Learned model, estimation device, and estimation method

Abstract

Provided are: a learned model, with which index data relating to the kinetics of contents in a container can be evaluated with high accuracy; an estimation device; and an estimation method. This learned model is trained using, as an explanatory variable, at least one of data relating to the shape of a stirrer provided in a container, data relating to the shape of a sparger provided in the container, data relating to the shape of the container, and data relating to the physical properties, the composition, the chemical properties, the biochemical properties or the concentration of contents accommodated in the container, and also using, as an objective variable, a mass transfer capacity coefficient (kLa), a stirring power (P / V), a respiratory rate, the concentration of bacterial bodies, the concentration of a product from the bacterial bodies and cells, or a titer.

Description

Trained model, estimation device, and estimation method

[0001] The present invention relates to a trained model, an estimation device, and an estimation method.

[0002] Devices known as stirred tanks and bubble columns are used as reactors that dissolve components in gas (such as oxygen) into the liquid inside the container by blowing (aerating) a gas into the liquid inside the container or by vibrating the container, causing the components in the gas to dissolve in the liquid and undergo a chemical reaction with the components contained in the liquid, or as culture tanks that supply oxygen to microorganisms or animal cells in the liquid or remove carbon dioxide and other gases to cultivate them, and are used in fields such as petrochemistry, chemistry, pharmaceuticals, and life sciences.

[0003] In such stirred vessels and bubble columns that involve gas aeration, the gas aerated into the liquid disperses in the liquid as bubbles or gas masses. However, the dissolution of gas components from the gas into the liquid or the transfer of dissolved components from the liquid to the gas (mass transfer) occurs at the gas-liquid interface (gas-liquid interface) between these bubbles and gas masses, and the mass transfer rate depends on the total gas-liquid interfacial area and the dispersion state of the bubbles and gas masses in the liquid. To maximize the mass transfer rate, the equipment is designed to increase the gas-liquid interfacial area and to disperse the bubbles and gas masses evenly throughout the liquid.

[0004] A well-known indicator of this is the volumetric mass transfer coefficient kL*a (hereinafter referred to as kL), which is calculated by multiplying the gas-liquid interfacial area concentration a (gas-liquid interfacial area per unit liquid volume) by kL, which is an indicator of the diffusion rate of gas components into the liquid (see, for example, Patent Document 1). Particularly when scaling up from a small laboratory scale to a large commercial scale, the design is considered so that the kL of the commercial equipment is equivalent to that of the laboratory scale. Similarly, the stirring power per unit liquid volume (P / V), respiration rate, bacterial and cell concentrations, and the concentrations and titers of products from bacterial and cell masses are also indicators that are considered so that the commercial scale is equivalent to that of the laboratory scale.

[0005] Japanese Patent Application Laid-Open No. 2001-75947

[0006] The kLa and P / V of stirred tanks and bubble columns are evaluated using correlation equations established based on experimental results and physical considerations. However, for example, the kLa in a stirred tank is significantly affected by the operating conditions of the agitator, fluid properties, and agitation shape, making it difficult to accurately evaluate the kLa under all conditions using correlation equations. Furthermore, there are no established correlation equations for evaluating the respiration rate, bacterial and cell concentrations, and the concentrations and titers of products from bacterial and cell products, etc., based on the shape of the agitator or sparger, the shape of the vessel, and the physical properties of the contents of the tank, without relying on actual operation of the equipment. This makes it difficult to accurately evaluate these parameters.

[0007] An object of the present invention is to provide a trained model, an estimation device, and an estimation method that are capable of accurately evaluating index data related to the dynamics of the contents of a container, such as the above-mentioned kLa, P / V, respiration rate, bacterial and cell concentrations, and concentrations and titers of products from bacterial and cell concentrations.

[0008] According to the present invention, there is provided a trained model having the following configuration: [1] A trained model that uses at least one of data relating to the shape of a stirrer provided in a vessel, data relating to the shape of a sparger provided in the vessel, data relating to the shape of the vessel, and data relating to the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the vessel as explanatory variables, and that has been trained using a volumetric mass transfer coefficient (kLa), a stirring power (P / V), a respiration rate, bacterial body and cell concentration, and the concentration or titer of a product from bacterial bodies and cells as objective variables.

[0009] According to the present invention, at least one of data relating to the shape of the agitator, data relating to the shape of the sparger, data relating to the shape of the container, and data relating to the physical properties, composition, chemical properties, or biochemical properties of the contents (explanatory variables) can be input into a trained model, and the output target variables can be obtained, such as the mass transfer capacity coefficient (kLa), the stirring power (P / V), the respiration rate, the bacterial mass and cell concentration, and the concentration or titer of the product from the bacterial mass and cells. This makes it possible to evaluate index data relating to the dynamics of the contents of the container with greater accuracy than conventional techniques.

[0010] Various embodiments of the present invention are exemplified below. The embodiments described below can be combined with each other. [2] The trained model described in [1], trained using data related to the shape of the agitator, data related to the shape of the sparger, data related to the shape of the container, and data related to the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container as explanatory variables. [3] The trained model described in [1] or [2], wherein the data related to the shape of the agitator includes data related to the shapes of agitating blades constituting the agitator, and the data related to the shapes of the agitating blades and the sparger include data in which a numerical label is assigned for each type of shape, or include quantified data in which a first numerical value is assigned to an area occupied by the agitating blades and the sparger in three-dimensional space and a second numerical value is assigned to an area not occupied by the agitating blades and the sparger in three-dimensional space. [4] An estimation device comprising an input unit and a dependent variable acquisition unit, wherein the input unit inputs the explanatory variables to the trained model described in any one of [1] to [3], and the dependent variable acquisition unit acquires the dependent variable output from the trained model when the explanatory variables are input to the trained model. [5] The estimation device described in [4], further comprising an explanatory variable acquisition unit, wherein the explanatory variable acquisition unit acquires the value of the explanatory variable input to the trained model when the value of the dependent variable output from the trained model becomes a predetermined value. [6] The estimation device described in [5], wherein, of first data and second data constituting the explanatory variables, the value of the first data is a fixed value and the value of the second data is a variable value, and the explanatory variable acquisition unit acquires the value of the second data input to the trained model when the value of the dependent variable output from the trained model becomes a predetermined value. [7] An estimation method, comprising: inputting the explanatory variables into a trained model described in any one of [1] to [3]; and, when the explanatory variables are input into the trained model, acquiring the objective variable output from the trained model.

[0011] According to the present invention, index data relating to the dynamics of the contents of a container can be evaluated with high accuracy.

[0012] It is a block diagram showing the functional configuration of an estimation device according to an embodiment of the present invention. It is a flowchart showing an example of estimation processing according to an embodiment of the present invention. It is an image diagram showing changes in values ​​of output data according to the number of days of culture.

[0013] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described with reference to the accompanying drawings, in which the same reference numerals denote the same or similar components.

[0014] (Configuration of Estimation Device 10) FIG. 1 is a block diagram showing the functional configuration of the estimation device 10 according to this embodiment. Using a trained model, the estimation device 10 estimates (acquires) index data related to the dynamics of the contents of a container, such as kLa (volumetric mass transfer coefficient), P / V (agitation power per unit liquid volume), respiration rate, bacterial and cell concentrations, and concentrations or titers of products from bacterial and cell cultures. Here, mass transfer refers to the dissolution of components in a gas (e.g., oxygen) into a liquid or the transfer of dissolved components from a liquid to a gas by blowing (aerating) a gas into a liquid in a container or by vibrating the container in a stirred tank or bubble column that involves gas aeration.

[0015] Stirred tanks and bubble columns are used as reactors for dissolving components in a gas into a liquid and causing a chemical reaction with the components contained in the liquid, or as culture tanks for culturing microorganisms, animal cells, etc. in a liquid by supplying oxygen, etc., or degassing carbon dioxide, etc., and are applied in the fields of petrochemistry, chemistry, pharmaceuticals, life sciences, etc.

[0016] 1, the estimation device 10 includes a trained model storage unit 1, an operation reception unit 2, an explanatory variable input unit 3, a target variable acquisition unit 4, a determination unit 5, an explanatory variable acquisition unit 6, and a display unit 7. The explanatory variable input unit 3 functions as the "input unit" of the present invention.

[0017] Each component of the estimation device 10 may be implemented by software or hardware. When implemented by software, various functions can be realized by a CPU executing a computer program. The program may be stored in an internal storage unit or a computer-readable non-transitory recording medium. Alternatively, the program may be read from an external storage unit and implemented using so-called cloud computing. When implemented by hardware, the components may be implemented by various circuits such as an ASIC, an FPGA, or a DRP (Dynamically Reconfigurable Processor). This embodiment deals with various pieces of information and concepts that encompass them. These are represented by high or low signal values ​​as a binary bit set consisting of 0 or 1, and communication and calculations can be performed using the above software or hardware aspects.

[0018] The trained model memory unit 1 is a database that stores the kLa model 1A, P / V model 1B, respiration rate model 1C, bacterial body and cell concentration model 1D, and titer model 1E as trained models trained by a machine learning unit (not shown).

[0019] The machine learning unit performs machine learning using training data (teacher data) stored in a training data storage unit (not shown). That is, the machine learning unit inputs multiple sets of training data into the training model, and generates a trained model by having the training model learn the correlation between the input data and output data that make up the training data through machine learning (so-called supervised learning). For example, a CNN (Convolutional Neural Network) or the like is used for the trained model. Furthermore, for example, deep learning or the like is used as the learning algorithm for machine learning.

[0020] In this embodiment, the trained model is described as a CNN; however, the trained model is not limited to a CNN, and may be trained using any learning algorithm, such as a neural network other than a CNN, a Support Vector Machine (SVM), a Bayesian network, or a regression tree.

[0021] The kLa model 1A is a trained model that uses, as input data (explanatory variables) constituting the training data, for example, data regarding the shape of the agitator installed in the container, data regarding the shape of the sparger installed in the container, data regarding the shape of the container, and data regarding the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container, and uses kLa (mass transfer capacity coefficient) as output data (objective variable) constituting the training data.

[0022] Examples of data related to the shape of the agitator include data related to the shape of the agitator blades that make up the agitator, such as the type of agitator blade, the agitator blade diameter, and the number of agitator blades, which are set according to the shape of the agitator blade (paddle (inclined / not inclined), disk turbine (inclined / not inclined), propeller, anchor, etc.). The data related to the shape of the agitator blade includes data with a numerical label attached to each type of shape, or data quantified by assigning a first numerical value (e.g., 1) to the area occupied by the agitator blade in three-dimensional space and a second numerical value (e.g., 0) to the area not occupied by the agitator blade in three-dimensional space. The shape data of the agitator blade quantified by the first numerical value and the second numerical value can be expressed as a single numerical value, for example, by taking the average value and standard deviation for the entire device, and can be used as an explanatory variable in a learning model. It is also possible to combine the method of assigning a numerical label to each type of agitator blade shape with the method of quantifying by assigning a first numerical value (e.g., 1) to the area that the agitator blade occupies in three-dimensional space and a second numerical value (e.g., 0) to the area that the agitator blade does not occupy in three-dimensional space. In this case, by assigning a numerical label to each type of agitator blade shape (paddle (inclined / not inclined), disk turbine (inclined / not inclined), propeller, anchor, etc.) and further expressing the presence or absence of an agitator blade occupying three-dimensional space numerically, the shape of the agitator blade can be expressed more precisely as numerical data.

[0023] Examples of data related to the shape of a sparger provided in a vessel include the type of aeration sparger, the diameter of the aeration sparger, the diameter of the aeration sparger holes, and the number of the aeration sparger holes, which are set according to the shape of the sparger (ring-shaped, cylindrical, circular). The data related to the shape of the sparger includes data in which a numerical label is attached for each type of shape, or includes data quantified by assigning a first numerical value (e.g., 1) to an area occupied by the sparger in three-dimensional space and a second numerical value (e.g., 0) to an area not occupied by the sparger in three-dimensional space.

[0024] The data on the shape of the container is data on the shape of the container itself that contains the contents, and includes, for example, the type of container shape (cylindrical, rectangular, etc.), the dimensions of the container (inner diameter (in the case of a cylinder), width (in the case of a rectangle), height, etc.), and the presence or absence of a baffle (in the case of a mixing vessel).

[0025] Data regarding the physical properties, composition, chemical properties, biochemical properties, or concentrations of the contents includes, for example, the physical properties, composition, chemical properties, and biochemical properties specific to the materials contained in the contents, the physical properties of the entire contents within the container, and the concentrations of each material contained in the contents.

[0026] The physical properties inherent to the material contained in the contents include, for example, the density, viscosity, elasticity, non-Newtonianity, etc. inherent to the material. The composition inherent to the material contained in the contents includes, for example, the elements constituting the material and their contents, the ions dissolved in the material and their contents, the presence or absence of surfactants, the sugar content, the amino acid content, the vitamin content, the inorganic salt content, the lipid component content, and the trace metal content. The chemical properties inherent to the material contained in the contents include, for example, the reaction activity, activation energy, reaction rate, reaction rate constant, etc. in the chemical reaction of the material. The biochemical properties inherent to the material contained in the contents include, for example, the presence or absence of mycelium formation (if the material is a cell line), the doubling time, and the growth factors contained (if the material is a culture medium). The materials contained in the contents are, for example, culture medium / culture solution and cells. In this case, the physical properties specific to the materials contained in the contents, such as the fluid properties of the culture medium / culture solution (density, viscosity, non-Newtonianity, etc.), culture medium composition (presence or absence of surfactants, type of growth factor, concentration of sugars contained in the culture medium, concentration of amino acids contained in the culture medium, concentration of vitamins contained in the culture medium, concentration of inorganic salts contained in the culture medium, concentration of lipid components contained in the culture medium, concentration of trace metals contained in the culture medium, etc.), and type of cell line / strain (mass, viscosity, elasticity, size, shape, presence or absence of mycelium formation, doubling time, etc.), are used as explanatory variables for the learning model. The physical properties, composition, chemical properties, and biochemical properties specific to the materials contained in the contents are parameters that do not change even after the device containing the contents is operated, or parameters whose values ​​under operating conditions such as temperature and pressure are known.

[0027] The physical properties of the entire contents in the container include, for example, the temperature, osmotic pressure, oxidation-reduction potential (ORP), and pH of the contents. The concentrations of each material contained in the contents include, for example, the seeding amount and cell concentration (when the contents are culture medium and cells). Data regarding the physical properties of the entire contents in the container used as explanatory variables of the learning model may be initial values ​​before the container starts operating, predicted values ​​obtained by simulating the operation of the container after the container starts operating (for example, computational fluid dynamics simulation, e.g., CFD (Computational Fluid Dynamics) simulation), or actual measured values ​​obtained continuously after the container starts operating. For example, the cell concentration and pH may be set as initial values ​​input by the user, as predicted values ​​(mean value within the container + standard deviation value) obtained by CFD (Computational Fluid Dynamics) simulation, or as mean value within the device + standard deviation value obtained by a multipoint measurement method after operation has started.

[0028] The training data for the trained model preferably includes data on the physical properties, composition, chemical properties, biochemical properties, or concentrations of the contents contained in the container, particularly the physical properties, composition, chemical properties, and biochemical properties specific to the materials contained in the contents, as well as the physical properties of the entire contents in the container and the initial values ​​of the concentrations of each material contained in the contents. Because these data can be obtained without operating the actual device, they can be easily acquired as explanatory variables for the trained model to predict the performance of the actual device in device design. More preferably, in addition to the physical properties, composition, chemical properties, and biochemical properties specific to the materials contained in the contents, the physical properties of the entire contents in the container, and the initial values ​​of the concentrations of each material contained in the contents, the training data may further include values ​​predicted by simulation for the physical properties of the entire contents in the container and the concentrations of each material contained in the contents. Including not only the initial values ​​of these values ​​but also the values ​​after device operation as explanatory variables allows for more accurate estimation of the value of the target variable.

[0029] In another example, in addition to the physical properties, composition, chemical properties, and biochemical properties specific to the materials contained in the contents, the initial values ​​of the physical properties of the entire contents in the container and the concentrations of each material contained in the contents, and values ​​predicted by simulation of the physical properties of the entire contents in the container and the concentrations of each material contained in the contents, it is also preferable to include measured values ​​of the physical properties of the entire contents in the container and the concentrations of each material contained in the contents obtained during actual operation of the device.Since the physical properties of the entire contents in the container and the concentrations of each material contained in the contents are values ​​that change from moment to moment as the device operates, by inputting these data as explanatory variables during operation, the trained model can be used to evaluate actual operation.

[0030] The kLa model 1A may be trained using the gas superficial velocity, aeration rate, turbulent energy dissipation rate, gas holdup, bubble diameter, dissolved oxygen (DO) control method (upper and lower limits, etc.), agitator rotation speed, culture medium input / discharge conditions, shear stress, and culture medium feed rate / perfusion rate (in the case of fed-batch culture / perfusion culture) as input data (explanatory variables) constituting the training data, and kLa (mass transfer capacity coefficient) as output data (objective variable) constituting the training data. Configuring the kLa model 1A to include one or more of these parameters as input data can further improve the prediction accuracy of kLa.

[0031] The gas superficial velocity is a set value calculated from the aeration rate and vessel diameter entered by the user. The aeration rate, rotation speed, dissolved oxygen control method, and fed-batch rate / perfusion rate (culture medium input / output conditions) are set by values ​​entered by the user. The turbulent energy dissipation rate is a value theoretically calculated from P / V (agitation power) or the in-apparatus mean value + standard deviation value obtained by multi-point measurement techniques in CFD simulations or experiments. The gas holdup is a value calculated from the liquid level difference with / without aeration or the in-apparatus mean value + standard deviation value obtained by multi-point measurement techniques in CFD simulations or experiments. The bubble diameter is a value measured using photographic images, etc., or the in-apparatus mean value + standard deviation value obtained by multi-point measurement techniques in CFD simulations or experiments. The shear stress is set by a value entered by the user or the in-apparatus mean value + standard deviation value obtained by multi-point measurement techniques in CFD simulations or experiments. The explanatory variables of the kLa model 1A may include kLa approximately estimated using a conventional approximation formula that approximately derives kLa. By including this as an explanatory variable, the kLa model 1A can be evaluated taking into account the correlation between the kLa estimated using the approximation formula and the kLa output by the model 1A.

[0032] The learning data for kLa, which is the objective variable of the kLa model 1A, is acquired by a known method. For example, kLa may be determined experimentally using an actual device (e.g., a static method in which changes in the dissolved oxygen concentration in the liquid are measured using an oxygen electrode installed in a container and the increase curve of the dissolved oxygen concentration is analyzed), or may be determined by CFD simulation. Preferably, the learning data is composed of data obtained experimentally using the actual device. However, if there is insufficient data on the actual device, the learning data may include data obtained by CFD simulation in addition to data obtained experimentally using the actual device. Note that, because the trained model according to the present invention is a model that replaces conventional approximation equations for determining kLa, it is preferable that kLa determined using conventional correlation equations that do not involve experiments using an actual device is not included in the learning data for the objective variable of the kLa model 1A.

[0033] P / V model 1B is a trained model that uses, for example, data regarding the shape of the agitator installed in the container and data regarding the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container as input data (explanatory variables) that constitute the training data, and kLa (agitation power per unit liquid volume) as output data (objective variable) that constitutes the training data.

[0034] Examples of data relating to the shape of the agitator include data relating to the shape of the agitator blades that make up the agitator, such as the power number and agitator blade diameter that are set according to the shape of the agitator blades. The data relating to the shape of the agitator blades includes data in which a numerical label is attached for each type of shape, or data quantified by assigning a first numerical value (e.g., 1) to the area occupied by the agitator blades in three-dimensional space and a second numerical value (e.g., 0) to the area not occupied by the agitator blades in three-dimensional space. Examples of data relating to the physical properties, composition, chemical properties, or biochemical properties of the contents include the culture medium density that is set according to the culture medium.

[0035] The P / V model 1B may be trained using data regarding the shape of a sparger installed in the vessel, data regarding the shape of the vessel, operating parameters of the agitator, or aeration conditions as input data (explanatory variables) constituting the training data, and the P / V as output data (target variable) constituting the training data. Here, an example of the operating parameters of the agitator is the rotation speed set by a value input by the user. An example of the aeration conditions is the aeration flow rate set by a value input by the user. The explanatory variables of the P / V model 1B may include a P / V approximately estimated using a conventional approximation formula (such as the Michel-Miller empirical formula) that approximately derives the P / V. By including this as an explanatory variable, the P / V model 1B can be evaluated taking into account the correlation between the P / V estimated by the approximation formula and the P / V output by the model 1B.

[0036] The learning data for P / V, which is the objective variable of P / V model 1B, is obtained by a known method. For example, the stirring power can be calculated by measuring the electric power required to operate the stirrer in an actual machine and dividing this by the volume of the container. As with kLa model 1A, it is preferable that P / V calculated using a conventional correlation equation in an actual machine not be included in the learning data for the objective variable of P / V model 1B.

[0037] The respiration rate model 1C is a trained model that uses, for example, data regarding the shape of the agitator installed in the container, data regarding the shape of the sparger installed in the container, data regarding the shape of the container, and data regarding the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container as input data (explanatory variables) that constitute the training data, and trains the respiration rate as output data (objective variable) that constitutes the training data.

[0038] The data relating to the shape of the agitator is, for example, data relating to the shape of the agitator blades constituting the agitator, including the type of agitator blade, the agitator blade diameter, and the number of agitator blades that are set according to the shape of the agitator blade. The data relating to the shape of the agitator blade includes data in which a numerical label is attached for each type of shape, or includes data quantified by assigning a first numerical value (e.g., 1) to the area occupied by the agitator blade in three-dimensional space and a second numerical value (e.g., 0) to the area not occupied by the agitator blade in three-dimensional space.

[0039] Examples of data related to the shape of a sparger provided in the vessel include the type of aeration sparger, the diameter of the aeration sparger, the diameter of the aeration sparger holes, and the number of holes for the aeration sparger, which are set according to the shape of the sparger. The data related to the shape of the sparger includes data in which a numerical label is attached for each type of shape, or includes data quantified by assigning a first numerical value (e.g., 1) to an area occupied by the sparger in three-dimensional space and a second numerical value (e.g., 0) to an area not occupied by the sparger in three-dimensional space.

[0040] The data on the shape of the container is data on the shape of the container itself that contains the contents, and includes, for example, the type of container shape (cylindrical, rectangular, etc.), the dimensions of the container (inner diameter (in the case of a cylinder), width (in the case of a rectangle), height, etc.), and the presence or absence of a baffle (in the case of a mixing vessel).

[0041] Data regarding the physical properties, composition, chemical properties, biochemical properties, or concentrations of the contents includes, for example, the physical properties, composition, chemical properties, and biochemical properties specific to the materials contained in the contents, the physical properties of the entire contents within the container, and the concentrations of each material contained in the contents.

[0042] The respiration rate model 1C may be trained using the gas superficial velocity, aeration rate, turbulent energy dissipation rate, gas holdup, bubble diameter, dissolved oxygen (DO) control method (upper and lower limits, etc.), agitator rotation speed, shear stress, and culture medium feed rate / perfusion rate (in the case of fed-batch culture / perfusion culture) as input data (explanatory variables) constituting the training data, and the respiration rate as output data (target variable) constituting the training data. Configuring the respiration rate model 1C to include one or more of these parameters as input data can further improve the accuracy of respiration rate prediction.

[0043] The gas superficial velocity is a set value calculated from the aeration rate and vessel diameter entered by the user. The aeration rate, rotation speed, dissolved oxygen control method, and fed-batch rate / perfusion rate (culture medium input / output conditions) are set by values ​​entered by the user. The turbulent energy dissipation rate is a value theoretically calculated from P / V (agitation power) or the in-apparatus mean value + standard deviation value obtained by multi-point measurement techniques in CFD simulations or experiments. The gas holdup is a value calculated from the liquid level difference with / without aeration or the in-apparatus mean value + standard deviation value obtained by multi-point measurement techniques in CFD simulations or experiments. The bubble diameter is a value measured using photographic images, etc., or the in-apparatus mean value + standard deviation value obtained by multi-point measurement techniques in CFD simulations or experiments. The shear stress is set by a value entered by the user or the in-apparatus mean value + standard deviation value obtained by multi-point measurement techniques in CFD simulations or experiments.

[0044] The learning data of the respiration rate, which is the response variable of the respiration rate model 1C, is acquired by a known method. For example, the learning data of the respiration rate is calculated from the oxygen concentration and the exhaust gas concentration measured in the actual machine during operation.

[0045] The bacterial body and cell concentration model 1D is a trained model that is trained using, for example, data on the shape of a stirrer provided in the container, data on the shape of a sparger provided in the container, data on the shape of the container, and data on the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container as input data (explanatory variables) that constitute the training data, and the bacterial body and cell concentration as output data (objective variables) that constitute the training data.

[0046] The data relating to the shape of the agitator is, for example, data relating to the shape of the agitator blades constituting the agitator, including the type of agitator blade, the agitator blade diameter, and the number of agitator blades that are set according to the shape of the agitator blade. The data relating to the shape of the agitator blade includes data in which a numerical label is attached for each type of shape, or includes data quantified by assigning a first numerical value (e.g., 1) to the area occupied by the agitator blade in three-dimensional space and a second numerical value (e.g., 0) to the area not occupied by the agitator blade in three-dimensional space.

[0047] Examples of data related to the shape of a sparger provided in the vessel include the type of aeration sparger, the diameter of the aeration sparger, the diameter of the aeration sparger holes, and the number of holes for the aeration sparger, which are set according to the shape of the sparger. The data related to the shape of the sparger includes data in which a numerical label is attached for each type of shape, or includes data quantified by assigning a first numerical value (e.g., 1) to an area occupied by the sparger in three-dimensional space and a second numerical value (e.g., 0) to an area not occupied by the sparger in three-dimensional space.

[0048] The data on the shape of the container is data on the shape of the container itself that contains the contents, and includes, for example, the type of container shape (cylindrical, rectangular, etc.), the dimensions of the container (inner diameter (in the case of a cylinder), width (in the case of a rectangle), height, etc.), and the presence or absence of a baffle (in the case of a mixing vessel).

[0049] Data regarding the physical properties, composition, chemical properties, biochemical properties, or concentrations of the contents includes, for example, the physical properties, composition, chemical properties, and biochemical properties specific to the materials contained in the contents, the physical properties of the entire contents within the container, and the concentrations of each material contained in the contents.

[0050] The bacterial body and cell concentration model 1D may be trained using the gas superficial velocity, aeration rate, turbulent energy dissipation rate, gas holdup, bubble diameter, dissolved oxygen (DO) control method (upper and lower limits, etc.), agitator rotation speed, shear stress, culture medium feed rate / perfusion rate (in the case of fed-batch culture / perfusion culture), timing of addition of fed-batch medium (in the case of fed-batch culture / perfusion culture), fed-batch culture method, number of days of culture, cell (bacterial body) survival rate, and viable cell (bacterial body) density as input data (explanatory variables) constituting the training data, and the bacterial body and cell concentration as output data (objective variables) constituting the training data.

[0051] The gas superficial velocity is a set value calculated from the aeration rate and vessel diameter entered by the user. The aeration rate, rotation speed, dissolved oxygen control method, and fed-batch rate / perfusion rate are set by values ​​entered by the user. The turbulent energy dissipation rate is a value theoretically calculated from P / V (agitation power) or the value of the in-apparatus mean value + standard deviation obtained by a multi-point measurement method in a CFD simulation or experiment. The gas holdup is a value calculated from the liquid level difference with / without aeration or the value of the in-apparatus mean value + standard deviation obtained by a multi-point measurement method in a CFD simulation or experiment. The bubble diameter is a value measured using photographic images, etc., or the value of the in-apparatus mean value + standard deviation obtained by a multi-point measurement method in a CFD simulation or experiment. The shear stress is set by a value entered by the user or the value of the in-apparatus mean value + standard deviation obtained by a multi-point measurement method in a CFD simulation or experiment.

[0052] Furthermore, the bacterial cell and cell concentration model 1D may use the concentration of products from bacterial cells and cells as the target variable in addition to or instead of the bacterial cell and cell concentrations. Products from bacterial cells and cells include proteins, genes, etc. that are produced in the bacterial cells and cells and secreted to the outside of the bacterial cells and cells. Proteins include target products (antibodies, etc.) and inhibitors that inhibit cell growth, etc.

[0053] The learning data of the bacterial cell and cell concentrations and / or the concentrations of products from the bacterial cells and cells, which are the response variables of the bacterial cell and cell concentration model 1D, are acquired by a known method. For example, this learning data includes the bacterial cell and cell concentrations and / or the concentrations of products from the bacterial cells and cells measured by sampling a portion of the contents of an actual machine during operation or by an optical sensor.

[0054] The titer model 1E is a trained model that uses, for example, data regarding the shape of the agitator installed in the container, data regarding the shape of the sparger installed in the container, data regarding the shape of the container, and data regarding the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container as input data (explanatory variables) that constitute the training data, and trains the titer as output data (objective variable) that constitutes the training data.

[0055] The data relating to the shape of the agitator is, for example, data relating to the shape of the agitator blades constituting the agitator, including the type of agitator blade, the agitator blade diameter, and the number of agitator blades that are set according to the shape of the agitator blade. The data relating to the shape of the agitator blade includes data in which a numerical label is attached for each type of shape, or includes data quantified by assigning a first numerical value (e.g., 1) to the area occupied by the agitator blade in three-dimensional space and a second numerical value (e.g., 0) to the area not occupied by the agitator blade in three-dimensional space.

[0056] Examples of data related to the shape of a sparger provided in the vessel include the type of aeration sparger, the diameter of the aeration sparger, the diameter of the aeration sparger holes, and the number of holes for the aeration sparger, which are set according to the shape of the sparger. The data related to the shape of the sparger includes data in which a numerical label is attached for each type of shape, or includes data quantified by assigning a first numerical value (e.g., 1) to an area occupied by the sparger in three-dimensional space and a second numerical value (e.g., 0) to an area not occupied by the sparger in three-dimensional space.

[0057] The data on the shape of the container is data on the shape of the container itself that contains the contents, and includes, for example, the type of container shape (cylindrical, rectangular, etc.), the dimensions of the container (inner diameter (in the case of a cylinder), width (in the case of a rectangle), height, etc.), and the presence or absence of a baffle (in the case of a mixing vessel).

[0058] Data regarding the physical properties, composition, chemical properties, biochemical properties, or concentrations of the contents includes, for example, the physical properties, composition, chemical properties, and biochemical properties specific to the materials contained in the contents, the physical properties of the entire contents within the container, and the concentrations of each material contained in the contents.

[0059] The titer model 1E may be trained using the gas superficial velocity, aeration rate, turbulent energy dissipation rate, gas holdup, bubble diameter, dissolved oxygen (DO) control method (upper and lower limits, etc.), agitator rotation speed, culture medium feed rate / perfusion rate (in the case of fed-batch culture / perfusion culture), shear stress, timing of addition of fed-batch medium (in the case of fed-batch culture / perfusion culture), fed-batch culture method, number of days of culture, cell (bacterial body) survival rate, and viable cell (bacterial body) density as input data (explanatory variables) constituting the training data, and the titer as output data (objective variable) constituting the training data.

[0060] The gas superficial velocity is a set value calculated from the aeration rate and vessel diameter entered by the user. The aeration rate, rotation speed, dissolved oxygen control method, and fed-batch rate / perfusion rate are set by values ​​entered by the user. The turbulent energy dissipation rate is a value theoretically calculated from P / V (agitation power) or the value of the in-apparatus mean value + standard deviation obtained by a multi-point measurement method in a CFD simulation or experiment. The gas holdup is a value calculated from the liquid level difference with / without aeration or the value of the in-apparatus mean value + standard deviation obtained by a multi-point measurement method in a CFD simulation or experiment. The bubble diameter is a value measured using photographic images, etc., or the value of the in-apparatus mean value + standard deviation obtained by a multi-point measurement method in a CFD simulation or experiment. The shear stress is set by a value entered by the user or the value of the in-apparatus mean value + standard deviation obtained by a multi-point measurement method in a CFD simulation or experiment.

[0061] The training data of the potency, which is the objective variable of the potency model 1E, is acquired by a known method. For example, the training data of the potency includes potency values ​​measured by sampling some of the contents in an actual machine during operation or measured by an optical sensor.

[0062] The operation reception unit 2 is configured, for example, with a keyboard and a mouse, and receives various user inputs. In this embodiment, the operation reception unit 2 selects the kLa model 1A, P / V model 1B, respiration rate model 1C, bacterial body and cell concentration model 1D, or titer model 1E stored in the trained model storage unit 1, and receives a setting operation from the user to set input data (explanatory variables) to be input into the selected model. The operation reception unit 2 then outputs setting operation information indicating the content of the received setting operation (model, input data) to the explanatory variable input unit 3. Here, the operation reception unit 2 may receive input data one set at a time, or may receive multiple datasets simultaneously. When receiving multiple datasets simultaneously, for example, the operation reception unit 2 is configured to receive a range setting for each input data included in the dataset, thereby receiving a set of numerical values ​​for each input data included in the range.

[0063] When the model indicated by the setting operation information output from the operation receiving unit 2 is the kLa model 1A, the explanatory variable input unit 3 inputs the input data indicated by the setting operation information to the kLa model 1A. Here, the input data input to the kLa model 1A is input data that constitutes the learning data used for learning the kLa model 1A, that is, data on the shape of the agitator provided in the container, data on the shape of the sparger provided in the container, data on the shape of the container, and data on the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container.

[0064] When the model indicated by the setting operation information output from the operation receiving unit 2 is P / V model 1B, the explanatory variable input unit 3 inputs the input data indicated by the setting operation information to P / V model 1B. Here, the input data input to P / V model 1B is input data that constitutes the learning data used for learning P / V model 1B, i.e., data on the shape of the agitator provided in the container, data on the shape of the container, and data on the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container.

[0065] When the model indicated by the setting operation information output from the operation receiving unit 2 is the respiration rate model 1C, the explanatory variable input unit 3 inputs the input data indicated by the setting operation information to the respiration rate model 1C. Here, the input data input to the respiration rate model 1C is input data that constitutes the learning data used for learning the respiration rate model 1C, i.e., data on the shape of the agitator provided in the container, data on the shape of the container, data on the shape of the sparger provided in the container, and data on the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container.

[0066] When the model indicated by the setting operation information output from the operation receiving unit 2 is the bacterial cell and cell concentration model 1D, the explanatory variable input unit 3 inputs the input data indicated by the setting operation information to the bacterial cell and cell concentration model 1D. Here, the input data input to the bacterial cell and cell concentration model 1D is input data that constitutes the learning data used for learning the bacterial cell and cell concentration model 1D, i.e., data on the shape of a stirrer provided in the container, data on the shape of a sparger provided in the container, data on the shape of the container, and data on the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container.

[0067] When the model indicated by the setting operation information output from the operation receiving unit 2 is the potency model 1E, the explanatory variable input unit 3 inputs the input data indicated by the setting operation information into the potency model 1E. Here, the input data input into the potency model 1E is input data that constitutes the learning data used for learning the potency model 1E, i.e., data on the shape of the agitator provided in the container, data on the shape of the sparger provided in the container, data on the shape of the container, and data on the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container.

[0068] When input data is input to the kLa model 1A by the explanatory variable input unit 3, the objective variable acquisition unit 4 acquires kLa output from the kLa model 1A. Then, the objective variable acquisition unit 4 outputs information indicating the acquired kLa to the display unit 7 to display kLa. This allows a user referring to the display unit 7 to evaluate kLa, which is one of the index data related to the dynamics of the contents of the container, with higher accuracy than in conventional techniques that use a correlation equation for evaluation. Furthermore, the objective variable acquisition unit 4 outputs information indicating the acquired kLa to the determination unit 5.

[0069] When input data is input to the P / V model 1B by the explanatory variable input unit 3, the objective variable acquisition unit 4 acquires the P / V output from the P / V model 1B. Then, the objective variable acquisition unit 4 outputs information indicating the acquired P / V to the display unit 7 to display the P / V. This allows a user who refers to the display unit 7 to evaluate the P / V, which is one piece of index data related to the dynamics of the contents of the container, with higher accuracy than in conventional techniques that use correlation equations for evaluation. In addition, the objective variable acquisition unit 4 outputs information indicating the acquired P / V to the determination unit 5.

[0070] When input data is input to the respiration rate model 1C by the explanatory variable input unit 3, the objective variable acquisition unit 4 acquires the respiration rate output from the respiration rate model 1C. Then, the objective variable acquisition unit 4 outputs information indicating the acquired respiration rate to the display unit 7 to display the respiration rate. This allows a user referring to the display unit 7 to accurately evaluate the respiration rate, which is one of the index data related to the dynamics of the contents of the container, which has previously been difficult to evaluate without relying on the operation of the actual machine. In addition, the objective variable acquisition unit 4 outputs information indicating the acquired respiration rate to the determination unit 5.

[0071] When input data is input to the bacterial cell and cell concentration model 1D by the explanatory variable input unit 3, the objective variable acquisition unit 4 acquires the bacterial cell and cell concentrations output from the bacterial cell and cell concentration model 1D. The objective variable acquisition unit 4 then outputs information indicating the acquired bacterial cell and cell concentrations to the display unit 7, causing the display unit 7 to display the bacterial cell and cell concentrations. This allows a user referring to the display unit 7 to accurately evaluate the bacterial cell and cell concentrations, which are index data related to the dynamics of the contents of a container, which has previously been difficult to evaluate without relying on actual machine operation. The objective variable acquisition unit 4 also outputs the acquired information indicating the bacterial cell and cell concentrations to the determination unit 5.

[0072] When input data is input to the titer model 1E by the explanatory variable input unit 3, the objective variable acquisition unit 4 acquires the titer output from the titer model 1E. Then, the objective variable acquisition unit 4 outputs information indicating the acquired titer to the display unit 7 to display the titer. This allows a user who refers to the display unit 7 to accurately evaluate the titer, which is one of the index data related to the dynamics of the contents of a container, which has previously been difficult to evaluate without relying on the operation of an actual machine. In addition, the objective variable acquisition unit 4 outputs information indicating the acquired titer to the determination unit 5.

[0073] The determination unit 5 determines whether the value of kLa output from the objective variable acquisition unit 4 is a predetermined value, and outputs information indicating the determination result to the explanatory variable acquisition unit 6. Here, the predetermined value is a desired value of kLa designed so that the kLa in the commercial equipment is equivalent to that in the laboratory scale when scaling up from a small laboratory scale to a large commercial scale, for example, and may be one value or multiple values.

[0074] The determination unit 5 determines whether the P / V value output from the objective variable acquisition unit 4 is a predetermined value, and outputs information indicating the determination result to the explanatory variable acquisition unit 6. Here, the predetermined value is a desired value of P / V designed so that the P / V in the commercial equipment is equivalent to that in the laboratory scale, for example, when scaling up from a small laboratory scale to a large commercial scale, and may be one value or multiple values.

[0075] The determination unit 5 determines whether the respiration rate value output from the objective variable acquisition unit 4 is a predetermined value, and outputs information indicating the determination result to the explanatory variable acquisition unit 6. Here, the predetermined value is a desired value of the respiration rate designed so that the respiration rate in the commercial device is equivalent to that in the laboratory scale, for example, when scaling up from a small laboratory scale to a large commercial scale, and may be one value or multiple values.

[0076] The determination unit 5 determines whether the values ​​of the bacterial body and cell concentrations output from the objective variable acquisition unit 4 are predetermined values, and outputs information indicating the determination result to the explanatory variable acquisition unit 6. Here, the predetermined value is a desired value of the bacterial body and cell concentrations designed so that the bacterial body and cell concentrations in the commercial equipment are equivalent to those on the lab scale, for example, when scaling up from a small lab scale to a large commercial scale, and may be one value or multiple values.

[0077] The determination unit 5 determines whether the titer value output from the objective variable acquisition unit 4 is a predetermined value, and outputs information indicating the determination result to the explanatory variable acquisition unit 6. Here, the predetermined value is a desired value of the titer designed so that the kLa in the commercial apparatus is equivalent to that in the laboratory scale when scaling up from a small laboratory scale to a large commercial scale, for example, and may be one value or multiple values.

[0078] The explanatory variable acquisition unit 6 refers to the determination result output from the determination unit 5, and when the value of kLa output from the objective variable acquisition unit 4 is a predetermined value, acquires the input data (explanatory variables) that was input to the kLa model 1A when the value of kLa was output from the kLa model 1A, and outputs information indicating the acquired input data to the display unit 7 to display the input data. This allows a user referring to the display unit 7 to easily understand the input data without having to search blindly for input data that will result in a desired value of kLa.

[0079] The explanatory variable acquisition unit 6 refers to the determination result output from the determination unit 5, and when the P / V value output from the objective variable acquisition unit 4 becomes a predetermined value, it acquires the input data (explanatory variables) that was input to the P / V model 1B when the P / V value was output from the P / V model 1B, and outputs information indicating the acquired input data to the display unit 7 to display the input data. This allows a user referring to the display unit 7 to easily understand the input data without having to search blindly for input data that will result in a desired P / V value.

[0080] The explanatory variable acquisition unit 6 refers to the determination result output from the determination unit 5, and when the respiration rate value output from the objective variable acquisition unit 4 is a predetermined value, acquires the input data (explanatory variables) that was input to the respiration rate model 1C when the respiration rate value was output from the respiration rate model 1C, and outputs information indicating the acquired input data to the display unit 7 to display the input data. This allows a user referring to the display unit 7 to easily understand the input data without having to search blindly for input data that will result in a desired respiration rate value.

[0081] The explanatory variable acquisition unit 6 refers to the determination result output from the determination unit 5, and when the bacterial body and cell concentration values ​​output from the objective variable acquisition unit 4 are predetermined values, it acquires input data (explanatory variables) that were input to the bacterial body and cell concentration model 1D when the bacterial body and cell concentration values ​​were output from the bacterial body and cell concentration model 1D, and outputs information indicating the acquired input data to the display unit 7 to display the input data. This allows a user referring to the display unit 7 to easily understand the input data without having to search blindly for input data that will result in desired bacterial body and cell concentration values.

[0082] The explanatory variable acquisition unit 6 refers to the determination result output from the determination unit 5, and when the titer value output from the objective variable acquisition unit 4 is a predetermined value, it acquires the input data (explanatory variables) that was input to the titer model 1E when that titer value was output from the titer model 1E, and outputs information indicative of the acquired input data to the display unit 7 to display the input data. This allows a user referring to the display unit 7 to easily understand the input data without having to search blindly for input data that will result in a desired titer value.

[0083] The display unit 7 is configured by, for example, a touch panel display provided in the estimation device 10 , and displays various information output from the objective variable acquisition unit 4 and the explanatory variable acquisition unit 6 .

[0084] 2 is a flowchart showing an example of the estimation process performed by the estimation device 10 in this embodiment. The process of step S100 in Fig. 2 starts when the operation reception unit 2 selects the kLa model 1A, the P / V model 1B, the respiration rate model 1C, the bacterial body and cell concentration model 1D, or the titer model 1E stored in the trained model storage unit 1 and receives a setting operation from the user to set input data (explanatory variables) to be input to the selected model.

[0085] First, the explanatory variable input unit 3 inputs the input data set via the operation reception unit 2 into the trained model (kLa model 1A, P / V model 1B, respiration rate model 1C, bacterial body and cell concentration model 1D, or titer model 1E) selected via the operation reception unit 2 (step S100).

[0086] Next, the input data is input to the trained model by the explanatory variable input unit 3, and the objective variable acquisition unit 4 acquires the output data (objective variables: kLa, P / V, respiration rate, bacterial and cell concentrations, concentrations or titers of products from bacterial and cell products) output from the trained model, and displays the acquired output data on the display unit 7 (step S110).

[0087] Next, the determination unit 5 determines whether the value of the output data acquired by the objective variable acquisition unit 4 is a predetermined value (step S120). If the result of the determination is that the value of the output data is not the predetermined value (step S120, NO), the process proceeds to step S140.

[0088] On the other hand, if the value of the output data is equal to the predetermined value (step S120, YES), the explanatory variable acquisition unit 6 acquires the input data (explanatory variables) that were input to the trained model when the value of the output data was output from the trained model, and displays the acquired input data on the display unit 7 (step S130). Then, the process proceeds to step S140.

[0089] In step S140, the explanatory variable input unit 3 determines whether input of the input data set via the operation reception unit 2 has been completed for the trained model (kLa model 1A, P / V model 1B, respiration rate model 1C, bacterial body and cell concentration model 1D, or titer model 1E) selected via the operation reception unit 2. If the determination result shows that input of the input data has not been completed (NO in step S140), the process returns to before step S100. On the other hand, if input of the input data has been completed (YES in step S140), the estimation device 10 ends the estimation process shown in FIG. 2.

[0090] (Effects of the Present Embodiment) As described in detail above, in the present embodiment, the estimation device 10 includes an explanatory variable input unit 3 and a response variable acquisition unit 4. The explanatory variable input unit inputs explanatory variables (input data) to the trained model (kLa model 1A, P / V model 1B, respiration rate model 1C, bacterial body and cell concentration model 1D, or titer model 1E). When the explanatory variables are input to the trained model, the response variable acquisition unit 4 acquires the response variable (output data) output from the trained model. The trained model is trained using at least one of data on the shape of a stirrer installed in a vessel when gas is blown into the liquid in the vessel, data on the shape of a sparger installed in the vessel, data on the shape of the vessel, and data on the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the vessel as explanatory variables, and using the mass transfer capacity coefficient (kLa), the stirring power (P / V), the respiration rate, bacterial body and cell concentration, the concentration of a product from the bacterial body and cells, or the titer as response variables.

[0091] According to this embodiment, at least one of the explanatory variables (data related to the shape of the agitator, data related to the shape of the sparger, data related to the shape of the vessel, and data related to the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents) can be input into the trained model, and the output variables, such as the volumetric mass transfer coefficient (kLa), the stirring power per unit volume (P / V), the respiration rate, the bacterial and cell concentration, and the concentration or titer of the product from the bacterial and cell, can be obtained. This allows for more accurate evaluation of indicator data related to the dynamics of the contents of the vessel than conventional techniques. For example, conventional correlation equations for the volumetric mass transfer coefficient (kLa) and the stirring power per unit volume (P / V) only take into account certain parameters, such as the aeration rate and the agitator rotation speed, and do not take into account the influence of the shape of the agitator and sparger, or the physical properties, composition, chemical properties, biochemical properties, and concentration of the contents of the vessel. Therefore, in the scale-up design of an apparatus, even if a calculation using a correlation equation yields results in which the volumetric mass transfer coefficient (kLa) and the stirring power per unit volume (P / V) satisfy predetermined values, the volumetric mass transfer coefficient (kLa) and the stirring power per unit volume (P / V) may not satisfy the predetermined values ​​when the actual apparatus designed accordingly is operated. In contrast, the present invention is configured to calculate these parameters using a machine learning model that takes into account the shapes of the agitator, sparger, and vessel, which have complex effects as the apparatus is scaled up, as well as the physical properties, composition, chemical properties, biochemical properties, and concentration of the contents of the vessel, making it possible to calculate an accurate volumetric mass transfer coefficient (kLa) and the stirring power per unit volume (P / V).

[0092] Furthermore, according to this embodiment, at least one of data (explanatory variables) related to the shape of the agitator, data related to the shape of the sparger, data related to the shape of the vessel, and data related to the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents can be input into the trained model to obtain the respiration rate, bacterial and cell concentration, and the concentration or titer of the product from the bacterial and cell as the objective variable. Conventionally, these parameters were difficult to estimate at the design stage of the actual equipment because they were not dependent on the operation of the actual equipment. However, the present invention allows these parameters to be estimated at the design stage, making it possible to more reliably design an equipment that exhibits the desired performance during actual operation.

[0093] Furthermore, the trained model of the present invention can be used not only during design but also to estimate the dynamics of the contents during operation of an actual machine. Values ​​indicating the dynamics of the contents, such as kLa, P / V, respiration rate, bacterial and cell concentration, concentration of products from bacterial and cell products, and titer, fluctuate after operation begins. The trained model of the present invention can be used to estimate these in real time. In this case, the same parameters as those described above may be included. When the trained model of the present invention is used to estimate the dynamics of the contents during operation of an actual machine, the explanatory variables of the trained model may include measurements of the physical properties of the entire contents in the container and the concentrations of each material contained in the contents acquired during operation of the actual machine. By continuously acquiring these data throughout operation and inputting them into the trained model, more accurate estimation of the current dynamics is possible. Furthermore, the explanatory variables may include images taken through the side or top window of the container, measurements of the contents using an odor sensor, measurements of container vibrations and associated sounds, the viability of bacterial and cell viability, or the density of live bacterial and cell viability. When the captured image is included as an explanatory variable, for example, information regarding the color of the contents in the captured image (e.g., RGB values) is extracted and used as an explanatory variable, thereby allowing the correlation between the color of the contents and indicators related to the dynamics of the contents to be reflected in the trained model.

[0094] (Variations) In the above embodiment, the first data (e.g., data related to the shape of the agitator) and the second data (e.g., data related to the shape of the sparger) constituting the input data (explanatory variables) input to the trained model may have a fixed value for the first data and a variable value for the second data. In this case, the explanatory variable acquisition unit 6 acquires the value of the second data input to the trained model when the value of the output data (objective variable) output from the trained model becomes a predetermined value. This allows the user to easily acquire the second data without having to search blindly for second data that makes the value of the output data output from the trained model the desired value, while keeping the value of the first data fixed. This embodiment is useful, for example, when there is a constraint in designing an apparatus that requires changing only the shape of the sparger and not changing the agitator used to satisfy the desired kLa value.

[0095] In the above embodiment, the operation reception unit 2 selects multiple models from the kLa model 1A, P / V model 1B, respiration rate model 1C, bacterial body and cell concentration model 1D, and titer model 1E stored in the trained model storage unit 1 and receives setting operations from the user to set input data (explanatory variables) to be input to the selected multiple models. Here, the input data may be input data for each of the multiple models individually, or input data common to the multiple models may be received collectively. The operation reception unit 2 then outputs setting operation information indicating the content of the received setting operation (model, input data) to the explanatory variable input unit 3, and the target variable acquisition unit 4 acquires the output of each of the multiple models. The determination unit 5 determines whether the output of each of the multiple models is equal to its respective predetermined value. The explanatory variable acquisition unit 6 may acquire the explanatory variables when all of the outputs of the multiple models are equal to the predetermined value, or may acquire the explanatory variables when some of the outputs of the multiple models are equal to the predetermined value. However, in the scale-up design of an apparatus, it is preferable to design all outputs to be predetermined values. Therefore, it is preferable that the explanatory variable acquisition unit 6 acquires explanatory variables when the outputs of the multiple models are all predetermined values. For example, if the multiple models are kLa Model 1A and P / V Model 1B, the explanatory variable acquisition unit 6 may be configured to acquire the explanatory variables when the output of kLa Model 1A is a predetermined value and the output of P / V Model 1B is a predetermined value. In the scale-up design of an apparatus, it is particularly important whether the kLa and P / V values ​​are predetermined values. Therefore, it is preferable that the multiple models include at least kLa Model 1A and P / V Model 1B. More preferably, to ensure that the designed apparatus achieves the desired operation, it is preferable that the multiple models include all of kLa Model 1A, P / V Model 1B, Respiration Rate Model 1C, Bacterial and Cell Concentration Model 1D, and Titer Model 1E.

[0096] Furthermore, in the above embodiment, the objective variable of a first model selected from kLa Model 1A, P / V Model 1B, respiration rate Model 1C, bacterial cell and cell concentration Model 1D, and titer Model 1E may be included as an explanatory variable of a second model selected from kLa Model 1A, P / V Model 1B, respiration rate Model 1C, bacterial cell and cell concentration Model 1D, and titer Model 1E. The index data representing the dynamics of the contents, kLa, P / V, respiration rate, bacterial cell and cell concentration, concentration of products from bacterial cells and cells, and titer, are all correlated parameters. Therefore, by employing the index estimated by Model 1 as an explanatory variable of a model using other indexes as objective variables, more accurate estimation of the objective variable is possible. For example, as described above, kLa is an index indicating the mass transfer rate of gas to liquid. An increase in the mass transfer rate promotes the absorption of bacterial cells and cells in the liquid, increasing the respiration rate, which may result in an increase in bacterial cell and cell concentration, concentration of products from bacterial cells and cells, and titer. Therefore, k (the objective variable of the first model) estimated by k model 1A may be included as an explanatory variable (the explanatory variable of the second model) of respiration rate model 1C, bacterial body and cell concentration model 1D, or titer model 1E. Similarly, the value of the objective variable of P / V model 1B may be included as an explanatory variable of k model 1A, respiration rate model 1C, bacterial body and cell concentration model 1D, or titer model 1E, and the value of the objective variable of respiration rate model 1C may be included as an explanatory variable of bacterial body and cell concentration model 1D or titer model 1E.

[0097] Furthermore, while the above embodiment describes a culture tank with aeration and agitation, the target of the trained model of the present invention is not limited to this. For example, a culture tank without aeration or agitation (e.g., a shaking culture tank) may also be used. In this case, the explanatory variables of Models 1A to 1E may not include data regarding the shape of the agitator or the shape of the sparger provided in the vessel, but may include data regarding the shape of the vessel and data regarding the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the vessel. Furthermore, a culture tank without aeration but with agitation may also be used. In this case, the explanatory variables of Models 1A to 1E may not include data regarding the shape of the sparger provided in the vessel, but may include data regarding the shape of the agitator, data regarding the shape of the vessel, and data regarding the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the vessel. A culture tank with aeration but no agitation may also be used. In this case, the explanatory variables of Models 1A-1E may not include data on the shape of the agitator, but may include data on the shape of a sparger installed in the vessel, data on the shape of the vessel, and data on the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the vessel. A reactor (such as a bubble column) in which a chemical reaction (such as a chemical synthesis reaction or a chemical decomposition reaction) takes place may also be used as the trained model. Similarly, in this case, explanatory variables may be set depending on whether or not a stirrer and a sparger are present. In this case, data on the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the vessel may particularly include chemical properties specific to the material contained in the contents (e.g., reactivity, activation energy, reaction rate, and reaction rate constant in the chemical reaction of the material).

[0098] In addition, in the above embodiment, the output data (objective variables: kLa, P / V, respiration rate, bacterial body and cell concentration, concentration of product from bacterial body and cells, or titer) output from the trained model (kLa model 1A, P / V model 1B, respiration rate model 1C, bacterial body and cell concentration model 1D, or titer model 1E) may change over time.

[0099] FIG. 3 is an image diagram showing changes in the value of output data depending on the number of days of culture. In FIG. 3, curve L1 shows the change in the value of output data (kLa) output from kLa model 1A depending on the number of days of culture (1 to 7 days in the illustrated example). Curve L2 shows the change in the value of output data (P / V) output from P / V model 1B depending on the number of days of culture. Curve L3 shows the change in the value of output data (respiration rate) output from respiration rate model 1C depending on the number of days of culture. Curve L4 shows the change in the value of output data (cell concentration) output from bacterial cell and cell concentration model 1D depending on the number of days of culture. Curve L5 shows the change in the value of output data (titer) output from titer model 1E depending on the number of days of culture. As shown in FIG. 3, the values ​​of the output data (P / V, particularly cell concentration and titer) change significantly depending on the number of days of culture.

[0100] Taking into account the change in the output data values ​​shown in Figure 3, the values ​​of the output data (kLa, P / V, respiration rate, bacterial cell and cell concentration, concentration of product from bacterial cells and cells, or titer) output from the trained models (kLa model 1A, P / V model 1B, respiration rate model 1C, bacterial cell and cell concentration model 1D, or titer model 1E) may be the value when the culture days are the final day (7 days), the value when the culture days are the intermediate day (4 days), both the value when the culture days are the final day (7 days) and the value when the culture days are the intermediate day (4 days), or the average value for the entire period (7 days).

[0101] Furthermore, the kLa model 1A may be trained using the number of days of culture as input data (explanatory variables) constituting the training data, in addition to data on the shape of the agitator, data on the shape of the sparger, and data on the physical properties, composition, chemical properties, or biochemical properties of the contents, and kLa (volumetric mass transfer coefficient) as output data (objective variable) constituting the training data. In this case, the explanatory variable input unit 3 inputs the input data including the number of days of culture to the kLa model 1A.

[0102] Furthermore, the P / V model 1B may be trained using the number of days of culture as input data (explanatory variables) constituting the training data, in addition to data on the shape of the agitator and data on the physical properties, composition, chemical properties, or biochemical properties of the contents, and the P / V as output data (target variable) constituting the training data. In this case, the explanatory variable input unit 3 inputs the input data including the number of days of culture to the P / V model 1B.

[0103] Furthermore, the respiration rate model 1C may be trained using the number of days of culture as input data (explanatory variables) constituting the training data, in addition to data on the shape of the agitator, data on the shape of the sparger, data on the shape of the container, and data on the physical properties, composition, chemical properties, or biochemical properties of the contents, and the respiration rate as output data (target variable) constituting the training data. In this case, the explanatory variable input unit 3 inputs the input data including the number of days of culture to the respiration rate model 1C.

[0104] Furthermore, the bacterial body and cell concentration model 1D may be trained using the number of days of culture as input data (explanatory variables) constituting the training data, in addition to data on the shape of the agitator, data on the shape of the sparger, data on the shape of the vessel, and data on the physical properties, composition, chemical properties, or biochemical properties of the contents, and the bacterial body and cell concentrations and the concentrations of products from the bacterial body and cells as output data (objective variables) constituting the training data. In this case, the explanatory variable input unit 3 inputs the input data including the number of days of culture to the bacterial body and cell concentration model 1D.

[0105] Furthermore, the titer model 1E may be trained using the number of days of culture as input data (explanatory variables) constituting the training data, in addition to data on the shape of the agitator, data on the shape of the sparger, data on the shape of the container, and data on the physical properties, composition, chemical properties, or biochemical properties of the contents, and the titer as output data (objective variable) constituting the training data. In this case, the explanatory variable input unit 3 inputs the input data including the number of days of culture to the titer model 1E.

[0106] In addition, considering that the values ​​of the output data (kLa, P / V, respiration rate) change slightly depending on the number of culture days, mainly due to the influence of data regarding the physical properties, composition, chemical properties, or biochemical properties of the contents, data regarding the physical properties, composition, chemical properties, or biochemical properties of the contents expected during the expected number of culture days can be input into the trained models (kLa model 1A, P / V model 1B, respiration rate model 1C), and it can be confirmed whether the values ​​of the output data (kLa, P / V, respiration rate) output from the trained models satisfy the specified values.

[0107] Furthermore, the above-described embodiments are merely examples of specific embodiments for carrying out the present invention, and the technical scope of the present invention should not be construed as being limited by these embodiments. In other words, the present invention can be carried out in various forms without departing from the gist or main features thereof.

[0108] 1: Trained model storage device, 1A: kLa model, 1B: P / V model, 1C: Respiration rate model, 1D: Bacterial cell concentration model, 1E: Titer model, 2: Operation reception unit, 3: Explanatory variable input unit, 4: Objective variable acquisition unit, 5: Determination unit, 6: Explanatory variable acquisition unit, 7: Display unit, 10: Estimation device

Claims

1. A trained model having at least one of the following explanatory variables: data regarding the shape of a stirrer installed in a vessel; data regarding the shape of a sparger installed in the vessel; data regarding the shape of the vessel; and data regarding the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the vessel; and trained using the following objective variables: a mass transfer capacity coefficient (kLa), agitation power per unit volume (P / V), respiration rate, bacterial mass and cell concentration, product concentration from bacterial mass and cells, or titer.

2. A trained model as described in claim 1, trained using data regarding the shape of the agitator, data regarding the shape of the sparger, data regarding the shape of the container, and data regarding the physical properties, composition, chemical properties, biochemical properties, or concentration of the contents contained in the container as explanatory variables.

3. A trained model as described in claim 1, wherein the data regarding the shape of the agitator includes data regarding the shape of the agitator blades that constitute the agitator, and the data regarding the shape of the agitator blades and the data regarding the shape of the sparger include data with numerical labels assigned for each type of shape, or include quantified data in which a first numerical value is assigned to an area occupied by the agitator blades and the sparger in three-dimensional space and a second numerical value is assigned to an area not occupied by the agitator blades and the sparger in three-dimensional space.

4. An estimation device comprising an input unit and a dependent variable acquisition unit, wherein the input unit inputs the explanatory variables to a trained model according to any one of claims 1 to 3, and the dependent variable acquisition unit acquires the dependent variable output from the trained model when the explanatory variables are input to the trained model.

5. An estimation device as described in claim 4, further comprising an explanatory variable acquisition unit, wherein the explanatory variable acquisition unit acquires the value of the explanatory variable input to the trained model when the value of the objective variable output from the trained model becomes a predetermined value.

6. An estimation device as described in claim 5, wherein, of the first data and second data constituting the explanatory variable, the value of the first data is a fixed value and the value of the second data is a variable value, and the explanatory variable acquisition unit acquires the value of the second data input to the trained model when the value of the objective variable output from the trained model becomes a predetermined value.

7. An estimation method comprising: inputting the explanatory variables into a trained model according to any one of claims 1 to 3; and, when the explanatory variables are input into the trained model, obtaining the objective variable output from the trained model.

Images