-state 13C-MFA, SS-MFA), isotopically-instationary 13C-MFA (isotopically-instationary 13C-MFA, INST-MFA) and metabolically-instationary flux analysis (metabolically-instationary 13C-MFA, MNST-MFA). Among them, metabolically-instationary flux analysis is a new concept proposed in recent years, which can be applied to systems where metabolic flux changes over time.

[0004] In the actual metabolic process, metabolic flux and metabolite concentration change with time. Therefore, it is necessary to analyze metabolic flux on a spatial scale in order to reveal the heterogeneity of metabolic reactions in different tissues or cell regions. In conventional isotope non-steady-state flux analysis, ordinary differential equations are often used, but this method can only solve the metabolite isotope distribution when neither the metabolite concentration nor the metabolic flux changes with time. 13C-MFA technology does not propose an algorithm for calculating changes in metabolic non-steady-state flux when both metabolic flux and metabolite concentration change over time. At the same time, the existing metabolomics technology can only obtain the spatial distribution of metabolite concentrations, and cannot analyze the dynamic flux of metabolic reactions in metabolic networks, so it cannot directly capture the dynamic characteristics of metabolic reactions in tissues.

[0005] Therefore, a method is needed to calculate and predict the changes in non-steady-state metabolic flux in biological networks during metabolic processes. Summary of the invention

[0006] To achieve the above object, the present invention provides a method for calculating the non-steady-state flow change of biological network metabolism, comprising the following steps:

[0007] Loading a metabolic calculation model, which is used to predict the flux value of metabolic reactions at future time points based on the substrate mass isotope distribution;

[0008] The total system reaction time T and time step Δt are input into the metabolic calculation model to obtain the change state of the mass isotope distribution of the intermediate metabolites and the predicted state of the metabolic reaction flow; the change state of the mass isotope distribution of the intermediate metabolites is composed of the change value of the mass isotope distribution of the intermediate metabolites over time and the metabolic time point t; the predicted state is composed of the flow value of the metabolic reaction at the metabolic time point t and the metabolic time point t; the metabolic time point t is accumulated n times according to the time step Δt to form t+n, and t+n is less than or equal to the total system reaction time T.

[0009] Before loading the metabolic calculation model, construct the metabolic calculation model, which includes: an intermediate metabolite concentration change calculation unit, an intermediate metabolite mass isotope distribution change calculation unit, a sample processing unit and a neural network construction unit; generate metabolic data in the intermediate metabolite concentration change calculation unit and the intermediate metabolite mass isotope distribution change calculation unit, generate training samples according to the metabolic data by the sample processing unit, and train and generate the metabolic calculation model by the neural network construction unit according to the training samples;

[0010] The intermediate metabolite concentration change calculation unit calculates the change value of the concentration value at time t+1 over time according to the change value of the intermediate metabolite concentration at time t over time, the flow rate of generating the intermediate metabolite at time t, and the flow rate of consuming the intermediate metabolite;

[0011] The intermediate metabolite mass isotope distribution change calculation unit calculates the intermediate metabolite mass isotope distribution change state at time t+1 using a differential equation according to the change value of the intermediate metabolite concentration value at time t and the change value of the intermediate metabolite mass isotope distribution at time t.

[0012] The change state and the predicted state are represented by change vectors, and the data structure is a matrix M×N, where M is the length of the mass isotope distribution of the metabolite and N is the number of time points.

[0013] The basic parameters of the metabolic calculation model include: the mass isotope distribution vector X of the substrate input ;

[0014] The metabolic calculation model obtains the flow value of the intermediate metabolic reaction of the intermediate metabolite through sampling by the program package Cobrapy; sets some flow values ​​to change with time, and defines the flow value that changes with time and the metabolic time point to form a vector V_reaction.

[0015] Furthermore, the input data of the sample processing unit include: vector V_reaction, vector C_metabolite of the change of the concentration value of the intermediate metabolite over time, vector X_metabolite of the change of the mass isotope distribution of the intermediate metabolite;

[0016] In the training sample generated by the sample processing unit, the feature set includes: change vector C_metabolite, change vector X_metabolite and metabolic time point t;

[0017] In the training samples generated by the sample processing unit, the label set is the vector V_reaction.

[0018] Further, the input parameters of the intermediate metabolite concentration change calculation unit include: the change vector C_metabolite at time t, and the output parameters include: the change vector C_metabolite at time t+1, where time t+1 is t+Δt;

[0019] The calculation method used by the intermediate metabolite concentration change calculation unit is:

[0020] C_metabolit t + (∑Vt 生成 -∑Vt 消耗 )×∆t = C_metabolit t+1 ,

[0021] Among them, V t生成 is the sum of the flow rates of the substance generated at time t, V t消耗 It represents the total flow rate of the substance consumed at time t, and Δt is the time step.

[0022] At t = 0, the intermediate metabolite concentration value is randomly selected from the range of experimentally measured metabolite concentration values.

[0023] Furthermore, the input parameters of the intermediate metabolite mass isotope distribution calculation unit are: the change vector C_metabolite at time t, the change vector X_metabolite at time t, and the output is the change vector X_metabolite at time t, where time t+1 is t+Δt.

[0024] The calculation method used by the intermediate metabolite mass isotope distribution calculation unit is:

[0025]

[0026] Where t is the time point, Δt is the time step, V 1 、V 2 is the flow rate value of unimolecular and bimolecular reactions, V input is the flow value of the input reaction; x i is the isotope distribution with weight i, x i,input represents the isotope distribution of substrate weight i, c is the concentration value, P i represents the single-molecule reaction mapping matrix with weight i, Q i The mapping matrix representing the bimolecular reaction with weight i, T i represents the elimination matrix with intermediate reaction weight i, P i,input is the mapping matrix with input reaction weight i, S1-1 is the first substance before metabolism 13 The number of C atoms marked by C, S2-1 is the number of C atoms marked by C in the second substance before metabolism. 13 C is the number of C atoms labeled.

[0027] At t=0, the mass isotope distribution value of the intermediate metabolite at time t is natural 13 C isotope ratio.

[0028] Furthermore, a metabolic calculation model is constructed using a python package called paddle paddle. The metabolic calculation model has an input layer, a normalization layer, a fully connected layer, a dropout layer, and an output layer, and a grid search algorithm is used to determine model parameters.

[0029] According to the present invention, the calculation and prediction of the metabolic non-steady-state flow change of the biological network during the metabolic process are realized. In the implementation process, the calculation of the metabolic non-steady-state flow change is realized by the differential equation, and the calculation result is used as a training set to construct a model, realize the prediction of the metabolic flow value, and make the prediction result of the calculation model more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of a method for calculating changes in non-steady-state flow of biological network metabolism provided in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The present invention constructs a metabolic calculation model of a biological network metabolic process. In the processing of sample data of the metabolic calculation model, a differential equation is used to calculate the change in metabolic non-steady-state flow, and the calculation result is used as a training set to construct the model. In the present invention, the metabolic calculation model is loaded, and according to the substrate mass isotope distribution, the metabolic reaction flow value at a future time point after the basic time point is predicted with the metabolic time point t as the basic time point.

[0032] The specific implementation of the present invention is described in detail below with reference to the accompanying drawings.

[0033] The method for calculating the non-steady-state flow change of biological network metabolism provided by the present invention is as follows: Figure 1 As shown, the following steps are included:

[0034] Step S100: inputting the total system reaction time T and the time step Δt into the metabolic calculation model;

[0035] The metabolic time point t is accumulated n times according to the time step Δt to form t+n (n is a natural number), and the t+n is less than or equal to the total system reaction time T;

[0036] For example: the total system time is the experimental time, and Δt is 0.001 seconds.

[0037] Step S120: in the metabolic calculation model, obtaining the mass isotope distribution change state of the intermediate metabolite;

[0038] The flux value of the intermediate metabolic reaction of each intermediate metabolite is sampled through the program package Cobrapy, and the change state of the mass isotope distribution of the intermediate metabolite is composed of the change value of the mass isotope distribution of the intermediate metabolite over time and the metabolic time point t; the metabolic calculation model can generate a predicted state, which is composed of the flux value of the metabolic reaction at the metabolic time point t and the metabolic time point t.

[0039] The change state and the predicted state are represented by a change vector, and the data structure is a matrix M×N, where M is the length of the mass isotope distribution of the metabolite and N is the number of time points.

[0040] Since the metabolic process is the reaction process of organisms in nature: that is, under certain conditions, the substances that make up the organisms change. Instruments can monitor the decomposition and merging of various substances in the process of change, the new substances generated by decomposition and merging, and the decrease and increase of various indicators of the original substances. This change process cannot be interfered with artificially, but the substrate metabolic process can be simulated based on the status data that can be monitored by the instrument (such as mass isotope distribution and metabolite concentration), and the metabolic flow can be predicted based on the simulation.

[0041] Therefore, before loading the metabolic calculation model, a metabolic calculation model is constructed, and based on the metabolic data collected in the laboratory, the metabolic process of the substrate is predicted by the metabolic calculation model: the metabolic calculation model simulates the decomposition process of the metabolic process of the substrate and predicts the data corresponding to the metabolic process of the intermediate metabolites.

[0042] The metabolic calculation model is constructed using the existing Python package paddle paddle. The metabolic calculation model has a feedforward neural network with an input layer, a normalization layer, a fully connected layer, a dropout layer, and an output layer. The grid search algorithm is used to determine the model parameters of the neural network model, such as the learning rate and batch size.

[0043] The training data of the metabolic calculation model is obtained based on basic parameters, which include: the mass isotope distribution vector X of the substrate input ; The mass isotope distribution of system substrates is also called labeling degree. In the metabolic calculation model, the mass isotope distribution of substrates will not change over time.

[0044] The metabolic calculation model uses the Cobrapy package to sample and obtain the flow value of the intermediate metabolic reaction of each intermediate metabolite. When extracting, sampling can be performed by a given metabolic flow threshold;

[0045] Some flow values ​​are randomly set to change with time, so that the flow values ​​of intermediate and output reactions change with time. The flow values ​​that change with time and the metabolic time points constitute the vector V_reaction. For other metabolic reactions, the flow values ​​remain constant. On the other hand, the concentration values ​​of all intermediate metabolites change with time, that is, the concentration of each metabolite is the change vector C_metabolite; the initial mass isotope distribution of all intermediate metabolites is the change vector X_metabolite.

[0046] The logical structure of the metabolic calculation model is as follows Figure 1 As shown, it includes: a P111 intermediate metabolite concentration change calculation unit, a P112 intermediate metabolite mass isotope distribution change calculation unit, a P113 sample processing unit and a P114 neural network construction unit; specifically, metabolic data are generated in the P111 intermediate metabolite concentration change calculation unit and the P112 intermediate metabolite mass isotope distribution change calculation unit, and the P113 sample processing unit generates training samples according to the metabolic data, including: a vector C_metabolite of the change of the intermediate metabolite concentration value over time, and a vector X_metabolite of the change of the intermediate metabolite mass isotope distribution; and the P114 neural network construction unit trains and generates a metabolic calculation model according to the training samples.

[0047] The metabolic calculation model is used as a global network to collect the changes in the mass isotope distribution of the metabolites of the substrate over time during the complete metabolic process. Specifically, the data of the metabolite mass isotope distribution at several time points in the entire metabolic process are selected as the feature set of the sample, and the metabolic reaction flow values ​​at several time points in the entire metabolic process are selected as the label set of the sample.

[0048] The process of forming sample data by the metabolic process of intermediate metabolites can be regarded as a local network. The total system reaction time T and time step Δt of the local network are consistent with those of the global network. The calculation process of the local network can be reflected by the P111 intermediate metabolite concentration change calculation unit and the P112 intermediate metabolite mass isotope distribution calculation unit.

[0049] The P111 intermediate metabolite concentration change calculation unit is used to calculate the change in concentration value at time t+1 based on the change in concentration value of the intermediate metabolite at time t, the flow rate of generating the intermediate metabolite at time t, and the flow rate of consuming the intermediate metabolite; in the present invention, each change value is represented by a change vector.

[0050] Specifically, the input parameters of the intermediate metabolite concentration change calculation unit include: the change vector C_metabolite at time t, and the output parameters include: the change vector C_metabolite at time t+1, where time t+1 is t+Δt; when t=0, the intermediate metabolite concentration value is randomly selected from the range of metabolite concentration values ​​measured in the experiment, and the calculation formula of the change vector C_metabolite at time t+1 is as follows:

[0051] C_metabolit t + (∑Vt 生成 -∑Vt 消耗 )×∆t = C_metabolit t+1 ,

[0052] Among them, V t生成 is the sum of the flow rates of the substance generated at time t, V t消耗 It represents the total flow rate of the substance consumed at time t, and Δt is the time step.

[0053] The P112 intermediate metabolite mass isotope distribution calculation unit is used to calculate the change state of the intermediate metabolite mass isotope distribution at time t+1 using a differential equation according to the change value of the intermediate metabolite concentration value at time t and the change value of the intermediate metabolite mass isotope distribution at time t;

[0054] Specifically, the input parameters of the intermediate metabolite mass isotope distribution calculation unit are: the change vector C_metabolite at time t, the change vector X_metabolite at time t, and the output change vector X_metabolite at time t+1, where time t+1 is t+Δt, and the calculation formula is:

[0055]

[0056] Where t is the time point, Δt is the time step, V 1 、V 2 is the flow rate value of unimolecular and bimolecular reactions, V input is the flow value of the input reaction; x i is the isotope distribution with weight i, x i,input represents the isotope distribution of substrate weight i, c is the concentration value, P i represents the single-molecule reaction mapping matrix with weight i, Q i The mapping matrix representing the bimolecular reaction with weight i, T i represents the elimination matrix with intermediate reaction weight i, P i,input is the mapping matrix with input reaction weight i, S1-1 is the first substance before metabolism 13 The number of C atoms marked by C, S2-1 is the number of C atoms marked by C in the second substance before metabolism. 13 C is the number of C atoms labeled.

[0057] Among them, the flow value, isotope distribution, and concentration value are obtained through the conversion of vector V_reaction, change vector X_metabolite, and change vector C_metabolite, respectively.

[0058] The initial value of the intermediate metabolite concentration, that is, when t=0, the intermediate metabolite concentration value at time t is randomly selected from the range of metabolite concentrations measured experimentally.

[0059] The input data of the P113 sample processing unit include: vector V_reaction, vector C_metabolite of the change of the concentration value of the intermediate metabolite over time, vector X_metabolite of the change of the mass isotope distribution of the intermediate metabolite;

[0060] In the training sample generated by the sample processing unit, the feature set includes: change vector C_metabolite, change vector X_metabolite and metabolic time point t;

[0061] In the training samples generated by the sample processing unit, the label set is a vector V_reaction of intermediate metabolic reaction flow values.

[0062] The P114 neural network building unit uses the existing python package paddle paddle to build a feedforward neural network model with an input layer, a normalization layer, a fully connected layer, a dropout layer, and an output layer to train the neural network and predict the flow of metabolic reactions.

[0063] In this step, a metabolic calculation model based on a differential equation is constructed, and the mass isotope distribution of the substrate is input into the metabolic calculation model to predict the changing reaction flow value during the metabolic process.

[0064] Step S120: obtaining the predicted state of the metabolic reaction flow through the metabolic calculation model.

[0065] Through the computational model of the biological network metabolic process constructed by the present invention, in the processing of the computational model sample data, the calculation of the metabolic non-steady-state flow change is realized by using the differential equation. Through the differential equation, the metabolite mass isotope distribution can be effectively calculated when the metabolic flow and metabolite concentration change. According to the present invention, by introducing metabolic non-steady-state flow analysis, the dynamic changes of metabolic reactions in different regions of the tissue can be obtained, and the leap from the spatial distribution of metabolite concentration to the dynamic analysis of metabolic flow is achieved, and the reaction rate changes in the metabolic network can be captured spatially. It improves the understanding of the heterogeneity of metabolic networks of complex tissues (such as brain, tumors, plants, etc.), which helps to describe some complex physiological states.

[0066] The above disclosures are only several specific embodiments of the present invention; however, the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for calculating the non-steady-state flow change of biological network metabolism, characterized in that: The following steps are involved: Loading a metabolic calculation model, wherein the metabolic calculation model is used to predict the flow value of a metabolic reaction at a future time point according to the substrate mass isotope distribution; The total system reaction time T and the time step Δt are input into the metabolic calculation model to obtain the change state of the mass isotope distribution of the intermediate metabolites and the predicted state of the metabolic reaction flow; the change state of the mass isotope distribution of the intermediate metabolites is composed of the change value of the mass isotope distribution of the intermediate metabolites over time and the metabolic time point t; the predicted state is composed of the flow value of the metabolic reaction at the metabolic time point t and the metabolic time point t; Before loading the metabolic calculation model, construct the metabolic calculation model, the metabolic calculation model comprising: an intermediate metabolite concentration change calculation unit, an intermediate metabolite mass isotope distribution change calculation unit, a sample processing unit and a neural network construction unit; generate metabolic data in the intermediate metabolite concentration change calculation unit and the intermediate metabolite mass isotope distribution change calculation unit, generate training samples according to the metabolic data by the sample processing unit, and train and generate the metabolic calculation model by the neural network construction unit according to the training samples; The intermediate metabolite concentration change calculation unit calculates the change value of the concentration value at time t+1 over time according to the change value of the intermediate metabolite concentration at time t over time, the flow rate of generating the intermediate metabolite at time t, and the flow rate of consuming the intermediate metabolite; The intermediate metabolite mass isotope distribution change calculation unit calculates the intermediate metabolite mass isotope distribution change state at time t+1 using a differential equation according to the change value of the intermediate metabolite concentration value over time and the change value of the intermediate metabolite mass isotope distribution over time at time t; The basic parameters of the metabolic calculation model include: the mass isotope distribution vector X of the substrate input ; The metabolic calculation model obtains the flow value of the intermediate metabolic reaction by sampling through the program package Cobrapy; sets part of the flow value to change with time, and defines the flow value that changes with time and the metabolic time point to form a vector V_reaction; The input data of the sample processing unit includes: vector V_reaction, vector C_metabolite of change of concentration value of intermediate metabolites over time, vector X_metabolite of change of mass isotope distribution of intermediate metabolites; in the training samples generated by the sample processing unit, the feature set includes: change vector C_metabolite, change vector X_metabolite and metabolic time point t; in the training samples generated by the sample processing unit, the label set is vector V_reaction; The input parameters of the intermediate metabolite concentration change calculation unit include: the change vector C_metabolite at time t, and the output parameters include: the change vector C_metabolite at time t+1, where time t+1 is t+Δt; the calculation method used by the intermediate metabolite concentration change calculation unit is: C_metabolite t + (∑Vt 生成 -∑Vt 消耗 )×∆t = C_metabolit t+1 , Among them, V t生成 is the sum of the flow rates of the substance generated at time t, V t消耗 It represents the sum of the flow rate of the substance consumed at time t, and Δt is the time step; The input parameters of the intermediate metabolite mass isotope distribution calculation unit are: the change vector C_metabolite at time t, the change vector X_metabolite at time t, and the change vector X_metabolite at time t is output, wherein time t+1 is t+Δt; The calculation method used by the intermediate metabolite mass isotope distribution calculation unit is: Where t is the time point, Δt is the time step, V 1 、V 2 is the flow rate value of unimolecular and bimolecular reactions, V input is the flow value of the input reaction; x i is the isotope distribution with weight i, x i,input represents the isotope distribution of substrate weight i, c is the concentration value, P i represents the single-molecule reaction mapping matrix with weight i, Q i The mapping matrix representing the bimolecular reaction with weight i, T i represents the elimination matrix with intermediate reaction weight i, P i,input is the mapping matrix with input reaction weight i, S1-1 is the first substance before metabolism 13 The number of C atoms marked by C, S2-1 is the number of C atoms marked by C in the second substance before metabolism. 13 C is the number of C atoms labeled.

2. The method according to claim 1, characterized in that The change state and the predicted state are represented by a change vector, and the data structure is a matrix M×N, where M is the length of the mass isotope distribution of the metabolite and N is the number of time points.

3. The method according to claim 1, characterized in that At t = 0, the intermediate metabolite concentration value is randomly selected from the range of experimentally measured metabolite concentration values.

4. The method according to claim 1, characterized in that At t=0, the mass isotope distribution value of the intermediate metabolite at time t is natural 13 C isotope ratio.

5. The method according to claim 1, characterized in that The metabolic time point t is accumulated n times according to the time step Δt to form t+n, and the t+n is less than or equal to the total system reaction time T.

6. The method according to claim 1, characterized in that The metabolic calculation model is constructed by the python package paddlepaddle. The metabolic calculation model has an input layer, a normalization layer, a fully connected layer, a Dropout layer, and an output layer. The model parameters are determined using a grid search algorithm.

Images