Chromatographic column sample loading control method, device, equipment, medium and program product

Abstract

The embodiment of the invention discloses a chromatographic column sample loading control method, device and equipment, a medium and a program product. The method comprises the following steps: determining the current protein concentration at an outlet of a chromatographic column after starting to load a sample on the chromatographic column; based on the current protein concentration and the sample loading protein concentration, determining a current penetration ratio, and when the current penetration ratio reaches a preset penetration ratio, determining a partial penetration curve corresponding to the current sample loading time period; performing curve fitting based on the target network model and the partial penetration curve, and determining current chromatographic column parameter information; predicting a complete penetration curve corresponding to the chromatographic column based on the target network model and the current chromatographic column parameter information; determining a target sample loading moment based on the complete penetration curve and a target penetration proportion corresponding to the target medium utilization rate; when the current sample loading moment reaches the target sample loading moment, sample loading of the chromatographic column is stopped, so that online optimization control of the sample loading process is realized, and it is ensured that the target medium utilization rate can be reached in each sample loading process.

Description

Technical Field

[0001] The embodiments of the present invention relate to downstream purification technology of biological products, and more particularly to a chromatography column loading control method, device, equipment, medium and program product. Background Art

[0002] Affinity chromatography plays an extremely important role in the biopharmaceutical industry, mainly because it provides a highly specific method to purify proteins and other biomolecules. This technology is based on the natural affinity between molecules, such as the interaction between antigens and antibodies, enzymes and substrates, and receptors and ligands. It improves product purity while saving costs and process time, and has a wide range of applications in the biopharmaceutical industry.

[0003] At present, before loading the sample on the chromatography column for affinity chromatography, it is necessary to set a fixed number of cycles, sample loading volume and sample loading flow rate, so as to perform the number of cycles during the sample loading process and complete the sample loading according to the sample loading volume and sample loading flow rate in each cycle. However, in the process of realizing the present invention, the inventors found that there are at least the following problems in the prior art:

[0004] The existing method is to control the sample offline before loading, and the performance of the column filler will decay after multiple loadings, resulting in failure to achieve the target medium utilization rate during each loading of the column, thereby reducing economic benefits. Summary of the invention

[0005] The embodiments of the present invention provide a chromatography column loading control method, device, equipment, medium and program product to perform online optimization control on the chromatography column loading process and cope with the influence of chromatography column filler attenuation, thereby ensuring that the target medium utilization rate can be achieved during each loading process.

[0006] In a first aspect, an embodiment of the present invention provides a chromatography column loading control method, comprising:

[0007] After the chromatography column starts to load the sample, determining the current protein concentration at the outlet of the chromatography column;

[0008] Based on the current protein concentration and the loading protein concentration of the chromatography column, determining a current penetration ratio corresponding to the current loading moment, and when the current penetration ratio reaches a preset penetration ratio, determining a partial penetration curve corresponding to the current loading time period;

[0009] Performing curve fitting based on the target network model and the partial penetration curve to determine the current chromatography column parameter information, wherein the target network model is obtained by pre-training and is used to predict the complete penetration curve corresponding to the chromatography column parameter information;

[0010] Based on the target network model and the current chromatography column parameter information, predicting a complete penetration curve corresponding to the chromatography column;

[0011] Determining a target loading time corresponding to the target penetration ratio based on the complete penetration curve and the target medium utilization rate;

[0012] When the current sample loading time reaches the target sample loading time, the sample loading to the chromatography column is stopped.

[0013] In a second aspect, an embodiment of the present invention further provides a chromatography column loading control device, comprising:

[0014] A current protein concentration determination module, used to determine the current protein concentration at the outlet of the chromatography column after the chromatography column starts to load the sample;

[0015] A partial penetration curve determination module, for determining a current penetration ratio corresponding to a current loading moment based on the current protein concentration and the loading protein concentration of the chromatography column, and determining a partial penetration curve corresponding to a current loading time period when the current penetration ratio reaches a preset penetration ratio;

[0016] a current chromatography column parameter information determination module, used to perform curve fitting based on a target network model and the partial penetration curve to determine the current chromatography column parameter information, wherein the target network model is obtained by pre-training and is used to predict the complete penetration curve corresponding to the chromatography column parameter information;

[0017] A complete penetration curve prediction module, used to predict the complete penetration curve corresponding to the chromatography column based on the target network model and the current chromatography column parameter information;

[0018] A target loading time determination module, used to determine a target loading time corresponding to the target penetration ratio based on the complete penetration curve and the target penetration ratio corresponding to the target medium utilization rate;

[0019] The chromatography column sample loading stop module is used to stop loading the chromatography column when the current sample loading time reaches the target sample loading time.

[0020] In a third aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0021] one or more processors;

[0022] A memory for storing one or more programs;

[0023] When the one or more programs are executed by the one or more processors, the one or more processors implement the chromatography column loading control method provided in any embodiment of the present invention.

[0024] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the chromatography column loading control method provided in any embodiment of the present invention.

[0025] In a fifth aspect, an embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the chromatography column loading control method provided in any embodiment of the present invention.

[0026] One embodiment of the above invention has the following advantages or beneficial effects:

[0027] After the chromatography column starts to be loaded, the current protein concentration at the outlet of the chromatography column is determined in real time, and based on the current protein concentration and the loaded protein concentration of the chromatography column, the current penetration ratio corresponding to the current loading time is determined, and when the current penetration ratio reaches the preset penetration ratio, the partial penetration curve corresponding to the current loading time period is determined; based on the target network model obtained by pre-training and the partial penetration curve, the curve fitting is performed, so that the current chromatography column parameter information of the chromatography column can be quickly and accurately determined, so that the parameter information after the chromatography column filler decays can be accurately estimated, and based on the target network model and the current chromatography column parameter information, the complete penetration curve corresponding to the current chromatography column can be accurately predicted, so as to cope with the influence of the chromatography column filler decay; based on the complete penetration curve and the target penetration ratio corresponding to the target medium utilization rate, the target loading time corresponding to the target penetration ratio is determined, and when the current loading time reaches the target loading time, the chromatography column is stopped from loading, so that by controlling the loading time of the chromatography column loading process, it can be ensured that the target medium utilization rate can be achieved in each loading process, and the online optimization control of the chromatography column loading is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 It is a flow chart of a chromatography column loading control method provided by one embodiment of the present invention;

[0030] Figure 2 is a flow chart of another chromatography column loading control method provided by one embodiment of the present invention;

[0031] Figure 3 is a flow chart of another chromatography column loading control method provided by one embodiment of the present invention;

[0032] Figure 4 It is a structural schematic diagram of a chromatography column loading control device provided by one embodiment of the present invention;

[0033] Figure 5 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0035] Figure 1 This is a flow chart of a chromatography column loading control method provided by one embodiment of the present invention. This embodiment can be applied to the case of online optimization control of the chromatography column loading process. The method can be executed by a chromatography column loading control device, which can be implemented by software and / or hardware and integrated into an electronic device. Figure 1 As shown, the method specifically comprises the following steps:

[0036] S110, after the chromatography column starts to be loaded with sample, determine the current protein concentration at the outlet of the chromatography column.

[0037] Among them, loading refers to the process of adding the sample to be chromatographed to the chromatography column. The current protein concentration refers to the concentration of the mobile phase protein between the particles at the outlet of the chromatography column at the current moment.

[0038] Specifically, the sample is added to the chromatography column multiple times for chromatography treatment. For each loading process of the chromatography column, after the chromatography column starts loading, the current protein concentration at the outlet of the chromatography column is detected and determined in real time, so as to know the chromatography status of the loading process in real time.

[0039] Exemplarily, "determining the current protein concentration at the outlet of the chromatography column" in step S110 may include: obtaining the current ultraviolet value at the outlet of the chromatography column; and determining the current protein concentration at the outlet of the chromatography column based on a pre-established correspondence between the protein concentration and the ultraviolet value and the current ultraviolet value.

[0040] Specifically, at least 3 standard samples of known protein concentrations can be prepared in advance by diluting and concentrating the target protein, and ultraviolet spectroscopy is used to irradiate the standard sample with ultraviolet light of 280nm wavelength to obtain ultraviolet UV values. A linear relationship is established between the known protein concentration and the corresponding ultraviolet value, thereby obtaining a corresponding relationship between the protein concentration and the ultraviolet value. After the actual sampling begins, the current ultraviolet value at the outlet of the chromatography column can be collected in real time, and based on the pre-established corresponding relationship between the protein concentration and the ultraviolet value, the current protein concentration corresponding to the current ultraviolet value is determined, so that the current protein concentration can be accurately and quickly determined by collecting the ultraviolet value.

[0041] S120, determining a current penetration ratio corresponding to a current loading moment based on a current protein concentration and a loading protein concentration of the chromatography column, and determining a partial penetration curve corresponding to a current loading time period when the current penetration ratio reaches a preset penetration ratio.

[0042] Among them, the preset penetration ratio can be a pre-set, specified penetration ratio that the chromatography column needs to reach during the loading process. The current loading time period refers to the period from the start of loading to the current loading moment when the preset penetration ratio is reached. The partial penetration curve is generated based on the penetration ratio corresponding to each loading moment in the current loading time period. The partial penetration curve is the penetration curve of the front part of the complete penetration curve of the entire loading process. The preset penetration ratio is used to control the end time of the partial penetration curve. For example, in order to ensure that the partial penetration curve has sufficient fitting data to ensure the accuracy of subsequent curve fitting, the preset penetration ratio can be set to 50%.

[0043] Specifically, the current protein concentration is divided by the protein concentration of the sample loaded on the chromatography column to obtain the current penetration ratio at the current loading time, thereby realizing real-time detection of the penetration ratio during the loading process. When it is detected that the current penetration ratio reaches the preset penetration ratio, it indicates that the loading optimization can be performed at present, and a partial penetration curve corresponding to the current loading time period is generated at this time. It should be noted that when the loading optimization is subsequently performed based on the partial penetration curve, that is, during the execution of the subsequent steps S130-S150, the loading process of the chromatography column does not stop, and is always continuously executed, thereby realizing online optimization of the loading of the chromatography column.

[0044] S130, performing curve fitting based on the target network model and the partial penetration curve to determine the current chromatography column parameter information, wherein the target network model is obtained by pre-training and is used to predict the complete penetration curve corresponding to the chromatography column parameter information.

[0045] Among them, the current chromatography column parameter information may refer to the current parameter information of the chromatography column. Since the performance attenuation of the chromatography column filler will cause the chromatography column parameter information to change, it is necessary to determine the parameter information of the chromatography column at the current moment in real time, so as to ensure the accuracy of the loading control. The complete penetration curve refers to the penetration curve corresponding to the entire loading time period. The entire loading time period refers to the time period from the start time of loading to the end time of loading. The target network model can be a neural network model that predicts the complete penetration curve based on the chromatography column parameter information. The target network model can directly predict the complete penetration curve, or it can indirectly predict the complete penetration curve, such as first predicting the mobile phase protein concentration between the particles corresponding to each loading moment, and then dividing the mobile phase protein concentration by the loading protein concentration to obtain the complete penetration curve. The target network model can simulate the chromatography process through training, so that the penetration curve can be predicted directly or indirectly.

[0046] Exemplarily, the current chromatography column parameter information may include the current first chromatography column parameter information related to the chromatography column and the filler, and the current second chromatography column parameter information related to time and space. For example, the current first chromatography column parameter information may include: chromatography column height, filler particle average radius, sample protein concentration, apparent flow rate, chromatography column porosity, filler particle porosity, axial diffusion coefficient, pore diffusion coefficient, surface diffusion coefficient, membrane mass transfer coefficient, equilibrium dissociation constant and filler maximum adsorption capacity. The current second chromatography column parameter information may include: sample loading process time, radial coordinates within particles and axial coordinates within the chromatography column.

[0047] Among them, the current first chromatography column parameter information corresponds to a chromatography column in a physical sense, that is, chromatography columns with different performances have different current first chromatography column parameter information. The loading process time includes each loading moment in the entire loading process. The radial coordinate within the particle refers to the radial coordinate within the particle from the core of the filler particle to the particle surface. The axial coordinate within the chromatography column refers to the axial coordinate within the chromatography column from the chromatography column inlet to the chromatography column outlet. The complete penetration curve is generated based on the penetration ratio at the chromatography column outlet at each loading moment in the entire loading process. Among them, each penetration ratio is obtained by dividing the mobile phase protein concentration between particles at each loading moment by the loading protein concentration.

[0048] It should be noted that the current first chromatography column parameter information and the current second chromatography column parameter information both refer to different parameter information of the same chromatography column.

[0049] Specifically, based on the target network model, the penetration curve corresponding to the chromatography column parameter information can be predicted, and the predicted penetration curve can be curve-fitted with the partial penetration curve, and the current chromatography column parameter information corresponding to the partial penetration curve can be determined based on the fitting result, so that the current chromatography column parameter information can be accurately obtained by fitting the partial penetration curve using the target network model, thereby realizing the real-time determination of the physical parameter information of the chromatography column under the influence of filler attenuation, and thus being able to cope with the influence of chromatography column filler attenuation.

[0050] S140, predicting a complete penetration curve corresponding to the chromatography column based on the target network model and current chromatography column parameter information.

[0051] Specifically, if the target network model is used to directly predict the complete penetration curve, the current column parameter information is input into the target network model for curve prediction, and based on the output of the target network model, the complete penetration curve of the column in the entire chromatography process is directly obtained.

[0052] Exemplarily, step S140 may include: inputting the current chromatography column parameter information into the target network model to predict the protein concentration, and obtaining the mobile phase protein concentration between particles corresponding to each loading moment at the chromatography column outlet; based on the mobile phase protein concentration between particles and the loading protein concentration of the chromatography column, determining the complete penetration curve of the chromatography column during the entire loading process.

[0053] Specifically, if the target network model is used to predict the concentration of mobile phase protein to indirectly determine the complete penetration curve, the current column parameter information is input into the target network model to predict the protein concentration, and the mobile phase protein concentration between particles corresponding to each loading time at the outlet of the column is obtained based on the output of the target network model. The mobile phase protein concentration corresponding to each loading time is divided by the loading protein concentration of the column to obtain the penetration ratio corresponding to each loading time, and the loading time is used as the horizontal coordinate and the penetration ratio is used as the vertical coordinate to generate a complete penetration curve of the column during the entire loading process.

[0054] S150 . Determine a target sample loading time corresponding to the target penetration ratio based on the complete penetration curve and the target medium utilization rate.

[0055] Among them, the medium utilization rate can be an indicator for evaluating the chromatography performance during the sample loading process of the chromatography column. The target medium utilization rate can refer to the maximum medium utilization rate expected to be achieved during the chromatography process. The target penetration ratio refers to the penetration ratio that needs to be achieved when the target medium utilization rate is achieved, that is, the maximum penetration ratio. The target loading time refers to the loading time when the target penetration ratio is reached, that is, the optimal loading end time.

[0056] Specifically, the subsequent penetration curve after the current sample loading time period can be known in advance through the predicted complete penetration curve, so that the target sample loading time when the target penetration ratio is reached can be predicted in advance based on the complete penetration curve.

[0057] S160, when the current sample loading time reaches the target sample loading time, stop loading the sample on the chromatography column.

[0058] Specifically, during the chromatography column loading process, when the current loading time reaches the target loading time, it indicates that the current loading time reaches the target penetration ratio, that is, the medium utilization rate reaches the maximum. At this time, the chromatography column loading operation can be stopped, so as to ensure that the target medium utilization rate can be achieved during each loading process, and the online optimization control of the chromatography column loading is realized. The target medium rate is achieved by controlling the chromatography column loading time, avoiding stopping the loading too early or too late, and will not be affected by the attenuation of the chromatography column filler. At the same time, the target network model can be used to quickly perform curve fitting and curve prediction, meeting the real-time requirements of online control. For example, the time required for the entire process of fitting and prediction using the target network model is within 15 seconds, thereby meeting the real-time requirements.

[0059] The technical solution of this embodiment determines the current protein concentration at the outlet of the chromatography column in real time after the chromatography column starts to be loaded, and determines the current penetration ratio corresponding to the current loading time based on the current protein concentration and the loaded protein concentration of the chromatography column, and determines the partial penetration curve corresponding to the current loading time period when the current penetration ratio reaches the preset penetration ratio; based on the target network model obtained by pre-training and the partial penetration curve, the curve fitting is performed to quickly and accurately determine the current chromatography column parameter information of the chromatography column, so that the parameter information after the chromatography column filler attenuation can be accurately estimated, and based on the target network model and the current chromatography column parameter information, the complete penetration curve corresponding to the current chromatography column can be accurately predicted, so as to cope with the influence of the chromatography column filler attenuation; based on the complete penetration curve and the target penetration ratio corresponding to the target medium utilization rate, the target loading time corresponding to the target penetration ratio is determined, and when the current loading time reaches the target loading time, the chromatography column is stopped from loading, so that by controlling the loading time of the chromatography column loading process, it can be ensured that the target medium utilization rate can be achieved in each loading process, and the online optimization control of the chromatography column loading is realized.

[0060] Figure 2 A flowchart of another chromatography column loading control method provided by an embodiment of the present invention. Based on the above embodiments, this embodiment optimizes the step of "determining the current chromatography column parameter information by curve fitting based on the target network model and the partial penetration curve". The explanations of the terms that are the same or corresponding to the above embodiments are not repeated here.

[0061] See also Figure 2 Another chromatography column loading control method provided in this embodiment specifically comprises the following steps:

[0062] S210, after the chromatography column starts to be loaded with samples, determine the current protein concentration at the outlet of the chromatography column.

[0063] S220, determining a current penetration ratio corresponding to a current loading moment based on a current protein concentration and a loading protein concentration of the chromatography column, and determining a partial penetration curve corresponding to a current loading time period when the current penetration ratio reaches a preset penetration ratio.

[0064] S230. Obtain current third chromatography column parameter information and current second chromatography column parameter information in the current chromatography column parameter information.

[0065] Among them, the current chromatography column parameter information includes: the current first chromatography column parameter information related to the chromatography column and the filler, and the current second chromatography column parameter information related to time and space. The current first chromatography column parameter information can be divided into two types of information, that is, the current first chromatography column parameter information includes: the current third chromatography column parameter information under the first type and the current fourth chromatography column parameter information under the second type. Among them, the first type refers to the type of information that can be directly obtained. The second type refers to the type of information that needs to be obtained by curve fitting. For example, the current third chromatography column parameter information under the first type may include: chromatography column column height, filler particle average radius, sample protein concentration and apparent flow rate. The current fourth chromatography column parameter information under the second type may include: chromatography column porosity, filler particle porosity, axial diffusion coefficient, pore diffusion coefficient, surface diffusion coefficient, membrane mass transfer coefficient, equilibrium dissociation constant and filler maximum adsorption capacity.

[0066] It should be noted that the current first chromatography column parameter information (which includes the current third chromatography column parameter information and the current fourth chromatography column parameter information) and the current second chromatography column parameter information both refer to different parameter information of the same chromatography column.

[0067] Specifically, the current third chromatography column parameter information can be directly obtained based on the chromatography column used in the loading process and the loading information of the chromatography column. For example, the column height and the average radius of the filler particles can be directly obtained based on the chromatography column used in the loading process, and the loading protein concentration and the apparent flow rate can be directly obtained based on the loading information of the chromatography column. The current second chromatography column parameter information acquisition process can be: obtaining the loading process time in the actual generated partial penetration curve, and directly assigning the radial coordinates in the particles and the axial coordinates in the chromatography column, wherein the radial coordinates in the particles corresponding to the outlet are the normalized value 1, representing the particle surface; the axial coordinates in the chromatography column corresponding to the outlet are the normalized value 1, representing the outlet of the chromatography column.

[0068] S240. Perform curve fitting based on the target network model and the partial penetration curve to determine the current fourth chromatography column parameter information in the current chromatography column parameter information.

[0069] Specifically, based on the target network model, the penetration curve corresponding to the column parameter information can be predicted, and the predicted penetration curve is curve-fitted with the partial penetration curve, and the current fourth column parameter information corresponding to the partial penetration curve is determined based on the fitting result. By fitting the partial penetration curve, the current fourth column parameter information can be accurately obtained, and the real-time determination of the column physical parameters under the influence of filler attenuation is realized.

[0070] Exemplarily, step S240 may include: based on an adaptive particle swarm optimization method, minimizing the error value between the candidate penetration curve predicted by the target network model and the partial penetration curve to obtain the current fourth chromatography column parameter information.

[0071] Among them, the first column parameter information corresponding to the penetration curve at the outlet of each column is unique, so the curve fitting task can be conceptualized as a global optimization problem with bounded constraints, and the global optimal solution of this problem is the only correct first column parameter information corresponding to the penetration curve. Adaptive Particle Swarm Optimization (APSO) is an optimization algorithm based on swarm intelligence, which simulates the foraging behavior of bird flocks to find the optimal solution. In the problem of curve fitting, especially when used to find the parameters corresponding to the chromatography column, the adaptive particle swarm optimization method has high efficiency and accuracy.

[0072] Specifically, since the current third chromatography column parameter information and the current second chromatography column parameter information in the current chromatography column parameter information are known quantities that have been determined, it is only necessary to use the adaptive particle swarm optimization method to determine the unknown current fourth chromatography column parameter information. For example, using the adaptive particle swarm optimization method, the error value between the candidate penetration curve predicted by the target network model and the partial penetration curve is minimized, and the chromatography column parameter information input by the candidate penetration curve when the target network model predicts the minimization is obtained, and the fourth chromatography column parameter information in the chromatography column parameter information is used as the final current fourth chromatography column parameter information. By using the adaptive particle swarm optimization method, the current fourth chromatography column parameter information can be determined more quickly and accurately.

[0073] S250: predicting a complete penetration curve corresponding to the chromatography column based on the target network model and current chromatography column parameter information.

[0074] Specifically, the current second chromatography column parameter information, the current third chromatography column parameter information and the current fourth chromatography column parameter information are input as the current chromatography column parameter information into the target network model to predict the penetration curve, and a predicted complete penetration curve is obtained.

[0075] S260: Determine a target sample loading time corresponding to the target penetration ratio based on the complete penetration curve and the target medium utilization rate.

[0076] S270, when the current sample loading time reaches the target sample loading time, stop loading the sample on the chromatography column.

[0077] The technical solution of this embodiment can accurately determine the current fourth chromatography column parameter information by directly acquiring the current third chromatography column parameter information and the current second chromatography column parameter information and performing curve fitting based on the target network model and the partial penetration curve, thereby realizing real-time determination of the chromatography column physical parameters under the influence of filler attenuation.

[0078] On the basis of the above technical solutions, the step of "minimizing the error value between the candidate penetration curve and the partial penetration curve predicted by the target network model based on the adaptive particle swarm optimization method to obtain the current fourth chromatography column parameter information" can be carried out through the following steps S241-244:

[0079] S241. Generate a plurality of particles in a preset parameter search space, wherein the position of each particle represents a set of parameter information of a fourth chromatography column to be selected.

[0080] Among them, the number of fourth chromatography column parameter information is multiple, such as chromatography column porosity, filler particle porosity, axial diffusion coefficient, pore diffusion coefficient, surface diffusion coefficient, membrane mass transfer coefficient, equilibrium dissociation constant and filler maximum adsorption capacity. The preset parameter search space can be a search space generated in advance based on the value range corresponding to each fourth chromatography column parameter information. The value range can be determined based on the historical value of each fourth chromatography column parameter information. The position of the particle corresponds to the value of each set of fourth chromatography column parameter information one by one. Particles at different positions represent different values ​​of the fourth chromatography column parameter information, so that each particle represents a set of possible parameter solutions. The fourth chromatography column parameter information to be selected refers to possible parameter solutions.

[0081] Specifically, a certain number of particles N can be randomly generated in the preset parameter search space particles , thereby obtaining initial multiple sets of candidate fourth chromatography column parameter information.

[0082] S242, predicting a candidate penetration curve corresponding to each particle based on the target network model, the candidate fourth chromatography column parameter information represented by each particle, the current third chromatography column parameter information, and the current second chromatography column parameter information, and determining an error value between the candidate penetration curve and the partial penetration curve.

[0083] The candidate penetration curve refers to a predicted penetration curve generated by the chromatography column having the candidate fourth chromatography column parameter information. The candidate fourth chromatography column parameter information corresponds one-to-one to the candidate penetration curve.

[0084] Specifically, the parameter information of the fourth chromatography column to be selected represented by each particle, as well as the parameter information of the current third chromatography column and the current second chromatography column parameter information that have been determined before are input into the target network model for direct or indirect curve prediction, and the penetration curve to be selected corresponding to each particle is obtained. Since the predicted penetration curve to be selected is the complete penetration curve of the entire sampling time period, and the partial penetration curve is the local penetration curve of the current sampling time period, the penetration curve to be selected is intercepted to obtain the penetration curve to be selected in the current sampling time period. Based on a preset error function, such as a mean square error function, the error value between the penetration curve to be selected and the partial penetration curve in the current sampling time period is determined.

[0085] S243. Based on the error value corresponding to each particle, the individual optimal solution and the global optimal solution are updated, and the inertia weight and acceleration coefficient of each particle are adjusted.

[0086] Among them, the inertia weight is used to control the particle speed to balance the global and local search capabilities. The acceleration coefficient can include an individual acceleration coefficient and a global acceleration coefficient. The individual acceleration coefficient is used to adjust the tendency of the particle to move toward the individual optimal solution. The global acceleration coefficient is used to adjust the tendency of the particle to move toward the global optimal solution.

[0087] Specifically, for each example, the error value corresponding to the particle is compared with the historical best error value corresponding to the individual optimal solution. If the error value corresponding to the particle is less than the historical best error value, the individual optimal solution is updated to the parameter information of the fourth chromatography column to be selected corresponding to the particle, and the historical best error value is updated to the error value corresponding to the particle, thereby realizing the update of the individual optimal solution. After the individual optimal solution corresponding to each particle is updated, the error values ​​corresponding to all individual optimal solutions are compared, and the individual optimal solution with the smallest error value is used as the latest global optimal solution, thereby realizing the update of the global optimal solution.

[0088] In the process of adaptive particle swarm optimization, the inertia weight can be dynamically adjusted according to the algorithm's iteration process or the performance of the particles to balance the algorithm's exploration and development capabilities, and the individual acceleration coefficient and the global acceleration coefficient can be adjusted to promote the movement of particles to the area where the global optimal solution is located. By adaptively adjusting the inertia weight and acceleration coefficient of each particle, the particle's speed and position update strategy can be adjusted to enhance the search capability and avoid premature convergence.

[0089] S244, based on the adjusted inertia weight and acceleration coefficient and the updated individual optimal solution and global optimal solution, determine the new position of each particle, and perform iterative processing based on the new position of each particle until the iteration stop condition is met, and use the current global optimal solution as the current fourth chromatography column parameter information.

[0090] Specifically, based on the adjusted inertia weight and acceleration coefficient and the updated individual optimal solution and global optimal solution, the position of each particle is updated to obtain a new position of each particle. Based on the new position of each particle, that is, the new candidate fourth chromatography column parameter information corresponding to each particle is returned to perform the above steps S242-S244, so as to perform the next update iteration on the global optimal solution and the particle position until the iteration stop condition is met, such as reaching the maximum number of iterations, the error value reaches a preset threshold, or the error value changes tend to be stable, the iteration is stopped, and the current global optimal solution is used as the final current fourth chromatography column parameter information.

[0091] Exemplarily, the speed update formula in the adaptive particle swarm optimization process is:

[0092]

[0093] in, is the position of particle i at time d; is the speed of particle i at time d; w is the inertia weight, which is used to control the particle speed to balance the global and local search capabilities; c1 is the individual acceleration coefficient, which adjusts the tendency of the particle to move toward the individual optimal solution; is the optimal individual position of particle i at time d; c2 is the global acceleration coefficient, which adjusts the tendency of particles to move toward the global optimal solution; is the optimal global position of particle i at time d; and is a random number in the interval [0,1] to provide randomness for each iteration.

[0094] The position update formula in the adaptive particle swarm optimization process is:

[0095]

[0096] In the adaptive particle swarm optimization process, the inertia weight w is dynamically adjusted according to the algorithm's iteration process or the performance of the particles to balance the algorithm's exploration and development capabilities, and c1 and c2 are adjusted to promote the movement of particles to the area where the global optimal solution is located.

[0097] Figure 3 A flowchart of another chromatography column loading control method provided by an embodiment of the present invention, based on the above embodiments, this embodiment describes in detail the training process of the target network model before the chromatography column starts to load the sample. The explanations of the terms that are the same or corresponding to the above embodiments are not repeated here.

[0098] See also Figure 3 Another chromatography column loading control method provided in this embodiment specifically comprises the following steps:

[0099] S310: Generate a training data set, and build a target network model to be trained based on the target parameter values ​​of the model structure hyperparameters.

[0100] The training data set is a data set used to train the target network model. The model structure hyperparameter is a model structure parameter that needs to be determined in advance. The target parameter value is the final parameter value of the model structure hyperparameter. The target parameter value can be pre-set based on business requirements or the optimal parameter value determined through performance analysis. Model structure hyperparameters with different parameter values ​​correspond to different network models.

[0101] Exemplarily, the target network model may be a residual network model including a plurality of residual network blocks, wherein each residual network block includes a plurality of fully connected layers, and each fully connected layer includes a plurality of neurons. Accordingly, the model structure hyperparameters may include: the number of residual network blocks N block , the number of fully connected layers N layer and the number of neurons N neuron The structure of each residual network block is "activation-full connection-activation-full connection", with a total of N layer A fully connected layer is first activated with an activation function, which can help reduce the problem of vanishing gradients. Each residual network block passes information directly by adding the input to the output through identity mapping, and learns the residual between the input and output through an additional path, avoiding the problem of vanishing gradients in deep networks and improving the stability and efficiency of network model training.

[0102] Specifically, according to the input parameters and output parameters of the target network model to be trained, a training data set matching the target network model to be trained is generated. Based on the target parameter values ​​of the model structure hyperparameters, a specific network architecture is constructed to obtain the target network model to be trained.

[0103] Exemplarily, when the target network model indirectly determines the complete penetration curve by predicting the protein concentration, the training data set may include: multiple groups of sample chromatography column parameter information and actual protein concentration parameter information corresponding to each group of sample chromatography column parameter information. Among them, each group of sample chromatography column parameter information may include: sample first chromatography column parameter information related to the chromatography column and filler, and sample second chromatography column parameter information related to time and space. The actual protein concentration parameter information is the output label during model training. The actual protein concentration parameter information may include: the mobile phase protein concentration between particles, the mobile phase protein concentration within the particles, and the solid phase protein concentration within the particles.

[0104] Exemplarily, the “generating a training data set” in step S310 may include the following steps S311-S314:

[0105] S311. Based on the value range corresponding to each first chromatography column parameter information related to the chromatography column and the filler, sample and combine each first chromatography column parameter information to obtain multiple groups of sample first chromatography column parameter information.

[0106] Among them, the value range corresponding to each first chromatography column parameter information is determined in advance based on actual needs, and the value range composed of the minimum value and the maximum value has actual physical meaning. Specifically, taking each first chromatography column parameter information as a dimension, sampling and combination are performed in the multidimensional space composed of all the first chromatography column parameter information to obtain multiple groups of sample first chromatography column parameter information. For example, in the multidimensional space, sampling and combination are performed using the method based on the fractional factorial design (Fractional Factorial Design) in the design of experiments (Design of Experiments, DOE), that is, each first chromatography column parameter information is used as a factor, and each factor consists of two levels -1 and 1 corresponding to the maximum value and the minimum value, respectively, which are used to define the boundaries and vertices of the multidimensional space, so that the first chromatography column parameter information of multiple groups of samples can be obtained using the fractional factorial design. And / or, the sample points of the multidimensional parameter space are generated in the multidimensional space by Latin Hypercube Sampling (Latin Hypercube Sampling, LHS). The Latin hypercube sampling method includes the steps of separation, random sampling and combination. The separation step is to divide the value range of the first column parameter information into N input The random sampling step is to randomly select a value in each interval of each dimension, and generate N input The combination step refers to the N random values ​​drawn in each dimension. input The random values ​​are randomly combined to ensure that the value of each sample point in each dimension comes from a different interval, thereby obtaining multiple sample points, that is, multiple groups of sample first chromatography column parameter information.

[0107] S312, using the first chromatography column parameter information of each group of samples as input parameters, performing a chromatography column global solution on the universal rate model, and obtaining a continuous space solution including the second chromatography column parameter information.

[0108] Wherein, the general rate model (General Rate Model, GRM) is used to describe the physical laws and details of the chromatography process, such as intra-particle diffusion (pore diffusion), inter-particle diffusion (film diffusion) and adsorption kinetics, so that the material transfer process in the column can be more accurately described and predicted using the general rate model. Exemplarily, the general rate model is a group of partial differential equations with boundary conditions and initial conditions. The group of partial differential equations includes three partial differential equations, respectively: a partial differential equation for describing the mass transfer process in the chromatography column, a partial differential equation for describing the mass transfer process in the particle, and an adsorption isotherm equation for describing the adsorption process in the particle, such as the Langmuir equation. There are four boundary conditions in the general rate model, namely, the chromatography column inlet boundary condition, the chromatography column outlet boundary condition, the column filler particle surface boundary condition, and the column filler particle internal core boundary condition. There are three initial conditions in the general rate model, i.e., the initial values ​​of the three output parameters.

[0109] The input parameters of the universal rate model include: the first column parameter information related to the column and the filler, and the second column parameter information related to time and space. The output parameters of the universal rate model include: the mobile phase protein concentration between particles, the mobile phase protein concentration within the particles, and the solid phase protein concentration within the particles. The output parameters of the universal rate model are the output parameters of the target network model.

[0110] For example, the partial differential equation describing the mass transfer process in a chromatography column is:

[0111]

[0112] The partial differential equation used to describe the mass transfer process within the particle is:

[0113]

[0114] The adsorption isotherm equation used to describe the adsorption process within the particles is:

[0115] The column inlet boundary condition is:

[0116] The boundary condition at the column outlet is:

[0117] The boundary conditions on the surface of the filler particles in the column are:

[0118]

[0119] The core boundary conditions inside the column filler particles are:

[0120] The initial conditions are: c(t=0)=0; c p (t=0)=0;q(t=0)=0

[0121] Wherein, L is the height of the chromatography column; r p is the average radius of the filler particles; c0 is the sample protein concentration; u is the apparent flow rate; ε is the porosity of the chromatography column; ε p is the porosity of filler particles; D ax is the axial diffusion coefficient; D p is the pore diffusion coefficient; D s is the surface diffusion coefficient; k f is the membrane mass transfer coefficient; K d is the equilibrium dissociation constant; q max is the maximum adsorption capacity of the filler; these parameters are the first chromatography column parameter information that needs to be input into the general rate model.

[0122] t is the process time; r is the radial coordinate in the particle; x is the axial coordinate in the chromatography column; these parameters are the parameter information of the second chromatography column that needs to be input into the general rate model.

[0123] c is the protein concentration of the mobile phase between particles; c p is the concentration of the mobile phase protein in the particles; q is the concentration of the solid phase protein in the particles; these parameters are the output parameters of the general rate model.

[0124] Specifically, the first column parameter information of each group of samples is used as the input universal rate model, and the universal rate model with input parameter values ​​is solved in the entire column using numerical solution methods such as finite element method, finite difference method or orthogonal collocation method to obtain a spatial solution composed of the second column parameter information with continuous values, that is, a three-dimensional spatial solution composed of x, r, and t. The three-dimensional spatial solution includes: the mobile phase protein concentration c between particles corresponding to any axial coordinate in the column at any sample loading time, and the mobile phase protein concentration c in particles corresponding to any axial coordinate in the column and any radial coordinate in particles at any sample loading time. p and solid phase protein concentration q.

[0125] S313, grid point sampling is performed on the continuous space solution to obtain the sample second chromatography column parameter information and actual protein concentration parameter information corresponding to each grid point after sampling.

[0126] Specifically, for the continuous space solution solved based on the first chromatography column parameter information of each group of samples, at the initial condition (t = 0), the chromatography column inlet boundary (x = 0), the chromatography column outlet boundary (x = L), the particle core boundary (r = 0) and the particle surface (r = r p ) In the two-dimensional planes of the other two dimensions corresponding to the five boundaries (for example, t = 0 corresponds to the two-dimensional plane composed of x and r, x = 0 corresponds to the two-dimensional plane composed of t and r, etc.), N points are taken in each dimension in a uniform way and then combined, and a total of N points are taken 2 points, that is, N 2 The second column parameter information of each sample and the actual protein concentration parameter information corresponding to the second column parameter information of each sample. Among them, the particle surface boundary (r = rp) is the main location where the mobile phase between particles and the mass transfer process within the particles are generated in the general rate model, and the sampling density of the corresponding two-dimensional plane composed of x and t is greater than that of other boundaries.

[0127] In the three-dimensional space solution composed of x, r, t (excluding the boundary), the following three methods can be used to sample grid points:

[0128] The first method is uniform sampling: N samples are taken in each of the three dimensions x, r, and t. x , N r , N t Combine the points to get N x *N r *N t Grid points.

[0129] The second method is Latin hypercube sampling: grid point sampling is performed using the above LHS sampling including separation, random sampling and combination steps.

[0130] The third method is the penetration zone sampling method: divide the x dimension into N equal parts, which is equivalent to physically dividing the chromatography column into N sections, and confirm whether the mobile phase protein concentration c between the largest particles reaches 1% of the loaded protein concentration at the end of the loading time in each section. If it reaches, this section is defined as a penetration section, and if it does not reach, this section is defined as a non-penetration section. In each penetration section, take the area corresponding to the 5%, 50% and 95% points of the mobile phase protein concentration between the largest particles in the penetration section, and record it as the penetration zone. Take the numerical solutions of the three points of 5%, 50% and 95% and the grid points in the penetration zone for uniform sampling.

[0131] By using at least one of the above methods, the second chromatography column parameter information of multiple samples in the three-dimensional space solution excluding the boundary and the actual protein concentration parameter information corresponding to the second chromatography column parameter information of each sample can be obtained.

[0132] S314, normalizing the sample first chromatography column parameter information, the sample second chromatography column parameter information and the actual protein concentration parameter information to obtain a training data set.

[0133] Specifically, the first chromatography column parameter information of the sample, the second chromatography column parameter information of the sample and the actual protein concentration parameter information are normalized so that the normalized information is within the range of -1 to 1. For example, for the maximum and minimum values ​​with a magnitude difference within 1000 times, a linear formula can be used: Normalize to the range -1 to 1. For magnitude differences of more than 1000 times, the logarithmic formula can be used Normalized to the interval of -1 to 1. After normalization, the sample first chromatography column parameter information and the sample second chromatography column parameter information corresponding to the sample first chromatography column parameter information are combined into the sample chromatography column parameter information, and the actual protein concentration parameter information corresponding to the sample first chromatography column parameter information is used as the actual protein concentration parameter information corresponding to the sample chromatography column parameter information, thereby obtaining a training data set.

[0134] S320: Based on the training data set and the target parameter values ​​of the model training hyperparameters, the target network model to be trained is trained to obtain a trained target network model.

[0135] The model training hyperparameters are model training parameters that need to be determined in advance. The target parameter value is the final parameter value of the model training hyperparameters. The target parameter value can be pre-set based on business needs or the optimal parameter value determined through performance analysis. Model training hyperparameters with different parameter values ​​correspond to different model training processes.

[0136] Exemplarily, the model training hyperparameters may include: the number of samples in a single iteration, the number of training cycles in the initial stage, the number of training cycles in the later stage, the initial learning rate, and the final learning rate. The number of samples in a single iteration refers to the number of samples that need to be processed in one iteration. The number of training cycles in the initial stage refers to the number of cycles required for training in the initial stage, and the entire training data set is processed once as one cycle. The number of training cycles in the later stage refers to the number of cycles required for training in the later stage.

[0137] Specifically, based on the training data set and the target parameter values ​​of the model training hyperparameters, supervised model training is performed on the target network model to be trained to obtain the trained target network model to ensure the accuracy of the prediction information of the target network model.

[0138] Exemplarily, step S320 may include: in the initial stage, based on the first loss function, the training data set and the target parameter value of the model training hyperparameter, performing initial training on the target network model to be trained to obtain an initial network model after the initial training; wherein the first loss function is determined based on the model prediction loss sub-function, the universal rate model loss sub-function, the first weight corresponding to the model prediction loss sub-function, and the second weight corresponding to the universal rate model loss sub-function, and the first weight is greater than the second weight;

[0139] In the later stage, the initial network model is post-trained based on the second loss function, the training data set and the target parameter values ​​of the model training hyperparameters to obtain the target network model after the later training; wherein the second loss function is based on the model prediction loss sub-function, the universal rate model loss sub-function, the third weight corresponding to the model prediction loss sub-function, and the fourth weight corresponding to the universal rate model loss sub-function, the third weight is less than the first weight, and the fourth weight is greater than the second weight.

[0140] Among them, the model prediction loss sub-function is used to calculate the error value between the output result of the model to be trained and the actual result. The general rate model loss sub-function is the residual obtained by substituting the output result of the model to be trained into the general rate model as the objective function. The model prediction loss sub-function and the general rate model loss sub-function are weighted and summed to obtain the loss function. The weight value in the first loss function used in the initial training is different from the weight value in the second loss function used in the later training.

[0141] It should be noted that introducing the universal rate model into the training as part of the loss function can ensure that the learned solution conforms to the given universal rate model, so that the physical process of tomography can be accurately predicted and a physical information neural network, namely the target network model, can be formed.

[0142] Specifically, in the initial stage, the first weight corresponding to the model prediction loss sub-function is greater than the second weight corresponding to the general rate model loss sub-function, so as to use the specific information of the actual result (i.e., the numerical solution) to quickly guide the model to converge to a reasonable range of the solution. For example, the target network model to be trained is trained by minimizing the first loss function using the gradient descent algorithm or its variant Adam optimizer. Among them, the learning rate decreases exponentially from high to low with the number of training cycles to improve the training effect.

[0143] In the later stage, when the model begins to converge to a global unique solution, the weight corresponding to the model prediction loss sub-function is reduced, that is, the third weight is less than the first weight, and the weight corresponding to the general rate model loss sub-function is increased, that is, the fourth weight is greater than the second weight, so that the model pays more attention to satisfying the laws of physics and reduces the dependence on the actual results (i.e., the numerical solution), thereby correcting any potential errors introduced by the numerical solution. For example, the initial network model is trained by minimizing the second loss function using the gradient descent algorithm or its variant Adam optimizer. By dynamically adjusting the weight value of the loss function throughout the training process, the training effect can be further optimized and the accuracy of the model prediction information can be further improved.

[0144] S330, after the chromatography column starts to be loaded with sample, determine the current protein concentration at the outlet of the chromatography column.

[0145] S340, based on the current protein concentration and the loading protein concentration of the chromatography column, determine the current penetration ratio corresponding to the current loading moment, and when the current penetration ratio reaches the preset penetration ratio, determine the partial penetration curve corresponding to the current loading time period.

[0146] S350, performing curve fitting based on the target network model and the partial penetration curve to determine current chromatography column parameter information.

[0147] S360: predicting a complete penetration curve corresponding to the chromatography column based on the target network model and current chromatography column parameter information.

[0148] S370: Determine a target sample loading time corresponding to the target penetration ratio based on the complete penetration curve and the target medium utilization rate.

[0149] S380, when the current sample loading time reaches the target sample loading time, stop loading the sample on the chromatography column.

[0150] The technical solution of this embodiment constructs a target network model to be trained based on the target parameter values ​​of the model structure hyperparameters, and trains the target network model to be trained based on the training data set and the target parameter values ​​of the model training hyperparameters, thereby ensuring the model training results and further ensuring the accuracy of the prediction information of the trained target network model.

[0151] On the basis of the above technical solutions, before "building a target network model to be trained based on the target parameter values ​​of the model structure hyperparameters" in step S310, the following steps S315-S37 may also be included:

[0152] S315. Construct multiple candidate network models based on multiple candidate parameter values ​​of the model structure hyperparameters.

[0153] The candidate parameter value may refer to a parameter value that can be selected for a model structure hyperparameter. Specifically, based on the fully factorial design method in the design of experiments (DOE), multiple candidate parameter values ​​for each model structure hyperparameter may be combined, and a candidate network model with a corresponding structure may be constructed based on each combination result.

[0154] For example, after constructing each candidate network model, the number of trainable parameters Q of each candidate network model can be determined. params In order to keep the training data density of each candidate network model the same in subsequent comparisons, according to the number of trainable parameters Q params 10% is set as the number of samples in the training data set with the first column parameter information N training , generate training data sets, and obtain training data sets corresponding to each candidate network model.

[0155] S316: Based on the training data set corresponding to each candidate network model, each candidate network model is trained, and based on the same test data set, each candidate network model after training is tested to obtain a test performance indicator value corresponding to each candidate network model.

[0156] Among them, the test data set is a data set different from the training data set, which is used to test the performance of the trained model. For example, the test data set may include: a global test set and an application test set. The generation method of the global test set is the same as that of the training data set described above, which will not be repeated here. The application test set takes all other input parameters when x=1 and r=1 and the corresponding numerical solution of the mobile phase protein concentration c between particles as output parameters, as a test set of the mobile phase protein concentration c at the outlet of the chromatography column, so as to only test the prediction performance at the outlet. The test performance indicators may include but are not limited to: the model prediction loss value on the global test set (i.e., the root mean square error between the model prediction value and the numerical solution), the model prediction loss value on the application test set, the universal rate model loss value on the global test set, and the time used for each model prediction.

[0157] Specifically, all candidate network models are trained using the same model training hyperparameters in their respective training data sets, and the test performance indicator value corresponding to each candidate network model is determined in the same data test set.

[0158] S317. Based on the test performance indicator value corresponding to each candidate network model, determine the change relationship between the test performance indicator and the model structure hyperparameter, and based on the change relationship, determine the target parameter value of the model structure hyperparameter.

[0159] Specifically, analysis of variance (Analysis of Variance, ANOVA) can be sampled to determine the F value, p value, degree of freedom, mean square, total square and other results between each test performance index and each model structure hyperparameter, and based on the result, determine whether there is a significant difference, determine the degree of influence of each model structure hyperparameter on each test performance index, and after obtaining significant influencing factors, the change relationship between the model structure hyperparameter and the test performance index can be studied by the method of curve fitting. Based on the change relationship, regression analysis and response surface analysis are performed to determine the optimal target parameter value of the model structure hyperparameter, that is, the parameter value of the model structure hyperparameter when the test performance index is the highest. The model structure hyperparameters can also be optimized one by one in combination with the one-factor method (One Factor At a Time, OFAT) to obtain the optimal target parameter value of the model structure hyperparameter.

[0160] Exemplarily, based on the network model with the optimal model structure hyperparameters obtained after the optimization is completed, the training model training hyperparameters can be further optimized to determine the target parameter values ​​of the model training hyperparameters. The process of determining the target parameter values ​​of the model training hyperparameters is the same as the process of determining the target parameter values ​​of the model structure hyperparameters. For example, based on the full factorial method in the design experiment, the model training hyperparameters are combined, trained on their corresponding training data sets, and the test performance indicators are collected and analyzed on the same test data set to find the optimal parameter value combination of the model training hyperparameters, thereby completing the optimization of the model training and ensuring the model training effect.

[0161] The following is an embodiment of a chromatography column loading control device provided in an embodiment of the present invention. The device and the chromatography column loading control method of the above-mentioned embodiments belong to the same inventive concept. For details not fully described in the embodiment of the chromatography column loading control device, reference can be made to the embodiment of the above-mentioned chromatography column loading control method.

[0162] Figure 4 The schematic diagram of the structure of a chromatography column loading control device provided in an embodiment of the present invention is applicable to the case of online optimization control of the chromatography loading process of the chromatography column. Figure 4 As shown, the device specifically includes: a current protein concentration determination module 410, a partial penetration curve determination module 420, a current chromatography column parameter information determination module 430, a complete penetration curve prediction module 440, a target loading time determination module 450 and a chromatography column loading stop module 460.

[0163] Among them, the current protein concentration determination module 410 is used to determine the current protein concentration at the outlet of the chromatography column after the chromatography column starts to be loaded; the partial penetration curve determination module 420 is used to determine the current penetration ratio corresponding to the current loading moment based on the current protein concentration and the loading protein concentration of the chromatography column, and when the current penetration ratio reaches the preset penetration ratio, determine the partial penetration curve corresponding to the current loading time period; the current chromatography column parameter information determination module 430 is used to perform curve fitting based on the target network model and the partial penetration curve to determine the current chromatography column parameter information, wherein, The target network model is obtained by pre-training and is used to predict the complete penetration curve corresponding to the chromatography column parameter information; the complete penetration curve prediction module 440 is used to predict the complete penetration curve corresponding to the chromatography column based on the target network model and the current chromatography column parameter information; the target loading time determination module 450 is used to determine the target loading time corresponding to the target penetration ratio based on the complete penetration curve and the target medium utilization rate; the chromatography column stopping loading module 460 is used to stop loading the chromatography column when the current loading time reaches the target loading time.

[0164] The technical solution of this embodiment is to determine the current protein concentration at the outlet of the chromatography column in real time after the chromatography column starts to be loaded, and determine the current penetration ratio corresponding to the current loading time based on the current protein concentration and the loading protein concentration of the chromatography column, and when the current penetration ratio reaches the preset penetration ratio, determine the partial penetration curve corresponding to the current loading time period; based on the target network model obtained by pre-training and the partial penetration curve, curve fitting is performed to quickly and accurately determine the current chromatography column parameter information of the chromatography column, and based on the target network model and the current chromatography column parameter information, the complete penetration curve corresponding to the chromatography column can be accurately predicted, so that it will not be affected by the attenuation of the chromatography column filler; based on the complete penetration curve and the target penetration ratio corresponding to the target medium utilization rate, the target loading time corresponding to the target penetration ratio is determined, and when the current loading time reaches the target loading time, the chromatography column is stopped from loading, so that by controlling the loading time of the chromatography column loading process, it can be ensured that the target medium utilization rate can be achieved in each loading process, and the online optimization control of the chromatography column loading is realized.

[0165] Optionally, the current protein concentration determination module 410 is specifically configured to:

[0166] The current ultraviolet value at the outlet of the chromatography column is obtained; based on the pre-established correspondence between the protein concentration and the ultraviolet value and the current ultraviolet value, the current protein concentration at the outlet of the chromatography column is determined.

[0167] Optionally, the current chromatography column parameter information includes: current first chromatography column parameter information related to the chromatography column and filler, and current second chromatography column parameter information related to time and space;

[0168] Wherein, the current first chromatography column parameter information includes: the current third chromatography column parameter information under the first type and the current fourth chromatography column parameter information under the second type;

[0169] The current chromatography column parameter information determination module 430 includes:

[0170] A first information determination unit, configured to obtain the current third chromatography column parameter information and the current second chromatography column parameter information;

[0171] The second information determination unit is used to perform curve fitting based on the target network model and the partial penetration curve to determine the current fourth chromatography column parameter information.

[0172] Optionally, the second information determining unit is specifically configured to:

[0173] Based on the adaptive particle swarm optimization method, the error value between the candidate penetration curve predicted by the target network model and the partial penetration curve is minimized to obtain the current fourth chromatography column parameter information.

[0174] Optionally, the second information determining unit is specifically configured to:

[0175] A plurality of particles are generated in a preset parameter search space, wherein the position of each particle represents a set of candidate fourth chromatography column parameter information; based on the target network model, the candidate fourth chromatography column parameter information represented by each particle, the current third chromatography column parameter information and the current second chromatography column parameter information, a candidate penetration curve corresponding to each particle is predicted, and an error value between the candidate penetration curve and the partial penetration curve is determined; based on the error value corresponding to each particle, an individual optimal solution and a global optimal solution are updated, and an inertia weight and an acceleration coefficient of each particle are adjusted; based on the adjusted inertia weight and acceleration coefficient and the updated individual optimal solution and global optimal solution, a new position of each particle is determined, and iterative processing is performed based on the new position of each particle until an iteration stop condition is met, and the current global optimal solution is used as the current fourth chromatography column parameter information.

[0176] Optionally, the current second chromatography column parameter information includes: sample loading process time, radial coordinates within the particle, and axial coordinates within the chromatography column;

[0177] The current third chromatography column parameter information includes: chromatography column height, average radius of filler particles, sample protein concentration and apparent flow rate;

[0178] The current fourth chromatography column parameter information includes: chromatography column porosity, filler particle porosity, axial diffusion coefficient, pore diffusion coefficient, surface diffusion coefficient, membrane mass transfer coefficient, equilibrium dissociation constant and filler maximum adsorption capacity.

[0179] Optionally, the complete penetration curve prediction module 440 is specifically used to:

[0180] The current chromatography column parameter information is input into the target network model to predict the protein concentration, and the mobile phase protein concentration between particles corresponding to each loading moment at the chromatography column outlet is obtained; based on the mobile phase protein concentration between particles and the loading protein concentration of the chromatography column, the complete penetration curve of the chromatography column during the entire loading process is determined.

[0181] Optionally, the device further comprises:

[0182] A training data set generation module is used to generate a training data set and build a target network model to be trained based on the target parameter values ​​of the model structure hyperparameters;

[0183] The model training module is used to train the target network model to be trained based on the training data set and the target parameter value of the model training hyperparameter to obtain the trained target network model.

[0184] Optionally, the target network model is a residual network model including a plurality of residual network blocks, wherein each residual network block includes a plurality of fully connected layers, and each fully connected layer includes a plurality of neurons;

[0185] The model structure hyperparameters include: the number of residual network blocks, the number of fully connected layers, and the number of neurons;

[0186] The model training hyperparameters include: the number of samples in a single iteration, the number of training cycles in the initial stage, the number of training cycles in the later stage, the initial learning rate and the final learning rate.

[0187] Optionally, the training data set includes: multiple groups of sample chromatography column parameter information and actual protein concentration parameter information corresponding to each group of sample chromatography column parameter information;

[0188] The column parameter information of each set of samples includes: the first column parameter information of the sample related to the column and the filler, and the second column parameter information of the sample related to time and space; the actual protein concentration parameter information includes: the mobile phase protein concentration between particles, the mobile phase protein concentration within the particles, and the solid phase protein concentration within the particles;

[0189] The training data set generation module is specifically used to: based on the value range corresponding to each first chromatography column parameter information related to the chromatography column and the filler, sample and combine each first chromatography column parameter information to obtain multiple groups of sample first chromatography column parameter information; use each group of sample first chromatography column parameter information as an input parameter, perform chromatography column global solution on the universal rate model, and obtain a continuous space solution containing second chromatography column parameter information; perform grid point sampling on the continuous space solution to obtain sample second chromatography column parameter information and actual protein concentration parameter information corresponding to each grid point after sampling; normalize the sample first chromatography column parameter information, the sample second chromatography column parameter information and the actual protein concentration parameter information to obtain a training data set.

[0190] Optionally, the universal rate model is a group of partial differential equations with boundary conditions and initial conditions, the group of partial differential equations including a partial differential equation for describing a mass transfer process in a chromatography column, a partial differential equation for describing a mass transfer process in a particle, and an adsorption isotherm equation for describing an adsorption process in a particle; the boundary conditions include a chromatography column inlet boundary condition, a chromatography column outlet boundary condition, a surface boundary condition of a filler particle in the column, and an internal core boundary condition of a filler particle in the column;

[0191] The input parameters of the general rate model include: first chromatography column parameter information related to the chromatography column and the filler, and second chromatography column parameter information related to time and space;

[0192] The output parameters of the general rate model include: the mobile phase protein concentration between particles, the mobile phase protein concentration within particles, and the solid phase protein concentration within particles.

[0193] Optionally, the model training module is specifically used to:

[0194] In the initial stage, based on the first loss function, the training data set and the target parameter value of the model training hyperparameter, the target network model to be trained is initially trained to obtain an initial network model after the initial training; wherein the first loss function is determined based on the model prediction loss sub-function, the universal rate model loss sub-function, the first weight corresponding to the model prediction loss sub-function, and the second weight corresponding to the universal rate model loss sub-function, and the first weight is greater than the second weight;

[0195] In the later stage, based on the second loss function, the training data set and the target parameter value of the model training hyperparameter, the initial network model is post-trained to obtain the target network model after the later training; wherein, the second loss function is based on the model prediction loss sub-function, the universal rate model loss sub-function, the third weight corresponding to the model prediction loss sub-function, and the fourth weight corresponding to the universal rate model loss sub-function. The third weight is less than the first weight, and the fourth weight is greater than the second weight.

[0196] Optionally, the device further comprises:

[0197] The model structure hyperparameter value determination module is used to construct multiple candidate network models based on multiple candidate parameter values ​​of the model structure hyperparameters; train each candidate network model based on the training data set corresponding to each candidate network model, and test each candidate network model after training based on the same test data set to obtain the test performance index value corresponding to each candidate network model; determine the change relationship between the test performance index and the model structure hyperparameter based on the test performance index value corresponding to each candidate network model, and determine the target parameter value of the model structure hyperparameter based on the change relationship.

[0198] The chromatography column loading control device provided in the embodiment of the present invention can execute the chromatography column loading control method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the chromatography column loading control method.

[0199] It is worth noting that in the embodiment of the above-mentioned chromatography column loading control device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0200] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 5 A block diagram of an exemplary electronic device 12 suitable for use in implementing embodiments of the present invention is shown. Figure 5 The electronic device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0201] like Figure 5 As shown, the electronic device 12 is in the form of a general purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0202] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0203] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0204] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory 30 (RAM) and / or cache memory 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 5 not shown, usually called a "hard drive"). Although Figure 5 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The system memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0205] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0206] The electronic device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), may communicate with one or more devices that enable a user to interact with the electronic device 12, and / or may communicate with any device that enables the electronic device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the electronic device 12 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the electronic device 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0207] The processing unit 16 executes various functional applications and data processing by running the programs stored in the system memory 28, such as implementing the steps of a chromatography column loading control method provided by any embodiment of the present invention.

[0208] Of course, those skilled in the art will appreciate that the processor may also implement the technical solution of the chromatography column loading control method provided in any embodiment of the present invention.

[0209] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the chromatography column loading control method provided in any embodiment of the present invention are implemented.

[0210] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0211] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0212] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0213] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0214] An embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the chromatography column loading control method provided by any embodiment of the present invention.

[0215] In the process of implementation, the computer program product can be written in one or more programming languages ​​or a combination thereof to perform the computer program code of the present invention, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0216] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0217] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A chromatography column loading control method, characterized in that: include: After the chromatography column starts to load the sample, determining the current protein concentration at the outlet of the chromatography column; Based on the current protein concentration and the loading protein concentration of the chromatography column, determining a current penetration ratio corresponding to the current loading moment, and when the current penetration ratio reaches a preset penetration ratio, determining a partial penetration curve corresponding to the current loading time period; Performing curve fitting based on the target network model and the partial penetration curve to determine the current chromatography column parameter information, wherein the target network model is obtained by pre-training and is used to predict the complete penetration curve corresponding to the chromatography column parameter information; Based on the target network model and the current chromatography column parameter information, predicting a complete penetration curve corresponding to the chromatography column; Determining a target sample loading time corresponding to the target penetration ratio based on the complete penetration curve and the target medium utilization rate; When the current sample loading time reaches the target sample loading time, the sample loading to the chromatography column is stopped.

2. The method according to claim 1, characterized in that Determining the current protein concentration at the outlet of the chromatography column comprises: Obtaining the current UV value at the outlet of the chromatography column; Based on the pre-established correspondence between the protein concentration and the ultraviolet value and the current ultraviolet value, the current protein concentration at the outlet of the chromatography column is determined.

3. The method according to claim 1, characterized in that The current chromatography column parameter information includes: current first chromatography column parameter information related to the chromatography column and the filler, and current second chromatography column parameter information related to time and space; Wherein, the current first chromatography column parameter information includes: the current third chromatography column parameter information under the first type and the current fourth chromatography column parameter information under the second type; The method of performing curve fitting based on the target network model and the partial penetration curve to determine the current chromatography column parameter information includes: Acquiring the current third chromatography column parameter information and the current second chromatography column parameter information; Curve fitting is performed based on the target network model and the partial penetration curve to determine the current fourth chromatography column parameter information.

4. The method according to claim 3, characterized in that The method of performing curve fitting based on the target network model and the partial penetration curve to determine the current fourth chromatography column parameter information includes: Based on the adaptive particle swarm optimization method, the error value between the candidate penetration curve predicted by the target network model and the partial penetration curve is minimized to obtain the current fourth chromatography column parameter information.

5. The method according to claim 4, characterized in that The method of minimizing the error between the candidate penetration curve predicted by the target network model and the partial penetration curve based on the adaptive particle swarm optimization method to obtain the parameter information of the current fourth chromatography column includes: Generate multiple particles in a preset parameter search space, where the position of each particle represents a set of parameter information of a fourth chromatography column to be selected; Based on the target network model, the candidate fourth chromatography column parameter information represented by each particle, the current third chromatography column parameter information and the current second chromatography column parameter information, predicting the candidate penetration curve corresponding to each particle, and determining the error value between the candidate penetration curve and the partial penetration curve; Based on the error value corresponding to each particle, the individual optimal solution and the global optimal solution are updated, and the inertia weight and acceleration coefficient of each particle are adjusted; Based on the adjusted inertia weight and acceleration coefficient and the updated individual optimal solution and global optimal solution, the new position of each particle is determined, and iterative processing is performed based on the new position of each particle until the iteration stop condition is met, and the current global optimal solution is used as the current fourth chromatography column parameter information.

6. The method according to claim 3, characterized in that The current second chromatography column parameter information includes: sample loading process time, radial coordinates within the particle, and axial coordinates within the chromatography column; The current third chromatography column parameter information includes: chromatography column height, average radius of filler particles, sample protein concentration and apparent flow rate; The current fourth chromatography column parameter information includes: chromatography column porosity, filler particle porosity, axial diffusion coefficient, pore diffusion coefficient, surface diffusion coefficient, membrane mass transfer coefficient, equilibrium dissociation constant and filler maximum adsorption capacity.

7. The method according to claim 1, characterized in that The predicting a complete penetration curve corresponding to the chromatography column based on the target network model and the current chromatography column parameter information includes: Inputting the current chromatography column parameter information into the target network model to predict the protein concentration, and obtaining the mobile phase protein concentration between particles corresponding to each loading moment at the chromatography column outlet; Based on the mobile phase protein concentration between the particles and the loading protein concentration of the chromatography column, a complete breakthrough curve of the chromatography column during the entire loading process is determined.

8. The method according to any one of claims 1 to 7, characterized in that: The training process of the target network model includes: Generate a training data set and build a target network model to be trained based on the target parameter values ​​of the model structure hyperparameters; Based on the training data set and the target parameter values ​​of the model training hyperparameters, the target network model to be trained is trained to obtain a trained target network model.

9. The method according to claim 8, characterized in that The target network model is a residual network model including a plurality of residual network blocks, wherein each residual network block includes a plurality of fully connected layers, and each fully connected layer includes a plurality of neurons; The model structure hyperparameters include: the number of residual network blocks, the number of fully connected layers, and the number of neurons; The model training hyperparameters include: the number of samples in a single iteration, the number of training cycles in the initial stage, the number of training cycles in the later stage, the initial learning rate and the final learning rate.

10. The method according to claim 8, characterized in that The training data set includes: multiple groups of sample chromatography column parameter information and actual protein concentration parameter information corresponding to each group of sample chromatography column parameter information; The column parameter information of each set of samples includes: the first column parameter information of the sample related to the column and the filler, and the second column parameter information of the sample related to time and space; the actual protein concentration parameter information includes: the mobile phase protein concentration between particles, the mobile phase protein concentration within the particles, and the solid phase protein concentration within the particles; The generating of the training data set comprises: Based on the value range corresponding to each first chromatography column parameter information related to the chromatography column and the filler, each first chromatography column parameter information is sampled and combined to obtain multiple groups of sample first chromatography column parameter information; The first column parameter information of each group of samples is used as input parameters, and the universal rate model is solved in the entire column domain to obtain a continuous space solution including the second column parameter information; Grid point sampling is performed on the continuous space solution to obtain parameter information of the second chromatography column and actual protein concentration parameter information of the sample corresponding to each grid point after sampling; The first chromatography column parameter information of the sample, the second chromatography column parameter information of the sample and the actual protein concentration parameter information are normalized to obtain a training data set.

11. The method according to claim 10, characterized in that The general rate model is a group of partial differential equations with boundary conditions and initial conditions, wherein the group of partial differential equations includes a partial differential equation for describing the mass transfer process in the chromatography column, a partial differential equation for describing the mass transfer process in the particles, and an adsorption isotherm equation for describing the adsorption process in the particles; the boundary conditions include the chromatography column inlet boundary conditions, the chromatography column outlet boundary conditions, the surface boundary conditions of the filler particles in the column, and the internal core boundary conditions of the filler particles in the column; The input parameters of the general rate model include: first chromatography column parameter information related to the chromatography column and the filler, and second chromatography column parameter information related to time and space; The output parameters of the general rate model include: the mobile phase protein concentration between particles, the mobile phase protein concentration within particles, and the solid phase protein concentration within particles.

12. The method according to claim 8, characterized in that The training of the target network model to be trained based on the training data set and the target parameter value of the model training hyperparameter to obtain the trained target network model includes: In the initial stage, based on the first loss function, the training data set and the target parameter value of the model training hyperparameter, the target network model to be trained is initially trained to obtain an initial network model after the initial training; wherein the first loss function is determined based on the model prediction loss sub-function, the universal rate model loss sub-function, the first weight corresponding to the model prediction loss sub-function, and the second weight corresponding to the universal rate model loss sub-function, and the first weight is greater than the second weight; In the later stage, based on the second loss function, the training data set and the target parameter value of the model training hyperparameter, the initial network model is post-trained to obtain the target network model after the later training; wherein, the second loss function is based on the model prediction loss sub-function, the universal rate model loss sub-function, the third weight corresponding to the model prediction loss sub-function, and the fourth weight corresponding to the universal rate model loss sub-function. The third weight is less than the first weight, and the fourth weight is greater than the second weight.

13. The method according to claim 8, characterized in that Before building the target network model to be trained based on the target parameter values ​​of the model structure hyperparameters, it also includes: Based on multiple candidate parameter values ​​of the model structure hyperparameters, multiple candidate network models are constructed; Based on the training data set corresponding to each candidate network model, each candidate network model is trained, and based on the same test data set, each candidate network model after training is tested to obtain a test performance index value corresponding to each candidate network model; Based on the test performance index value corresponding to each candidate network model, the change relationship between the test performance index and the model structure hyperparameter is determined, and based on the change relationship, the target parameter value of the model structure hyperparameter is determined.

14. A chromatography column loading control device, characterized in that: include: A current protein concentration determination module, used to determine the current protein concentration at the outlet of the chromatography column after the chromatography column starts to load the sample; A partial penetration curve determination module, for determining a current penetration ratio corresponding to a current loading moment based on the current protein concentration and the loading protein concentration of the chromatography column, and determining a partial penetration curve corresponding to a current loading time period when the current penetration ratio reaches a preset penetration ratio; a current chromatography column parameter information determination module, used to perform curve fitting based on a target network model and the partial penetration curve to determine the current chromatography column parameter information, wherein the target network model is obtained by pre-training and is used to predict the complete penetration curve corresponding to the chromatography column parameter information; A complete penetration curve prediction module, used to predict the complete penetration curve corresponding to the chromatography column based on the target network model and the current chromatography column parameter information; A target loading time determination module, used to determine a target loading time corresponding to the target penetration ratio based on the complete penetration curve and the target penetration ratio corresponding to the target medium utilization rate; The chromatography column sample loading stop module is used to stop loading the chromatography column when the current sample loading time reaches the target sample loading time.

15. An electronic device, characterized in that: The electronic device comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the chromatography column loading control method as described in any one of claims 1 to 13.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the chromatography column loading control method as described in any one of claims 1 to 13 is implemented.

17. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the chromatography column loading control method as described in any one of claims 1 to 13.

Images