Metabolic stimulation-based culture medium optimization method and system

Abstract

The invention discloses a culture medium optimization method and system based on metabolic stimulation. The method comprises the following steps: (1) constructing a relation model of component concentration difference and cell response amount at each culture moment relative to an initial culture moment; (2) taking a preset cell response amount or a weighted value of a plurality of preset cell response amounts as an optimization target, and searching an optimal component concentration difference by adopting a global optimization algorithm; and (3) determining a control quantity according to a preset culture scheme by taking the optimal component concentration difference as a control target, obtaining a culture solution component addition concentration of the culture medium or the culture medium series under the culture scheme by adopting an optimal control algorithm, and developing a culture medium formula of the preset culture scheme. According to the method, synergistic adjustment among components of the culture medium at different culture moments is considered; modeling analysis is carried out on data by adopting machine learning, so that the research and development period of culture medium optimization can be shortened, and repeated tests are reduced. Meanwhile, nonlinear correlation among different components in the culture medium can be considered.

Description

Technical Field

[0001] The present invention belongs to the field of biotechnology, and more specifically, relates to a culture medium optimization method and system based on metabolic stimulation. Background Art

[0002] At present, some intelligent development methods for cell basal culture media have emerged. For example, Chinese patent document CN113450882A provides a method and system for developing basal culture media formula based on artificial intelligence. It applies machine learning algorithms to complex formula optimization processes, builds a high-quality and sufficient sample formula database, selects appropriate machine learning algorithms and optimization algorithms, and recommends the basal culture media formula that is most likely to have a good culture effect in a short time, thereby lowering the threshold for formula development. Another example is Chinese patent document CN113450868A, as an improvement on the former, which further improves the culture effect of developing basal culture media.

[0003] The principle of these culture medium optimization methods is to learn the culture effect of the current culture medium formula on a specific cell line or a series of cell lines, so a large amount of sample data is needed to map the response relationship between the complex components of the culture medium and the specific culture effect of a specific cell line. This principle requires a large sample size, a long time to develop experimental samples, and more importantly, poor generalization effect.

[0004] There are some culture medium development methods that use metabolic information, such as metabolic flux analysis. Although the mutual influence of culture medium components can be analyzed and modeled, the technology involved is complex and requires professional biological metabolism knowledge and equipment support. And it can only be applied to known metabolic pathways and gene regulation, which limits the scope of optimization. Chinese patent document CN114678085A proposes a method for developing feed medium using metabolic parameters, which assumes that the required metabolites are consumed linearly, and only considers the nutrients required for metabolism as material supply, without considering its metabolic stimulation. Therefore, it can only roughly simulate the metabolic consumption process, which is only suitable for the development of feed medium.

[0005] In summary, the current intelligent culture medium development methods based on culture medium formula-culture effect sample learning require a large sample size, a long development cycle, high experimental costs, and poor generalization ability; the methods based on metabolic analysis cover limited metabolic pathways; and the intelligent culture medium development methods based on metabolic parameters are only applicable to feed culture media. Summary of the invention

[0006] In response to the above defects or improvement needs of the prior art, the present invention provides a culture medium optimization method and system based on metabolic stimulation, which aims to simulate the specific process of cell growth in which cell growth is stimulated by the metabolism of culture medium components, and optimize the culture medium of the entire culture process according to preset culture objectives by controlling the concentration of extracellular culture fluid components, thereby solving the technical problems that the prior art cannot simulate metabolic processes, resulting in the need for a large amount of sample data, high cost, or the culture medium development method based on metabolic analysis is not accurate enough and can only be applied to feed culture medium.

[0007] To achieve the above object, according to one aspect of the present invention, a method for optimizing a culture medium based on metabolic stimulation is provided, comprising the following steps:

[0008] (1) constructing a relationship model between the concentration difference of the component at each culture time relative to the initial culture time and the cell response amount; the relationship model uses a regression model to simulate the mapping relationship between the concentration difference of the culture solution component and the cell response amount, and the cell response amount is a parameter of cell metabolism affected by the concentration of the culture solution component;

[0009] (2) using a preset cell response amount or a plurality of preset cell response amount weighted values ​​as the optimization target, and using a global optimization algorithm to search for the optimal component concentration difference of the relationship model between the component concentration difference and the cell response amount obtained in step (1);

[0010] (3) Taking the optimal component concentration difference obtained in step (2) as the control target, determining the control amount according to the preset culture scheme, using the optimal control algorithm to obtain the concentration of the culture fluid components added to the culture medium or culture medium series under the culture scheme, and developing the culture medium formula of the preset culture scheme.

[0011] Preferably, in the culture medium optimization method based on metabolic stimulation, the relationship model between the component concentration difference at each culture time relative to the initial culture time and the cell response amount is constructed according to the following method:

[0012] The experimentally obtained concentration difference of the extracellular culture medium component relative to the initial culture time is used as the training input, the cell response amount is used as the training supervision, the mean square error between the actual cell response amount and the predicted cell response amount is used as the loss value, and the regression model is trained with minimization of the loss value as the optimization goal to obtain a relationship model between the component concentration difference at each culture time relative to the initial culture time and the cell response amount.

[0013] Preferably, in the culture medium optimization method based on metabolic stimulation, the relationship model is preferably a regression model that can be differentiated or approximately differentiated with respect to input variables, such as a multi-layer neural network or Gaussian process regression.

[0014] Preferably, in the culture medium optimization method based on metabolic stimulation, the cell response quantities include: cell density, cell viability and culture effect; the culture effect is the culture target value, including the expression amount of the specified protein product and the acid-base peak and sugar form related to the quality.

[0015] Preferably, the culture medium optimization method based on metabolic stimulation, step (2) is specifically:

[0016] The concentration difference of each component is used as the optimization parameter, and the parameter interval is set to be 0 to the maximum component concentration value given by experience; the preset cell response amount or the weighted value of multiple preset cell response amounts is used as the fitness; global optimization is performed with the maximum fitness as the optimization goal, and the component concentration difference at convergence is output as the optimal component concentration difference; the global optimization algorithm adopts a heuristic optimization algorithm such as a genetic algorithm, simulated annealing, or an ant colony algorithm.

[0017] Preferably, in the culture medium optimization method based on metabolic stimulation, the controlled quantity in step (3) of step (3) includes the extracellular culture fluid component addition concentration vector A(t) at each culture time, the culture medium addition rate or the culture medium addition time; and the culture medium formula is: basal culture medium, feed culture medium, and / or perfusion culture medium formula.

[0018] Preferably, in the culture medium optimization method based on metabolic stimulation, the system equation of the optimal control algorithm in step (3) is obtained as follows: the system equation is fitted using the culture solution component concentration and cell density measurement values ​​for system process simulation, and the system equation is written as:

[0019]

[0020]

[0021] Where VCD(t) represents the viable cell density at the culture time t, which characterizes the current cell culture state; C(t) = {c i (t)} is the vector representing the concentration of the extracellular medium component at the culture time t, c i (t) is the concentration of extracellular culture fluid component i at culture time t, i is the component number; f(C(t)) is the cell growth rate prediction model, whose input is the vector C(t) of the extracellular culture fluid component concentration at culture time t, and the predicted output is the growth rate at culture time t g i (C(t)) unit cell component consumption rate prediction model, whose input is the vector C(t) of the extracellular culture fluid component concentration at culture time t, and whose prediction output is the consumption rate of extracellular culture fluid component i at culture time t;

[0022] (3-1) The cell growth rate prediction model f(C(t)) is obtained according to the following method:

[0023] The cell growth curve is fitted with the vector C(t) of the extracellular culture fluid component concentration obtained experimentally and the viable cell density VCD(t) at the culture time t, and the slope of the cell growth curve at the culture time t is taken as the cell growth rate obtained experimentally. Preferably, a cubic spline function is used for interpolation when fitting the cell growth curve;

[0024] The regression model is trained with the extracellular culture fluid component concentration C(t) and the cell growth rate μ(t) at the culture time t obtained experimentally to obtain the cell growth rate prediction model f(C(t)); the regression model is preferably an autoregressive model;

[0025] (3-2) Unit cell component consumption rate prediction model g i (C(t)) is obtained as follows:

[0026] The cell growth curve is fitted with the experimentally obtained extracellular culture fluid component concentration vector C(t) and the viable cell density VCD(t) at culture time t to calculate the unit cell component consumption rate of component i obtained experimentally. Where t' is the last training time, VCD avg It is the arithmetic mean of the density of living cells from time t′ to time t; preferably, a cubic spline function is used for interpolation when fitting the cell growth curve.

[0027] Preferably, in the culture medium optimization method based on metabolic stimulation, the optimal control algorithm in step (3) adopts a global optimization-based stochastic forward simulation or a model predictive control algorithm;

[0028] The specific steps of optimizing control using global optimization stochastic forward simulation are as follows:

[0029] The vector C(t) of the extracellular culture fluid component concentration is randomly initialized, and the cell culture state at different control value values ​​is simulated by the system equation to obtain the component concentration at the culture time t, and the difference r between it and the optimal component concentration obtained in step (2) is i Compare and calculate to obtain the corresponding cost function;

[0030] For discrete culture schemes, such as basal culture and fed-batch culture, the control amount is the concentration vector of the extracellular culture fluid components added at each culture time A(t) = {a i (t)}; where a i (t) is the concentration of extracellular culture medium component i added to the culture medium at culture time t;

[0031] For a continuous culture scheme, such as perfusion culture, the control amount is the concentration vector of the extracellular culture fluid components added during each culture period A(t) = {a i(t)}, and the rate of addition;

[0032] The cost function of the optimal control algorithm is:

[0033]

[0034] Among them, C i (0) is the concentration of extracellular culture fluid component i at the initial culture time, is the concentration of extracellular culture fluid component i at culture time t obtained by simulation based on the system equation, r i is the optimal component concentration difference obtained in step (2), w j is the weight of extracellular medium component i.

[0035] Use a heuristic optimization algorithm, take the control amount as the optimization variable, and perform global optimization with the minimum cost function as the optimization goal, and output the control amount of the preset training plan when convergence.

[0036] Preferably, the culture medium optimization method based on metabolic stimulation, step (3) specifically comprises the following steps:

[0037] (3-3) Developing basal medium and / or feed medium based on the optimal extracellular culture fluid component addition concentration vector A(t):

[0038] The developing basal culture medium specifically comprises: using the optimal extracellular culture fluid component addition concentration vector at the initial culture time as the basal culture medium component concentration;

[0039] The developing feed medium specifically comprises: adding the optimal extracellular culture fluid component concentration vector A(t) at the culture time t, and determining the feed medium component concentration according to the added volume of the feed medium.

[0040] According to another aspect of the present invention, there is provided a culture medium optimization system based on metabolic stimulation, wherein the system is an electronic device and / or a non-transitory computer-readable storage medium;

[0041] The electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the culture medium optimization method based on metabolic stimulation provided by the present invention when executing the program;

[0042] The non-transitory computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the culture medium optimization method based on metabolic stimulation provided by the present invention are implemented.

[0043] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0044] 1. The present invention takes into account the coordinated adjustment between culture medium components at different culture moments; the use of machine learning to model and analyze data can shorten the research and development cycle of culture medium optimization and reduce repeated experiments. At the same time, it can take into account the nonlinear correlation between different components in the culture medium. The culture medium development cycle is shortened from 6-9 months of culture medium optimization time to 1-2 weeks; the correspondence between the initial culture solution component concentration and the component consumption is learned through the model-based optimal control method, which solves the problem that the concentration of added components cannot be quantified, and at the same time reduces the number of experiments required.

[0045] 2. After the culture medium is optimized, the culture medium can be optimized according to the preset culture response target, such as a significant increase in cell density and protein expression, while the quality of the protein remains unchanged. The prepared culture medium is not affected by batch differences due to its good effect. And the culture medium obtained by this method can predict the upper limit of the culture medium's improvement in cell expression when the culture conditions of other non-culture medium components remain unchanged.

[0046] 3. This method can optimize the medium formula / feed formula using the used medium formula provided by others and the supernatant after culturing cells with the used medium or during the culturing process, which can be used for the medium of others and also as the feed of the medium of others. The optimized medium formula includes but is not limited to the basal medium formula, feed medium formula, and perfusion medium formula. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0048] The method for optimizing culture medium based on metabolic stimulation provided by the present invention comprises the following steps:

[0049] (1) constructing a relationship model between the component concentration difference at each culture time relative to the initial culture time and the cell response amount; the relationship model uses a regression model to simulate the mapping relationship between the culture solution component concentration difference and the cell response amount, the cell response amount is a parameter of cell metabolism affected by the concentration of the culture solution component, including: cell density, cell viability and culture effect; the culture effect is a culture target value, such as the expression amount and quality-related acid-base peaks, sugar forms, etc. of a specified protein product.

[0050] The relationship model between the component concentration difference at each culture time relative to the initial culture time and the cell response amount is preferably constructed according to the following method:

[0051] The experimentally obtained extracellular medium component concentration difference relative to the initial culture time is used as the training input, the cell response is used as the training supervision, the mean square error between the real cell response and the predicted cell response is used as the loss value, and the regression model is trained with the loss value minimized as the optimization goal to obtain the relationship model between the component concentration difference at each culture time relative to the initial culture time and the cell response. The relationship model is preferably a regression model that can be differentiated or approximately differentiated for the input variables, such as a multi-layer neural network or a Gaussian process regression, so as to facilitate the optimization and solution of the subsequent optimization control part.

[0052] (2) taking a preset cell response amount or a plurality of preset cell response amount weighted values ​​as the optimization target, using a global optimization algorithm to search for the optimal component concentration difference of the relationship model between the component concentration difference and the cell response amount obtained in step (1); specifically:

[0053] The concentration difference of each component is used as the optimization parameter, and the parameter range is set from 0 to the maximum component concentration value given by experience; the preset cell response amount or the weighted value of multiple preset cell response amounts is used as the fitness; global optimization is performed with the maximum fitness as the optimization goal, and the component concentration difference at convergence is output as the optimal component concentration difference.

[0054] The global optimization algorithm preferably adopts a heuristic optimization algorithm such as a genetic algorithm, simulated annealing, an ant colony algorithm, and the like.

[0055] (3) using the optimal component concentration difference obtained in step (2) as the control target, determining the control amount according to the preset culture scheme, using the optimal control algorithm to obtain the concentration of the culture solution components added to the culture medium or culture medium series under the culture scheme, and developing the culture medium formula of the preset culture scheme;

[0056] The control quantity includes the extracellular culture fluid component addition concentration vector A(t) at each culture time, the culture medium addition rate or culture medium addition time; the culture medium formula is: basal culture medium, feed culture medium, and / or perfusion culture medium formula.

[0057] The optimal control algorithm preferably adopts a random forward simulation (random shooting) based on global optimization, or a model predictive control algorithm (Model Predictive Control).

[0058] The system equation is obtained as follows: the system equation is fitted using the measured values ​​of the culture medium component concentration and the cell density for system process simulation. The system equation is written as:

[0059]

[0060]

[0061] Where VCD(t) represents the viable cell density at the culture time t, which characterizes the current cell culture state; C(t) = {c i (t)} is the vector representing the concentration of the extracellular medium component at the culture time t, c i (t) is the concentration of extracellular culture fluid component i at culture time t, i is the component number; f(C(t)) is the cell growth rate prediction model, whose input is the vector C(t) of the extracellular culture fluid component concentration at culture time t, and the predicted output is the growth rate at culture time t g i (C(t)) is a unit cell component consumption rate prediction model, whose input is the vector C(t) of the extracellular culture fluid component concentration at culture time t, and its predicted output is the consumption rate of extracellular culture fluid component i at culture time t.

[0062] (3-1) The cell growth rate prediction model f(C(t)) is obtained according to the following method:

[0063] The cell growth curve is fitted with the vector C(t) of the extracellular culture fluid component concentration obtained experimentally and the viable cell density VCD(t) at the culture time t, and the slope of the cell growth curve at the culture time t is taken as the cell growth rate obtained experimentally. Preferably, a cubic spline function is used for interpolation when fitting the cell growth curve;

[0064] The regression model is trained with the extracellular culture fluid component concentration C(t) and the cell growth rate μ(t) at the culture time t obtained experimentally to obtain the cell growth rate prediction model f(C(t)); the regression model is preferably an autoregressive model.

[0065] (3-2) Unit cell component consumption rate prediction model g i (C(t)) is obtained as follows:

[0066] The cell growth curve is fitted with the experimentally obtained extracellular culture fluid component concentration vector C(t) and the viable cell density VCD(t) at culture time t to calculate the unit cell component consumption rate of component i obtained experimentally. Where t' is the last training time, VCD avg is the arithmetic mean of the density of live cells from time t′ to time t; preferably, a cubic spline function is used for interpolation when fitting the cell growth curve;

[0067] The specific steps of optimizing control using global optimization stochastic forward simulation are as follows:

[0068] The vector C(t) of the extracellular culture fluid component concentration is randomly initialized, and the cell culture state at different control value values ​​is simulated by the system equation to obtain the component concentration at the culture time t, and the difference r between it and the optimal component concentration obtained in step (2) is i Compare and calculate to obtain the corresponding cost function;

[0069] For discrete culture schemes, such as basal culture and fed-batch culture, the control amount is the concentration vector of the extracellular culture fluid components added at each culture time A(t) = {a i (t)}; where a i (t) is the concentration of extracellular culture medium component i added to the culture medium at culture time t;

[0070] For a continuous culture scheme, such as perfusion culture, the control amount is the concentration vector of the extracellular culture fluid components added during each culture period A(t) = {a i (t)}, and the rate of addition;

[0071] The cost function of the optimal control algorithm is:

[0072]

[0073] Among them, C i (0) is the concentration of extracellular culture fluid component i at the initial culture time, is the concentration of extracellular culture fluid component i at culture time t obtained by simulation based on the system equation, r i is the optimal component concentration difference obtained in step (2), w i is the weight of the extracellular culture fluid component i, which can actually be taken as the reciprocal of the upper limit of the concentration of the empirical formula component to play a normalization role.

[0074] Use a heuristic optimization algorithm, take the control amount as the optimization variable, and perform global optimization with the minimum cost function as the optimization goal, and output the control amount of the preset training plan when convergence.

[0075] (3-3) Optimizing the culture medium of the preset culture scheme according to the control amount of the preset culture scheme:

[0076] The developing basal culture medium specifically comprises: using the optimal extracellular culture fluid component addition concentration vector at the initial culture time as the basal culture medium component concentration;

[0077] The developing feed medium specifically comprises: adding the optimal extracellular culture fluid component concentration vector A(t) at the culture time t, and determining the feed medium component concentration according to the added volume of the feed medium.

[0078] The development of the perfusion culture medium specifically includes: adding a concentration vector A(t) to the extracellular culture fluid components in each culture period, and determining the concentration of the perfusion culture medium components according to the addition rate.

[0079] The following are examples:

[0080] This embodiment takes a discrete culture scheme as an example, and provides a culture medium optimization method based on metabolic stimulation, comprising the following steps:

[0081] (1) Construct a relationship model between the concentration difference of the components at each culture time relative to the initial culture time and the cell response amount; the culture response amount used in this embodiment is: cell viability, cell density, protein product expression amount, denoted as: Y(t) = {y j (t)};

[0082] The relationship model between the component concentration difference at each culture time relative to the initial culture time and the cell response amount in this embodiment is constructed according to the following method:

[0083] Experimental design, to obtain the concentration of extracellular culture fluid components at each culture time under different initial culture fluid and feed conditions. Each set of experimental data is composed of the measured value of the basic culture fluid component at the initial time C(0) = {c i (0)} and the concentration of extracellular culture fluid components at different culture time t C(t) = {c i (t)} and the cell response quantity Y(t).

[0084] Calculate the consumption of the extracellular culture fluid components from the beginning to the end of the culture time T ΔX T :

[0085] ΔX T =C(T)V(T)-C(0)V(0)

[0086] Or for the case of fed-batch culture

[0087] ΔX T =C(T)V(T)-C(0)V(0)+Xf

[0088] Where Xf = {Xf i} represents the total feed addition concentration during the cultivation process.

[0089] The model data is constructed using the component consumption at each culture time relative to the initial culture time and the corresponding culture response as the output variable Yj.

[0090] The different culture medium formula data obtained from the experiment are divided into training sets and test sets. The prediction model is trained and optimized on the training set, and the model effect is evaluated on the test set.

[0091] This example uses a company's commercial culture medium and feed culture medium, and adds initial additives of different component concentrations for CHO cell culture through experimental design. The cell line used is a company's Dura series cell line. A total of 1088 groups of experiments were conducted (corresponding to different basal and feed culture medium), and the measurement data include the initial and 1, 3, 5, and 7 days of culture medium components (20 amino acids and other vitamins, metal trace elements) and response variables - including cell viability y1(t), cell density y2(t), and protein expression y3(t) on the tth day.

[0092] The input feature data was normalized and then divided into 80% training set and 20% test set. We used the catBoost algorithm based on symmetric decision tree to build a prediction model for the consumption of culture medium components, cell density and expression. The model loss function was defined as mean square error, and the catBoost model was optimized on the training set.

[0093]

[0094] (2) using a preset cell response amount or a plurality of preset cell response amount weighted values ​​as the optimization target, and using a global optimization algorithm to search for the optimal component concentration difference of the relationship model between the component concentration difference and the cell response amount obtained in step (1);

[0095] This embodiment uses a genetic algorithm for global optimization, as follows:

[0096] Genetic algorithm is a heuristic optimization algorithm. The core idea is to search the parameter space by simulating the genetic evolution process. Genetic operations include the following three basic genetic operators: selection, crossover, and mutation. The specific steps are as follows:

[0097] (1) Initialization: Set the evolutionary generation counter n = 0, set the maximum evolutionary generation N, and randomly generate M individuals as the initial population P(O).

[0098] In this embodiment, N=1000 and M=50.

[0099] S1. Take the concentration of each component as the optimization parameter and set the parameter range from 0 to the maximum component concentration value given by experience:

[0100] Element A1 A2 A3 A4 A5 A6 A7 min(mg / L) 350.00 500.00 300.00 200.00 150.00 200.00 100.00 max(mg / L) 1800.00 1000.00 700.00 450.00 400.00 400.00 1000.00

[0101] S2, individual evaluation: calculate the fitness of each individual in the population P(t): the protein expression y3(t) is the fitness;

[0102] S3, selection operation: Apply the selection operator to the group. The purpose of selection is to directly pass on the optimized individuals to the next generation or to generate new individuals through pairing and crossover and then pass on to the next generation. The selection operation is based on the fitness evaluation of individuals in the group.

[0103] S4, crossover operation: Apply the crossover operator to the population. The crossover operator plays a core role in the genetic algorithm. The actual crossover probability is 0.9.

[0104] S5. Mutation operation: Apply the mutation operator to the population. That is, change the gene values ​​at certain loci of the individual strings in the population. After the population P(t) undergoes selection, crossover, and mutation operations, the next generation population P(t+1) is obtained. The mutation probability is 0.1.

[0105] S6. Termination condition judgment: If n=N, the individual with the maximum fitness obtained in the evolution process is output as the optimal solution and the calculation is terminated.

[0106] (3) Taking the optimal component concentration difference obtained in step (2) as the control target, taking the extracellular culture fluid component addition concentration vector A(t) at each culture time as the control quantity, an optimal control algorithm is used to obtain the optimal extracellular culture fluid component addition concentration vector A(t) at different culture times, and developing a basal culture medium and / or a feed medium based on the optimal extracellular culture fluid component addition concentration vector A(t).

[0107] The optimal control algorithm in this embodiment adopts random forward simulation (random shooting) based on global optimization.

[0108] The system equation is obtained as follows: the system equation is fitted using the measured values ​​of the culture medium component concentration and the cell density for system process simulation. The system equation is written as:

[0109]

[0110]

[0111] Wherein, VCD(t) represents the viable cell density at the culture time t, which in this embodiment is the product of the cell viability y1(t) and the cell density y2(t), representing the current cell culture state; C(t)={c i (t)} is the vector representing the concentration of the extracellular medium component at the culture time t, c i (t) is the concentration of extracellular culture fluid component i at culture time t, i is the component number; f(C(t)) is the cell growth rate prediction model, whose input is the vector C(t) of the extracellular culture fluid component concentration at culture time t, and the predicted output is the growth rate at culture time t g i(C(t)) is a unit cell component consumption rate prediction model, whose input is the vector C(t) of the extracellular culture fluid component concentration at culture time t, and its predicted output is the consumption rate of extracellular culture fluid component i at culture time t.

[0112] (3-1) The cell growth rate prediction model f(C(t)) is obtained according to the following method:

[0113] The cell growth curve is fitted with the vector C(t) of the extracellular culture fluid component concentration obtained experimentally and the viable cell density VCD(t) at the culture time t, and the slope of the cell growth curve at the culture time t is taken as the cell growth rate obtained experimentally. Preferably, a cubic spline function is used for interpolation when fitting the cell growth curve;

[0114] The regression model is trained with the extracellular culture fluid component concentration C(t) and the cell growth rate μ(t) at the culture time t obtained experimentally to obtain the cell growth rate prediction model f(C(t)); the regression model is preferably an autoregressive model.

[0115] (3-2) Unit cell component consumption rate prediction model g i (C(t)) is obtained as follows:

[0116] The cell growth curve is fitted with the experimentally obtained extracellular culture fluid component concentration vector C(t) and the viable cell density VCD(t) at culture time t to calculate the unit cell component consumption rate of component i obtained experimentally. Where t' is the last training time, VCD avg is the arithmetic mean of the density of live cells from time t′ to time t; preferably, a cubic spline function is used for interpolation when fitting the cell growth curve;

[0117] For extracellular medium component i, the vector C(t) of the extracellular medium component concentration obtained experimentally and the unit cell component consumption rate q i Train the regression model to obtain the unit cell component consumption rate prediction model g i (C(t)); the regression model is preferably an autoregressive model.

[0118] The specific steps of optimizing control using global optimization stochastic forward simulation are as follows:

[0119] The vector C(t) of the extracellular culture fluid component concentration is randomly initialized, and the system equation is used to simulate the addition of the extracellular culture fluid component concentration vector A(t) = {a i (t)} is added to the culture solution to obtain the component concentration at the culture time t, and the difference r between it and the optimal component concentration obtained in step (2) is iCompare and calculate to get the corresponding cost function; where a i (t) is the concentration of extracellular culture fluid component i added to the culture fluid at culture time t; the cost function of the optimal control algorithm is:

[0120]

[0121] Among them, C i (0) is the concentration of extracellular culture fluid component i at the initial culture time, is the concentration of extracellular culture fluid component i at culture time t obtained by simulation based on the system equation, r i is the optimal component concentration difference obtained in step (2), w i is the weight of the extracellular culture fluid component i, which can actually be taken as the reciprocal of the upper limit of the concentration of the empirical formula component to play a normalization role.

[0122] A heuristic optimization algorithm is used to perform global optimization with the concentration vector A(t) of the extracellular culture fluid components added at each culture time as the optimization variable and the minimum cost function as the optimization goal. The concentration vector A(t) of the extracellular culture fluid components added at each culture time at the time of convergence is output.

[0123] (3-3) Developing basal medium and / or feed medium based on the optimal extracellular culture fluid component addition concentration vector A(t):

[0124] The developing basal culture medium specifically comprises: using the optimal extracellular culture fluid component addition concentration vector at the initial culture time as the basal culture medium component concentration;

[0125] The developing feed medium specifically comprises: adding the optimal extracellular culture fluid component concentration vector A(t) at the culture time t, and determining the feed medium component concentration according to the added volume of the feed medium.

[0126] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for optimizing culture medium based on metabolic stimulation, characterized in that: The following steps are involved: (1) constructing a relationship model between the concentration difference of the component at each culture time relative to the initial culture time and the cell response amount; the relationship model uses a regression model to simulate the mapping relationship between the concentration difference of the culture solution component and the cell response amount, and the cell response amount is a parameter of cell metabolism affected by the concentration of the culture solution component; (2) using a preset cell response amount or a plurality of preset cell response amount weighted values ​​as the optimization target, and using a global optimization algorithm to search for the relationship model between the component concentration difference and the cell response amount obtained in step (1) to obtain an optimal component concentration difference for the preset culture target; (3) Taking the optimal component concentration difference obtained in step (2) as the control target, determining the control amount according to the preset culture scheme, using the optimal control algorithm to obtain the concentration of the culture fluid components added to the culture medium or culture medium series under the culture scheme, and developing the culture medium formula of the preset culture scheme.

2. The method for optimizing culture medium based on metabolic stimulation according to claim 1, characterized in that: The relationship model between the component concentration difference at each culture time relative to the initial culture time and the cell response amount is constructed according to the following method: The experimentally obtained concentration difference of the extracellular culture medium component relative to the initial culture time is used as the training input, the cell response amount is used as the training supervision, the mean square error between the actual cell response amount and the predicted cell response amount is used as the loss value, and the regression model is trained with minimization of the loss value as the optimization goal to obtain a relationship model between the component concentration difference at each culture time relative to the initial culture time and the cell response amount.

3. The method for optimizing culture medium based on metabolic stimulation according to claim 2, characterized in that: The relationship model is preferably a regression model that can be differentiated or approximately differentiated with respect to input variables, such as a multi-layer neural network or Gaussian process regression.

4. The method for optimizing culture medium based on metabolic stimulation according to claim 1, wherein: The cell response includes: cell density, cell viability and culture effect; the culture effect is the culture target value, including the expression amount of the designated protein product and the acid-base peak and sugar form related to the quality.

5. The method for optimizing culture medium based on metabolic stimulation according to claim 1, characterized in that: Step (2) is specifically as follows: The concentration difference of each component is used as the optimization parameter, and the parameter interval is set from 0 to the maximum component concentration value given by experience; the preset cell response amount or the weighted value of multiple preset cell response amounts is used as the fitness; the global optimization is performed with the maximum fitness as the optimization goal, and the component concentration difference at the time of convergence is output as the optimal component concentration difference; The global optimization algorithm adopts a heuristic optimization algorithm such as a genetic algorithm, simulated annealing, or an ant colony algorithm.

6. The method for optimizing culture medium based on metabolic stimulation according to claim 1, characterized in that: The control quantity in step (3) includes the extracellular culture fluid component addition concentration vector A(t) at each culture time, the culture medium addition rate or the culture medium addition time; the culture medium formula is: basal culture medium, feed culture medium, and / or perfusion culture medium formula.

7. The method for optimizing culture medium based on metabolic stimulation according to claim 1, characterized in that: The system equation of the optimal control algorithm in step (3) is obtained as follows: the system equation is fitted using the measured values ​​of the culture solution component concentration and the cell density for system process simulation, and the system equation is written as: VCD(t)=VCD(0)exp(∫0 t f(C(t))dt) Where VCD(t) represents the viable cell density at the culture time t, which characterizes the current cell culture state; C(t) = {c i (t)} is the vector representing the concentration of the extracellular medium component at the culture time t, c i (t) is the concentration of extracellular culture fluid component i at culture time t, i is the component number; f(C(t)) is the cell growth rate prediction model, whose input is the vector C(t) of the extracellular culture fluid component concentration at culture time t, and the predicted output is the growth rate at culture time t g i (C(t)) unit cell component consumption rate prediction model, whose input is the vector C(t) of the extracellular culture fluid component concentration at culture time t, and whose prediction output is the consumption rate of extracellular culture fluid component i at culture time t; (3-1) The cell growth rate prediction model f(C(t)) is obtained according to the following method: The cell growth curve is fitted with the vector C(t) of the extracellular culture fluid component concentration obtained experimentally and the viable cell density VCD(t) at the culture time t, and the slope of the cell growth curve at the culture time t is taken as the cell growth rate obtained experimentally. Preferably, a cubic spline function is used for interpolation when fitting the cell growth curve; The regression model is trained with the extracellular culture fluid component concentration C(t) and the cell growth rate μ(t) at the culture time t obtained experimentally to obtain the cell growth rate prediction model f(C(t)); the regression model is preferably an autoregressive model; (3-2) Prediction model for unit cell component consumption rate g i (C(t)) is obtained as follows: The cell growth curve is fitted with the experimentally obtained extracellular culture fluid component concentration vector C(t) and the viable cell density VCD(t) at culture time t to calculate the unit cell component consumption rate of component i obtained experimentally. Where t' is the last training time, VCD avg It is the arithmetic mean of the density of living cells from time t′ to time t; preferably, a cubic spline function is used for interpolation when fitting the cell growth curve.

8. The method for optimizing culture medium based on metabolic stimulation according to claim 1, characterized in that: The optimal control algorithm in step (3) adopts a global optimization-based stochastic forward simulation or a model predictive control algorithm; The specific steps of optimizing control using global optimization stochastic forward simulation are as follows: The vector C(t) of the extracellular culture fluid component concentration is randomly initialized, and the cell culture state at different control value values ​​is simulated by the system equation to obtain the component concentration at the culture time t, and the difference r between it and the optimal component concentration obtained in step (2) is i Compare and calculate to obtain the corresponding cost function; For discrete culture schemes, such as basal culture and fed-batch culture, the control amount is the concentration vector of the extracellular culture fluid components added at each culture time A(t) = {a i (t)}; where a i (t) is the concentration of extracellular culture medium component i added to the culture medium at culture time t; For a continuous culture scheme, such as perfusion culture, the control amount is the concentration vector of the extracellular culture fluid components added during each culture period A(t) = {a i (t)}, and the rate of addition; The cost function of the optimal control algorithm is: Among them, C i (0) is the concentration of extracellular culture fluid component i at the initial culture time, is the concentration of extracellular culture fluid component i at culture time t obtained by simulation based on the system equation, r i is the optimal component concentration difference obtained in step (2), w i is the weight of extracellular culture fluid component i. Use a heuristic optimization algorithm, take the control amount as the optimization variable, and minimize the cost function as the optimization goal to perform global optimization, and output the control amount of the preset training plan when convergence.

9. The method for optimizing culture medium based on metabolic stimulation according to claim 1, wherein: Step (3) specifically includes the following steps: Optimize the culture medium of the preset culture scheme according to the control volume of the preset culture scheme: The developing basal culture medium specifically comprises: using the optimal extracellular culture fluid component addition concentration vector at the initial culture time as the basal culture medium component concentration; The developing feed medium specifically comprises: adding the optimal extracellular culture fluid component concentration vector A(t) at the culture time t, and determining the feed medium component concentration according to the added volume of the feed medium. The development of the perfusion culture medium specifically includes: adding a concentration vector A(t) to the extracellular culture fluid components in each culture period, and determining the concentration of the perfusion culture medium components according to the addition rate.

10. A culture medium optimization system based on metabolic stimulation, characterized in that: The system is an electronic device and / or a non-transitory computer-readable storage medium; The electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the culture medium optimization method based on metabolic stimulation according to any one of claims 1 to 9 are implemented; The non-transitory computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the culture medium optimization method based on metabolic stimulation as described in any one of claims 1 to 9 are implemented.

Images