Information processing apparatus, operation method for information processing apparatus, and operation program for information processing apparatus

Abstract

This information processing apparatus predicts, on the basis of spectrum data obtained by measuring electromagnetic waves emitted from a suspension in which a biological molecule is dispersed in a liquid as a component, the state of a target component in the suspension. The information processing apparatus is provided with a processor. The processor: acquires, by using a state prediction model calibrated by standard spectrum data obtained from a standard suspension in a standard measurement environment, target spectrum data obtained from a target suspension having an unknown state of a target component in a target measurement environment different from the standard measurement environment; acquires a correction coefficient derived from a comparison between the standard spectrum data and reference spectrum data that corresponds to the target spectrum data and that is obtained from the standard suspension in the target measurement environment; corrects the target spectrum data by using the correction coefficient to generate corrected target spectrum data; and causes, by applying the corrected target spectrum data to the state prediction model, the state prediction model to predict the state of the target component.

Description

Information processing device, information processing device operation method, and information processing device operation program

[0001] The technology of the present disclosure relates to an information processing device, an operating method for an information processing device, and an operating program for an information processing device.

[0002] For example, there are known manufacturing processes for biopharmaceuticals, which use biological molecules such as proteins, including monoclonal antibodies, as active ingredients. In these manufacturing processes, a suspension is often produced in which various components, including the active ingredient, are dispersed in a liquid. Monitoring the state (e.g., protein concentration or protein aggregate concentration) of target components (e.g., proteins or protein aggregates) in this suspension is important for the success of the ongoing manufacturing process.

[0003] The following technology has attracted attention as a technology for predicting the state of a target component because it involves fewer concerns about contamination. That is, this technology measures electromagnetic waves emitted from a suspension, such as Raman scattered light, and applies the resulting spectral data, such as Raman spectral data, to a state prediction model, such as a multivariate analysis or machine learning model, to allow the state of the target component to be predicted by the state prediction model.

[0004] Japanese Patent Laid-Open Publication No. 09-089775 describes a technique for correcting variations in Raman spectrum data due to fluctuations in the intensity of excitation light from a light source. Japanese Patent Laid-Open Publication No. 2022-552876 describes a technique for suppressing variations in multiple Raman spectrum data by performing derivative transformation and standard normal variation (SNV) on the multiple Raman spectrum data.

[0005] In the manufacturing process of biopharmaceuticals, the scale of the equipment used is gradually scaled up, such as initially using relatively small-scale equipment to set conditions, and then switching to relatively large-scale equipment once the conditions have been set. As the equipment scales up, the electromagnetic wave measurement environment also changes in various ways, for example, changing the flow cell, which is a measuring device for stably measuring electromagnetic waves.

[0006] When the electromagnetic wave measurement environment changes, the spectral data can sometimes differ even for suspensions of the same composition. As a result, while the predicted results for the state of the target component in suspensions of the same composition should be the same, different prediction results are obtained because the spectral data differs due to the measurement environment. In short, there is a risk that changes in the electromagnetic wave measurement environment could reduce the reliability of the predicted results for the state of the target component.

[0007] One embodiment of the technology disclosed herein provides an information processing device, an operating method for the information processing device, and an operating program for the information processing device that are capable of obtaining highly reliable prediction results for the state of a target component in a suspension even when the electromagnetic wave measurement environment changes.

[0008] The information processing device disclosed herein is an information processing device that predicts the state of a target component in a suspension based on spectral data obtained by measuring electromagnetic waves emitted from the suspension in which biological molecules are dispersed as components in a liquid, and is equipped with a processor. The processor uses a state prediction model calibrated with reference spectral data obtained from a reference suspension in a reference measurement environment to acquire target spectral data obtained from a target suspension in which the state of the target component is unknown in a target measurement environment different from the reference measurement environment, acquires a correction coefficient derived by comparing the reference spectral data with reference spectral data corresponding to the target spectral data, which is obtained from the reference suspension in the target measurement environment, corrects the target spectral data with the correction coefficient to generate corrected target spectral data, and applies the corrected target spectral data to the state prediction model, thereby causing the state prediction model to predict the state of the target component.

[0009] The correction factor is preferably derived based on the ratio of intensity values ​​of the standard spectral data and the reference spectral data corresponding to the components in the standard suspension.

[0010] The correction coefficients are preferably derived without the ratio of intensity values ​​corresponding to the measuring instruments of the electromagnetic waves.

[0011] The measuring device is preferably a flow cell having a flow path through which the suspension flows and an optical system for capturing electromagnetic waves, and the intensity value corresponding to the measuring device is preferably an intensity value corresponding to the optical system.

[0012] The reference measurement environment and the target measurement environment preferably differ in either the presence or absence of an electromagnetic wave measuring instrument and the type of instrument.

[0013] The measuring device is preferably a flow cell having a flow path through which the suspension flows and an optical system for capturing the electromagnetic waves.

[0014] There are a plurality of target measurement environments, and it is preferable that the plurality of target measurement environments use different types of flow cells.

[0015] It is preferable that correction coefficients are stored in advance in a memory unit for each of a plurality of target measurement environments, and the processor receives a specification of the type of flow cell used and obtains the correction coefficient corresponding to the specification by reading it from the memory unit.

[0016] The flow cells preferably differ in at least one of the distance between the incident surface of the optical system onto which the electromagnetic waves are incident and the wall surface of the flow channel facing the incident surface, the reflectivity of the wall surface, and the focal length of the optical system.

[0017] The state prediction model is preferably calibrated using the spectral data to be corrected.

[0018] The state of the target component is preferably the concentration of the target component.

[0019] Preferably, the component of interest is a protein.

[0020] Preferably, the protein is an antibody.

[0021] The electromagnetic wave is preferably Raman scattered light.

[0022] The state prediction model is preferably a machine learning model trained using the reference spectrum data as training data.

[0023] The method of operating an information processing device disclosed herein is a method of operating an information processing device that predicts the state of a target component in a suspension based on spectral data obtained by measuring electromagnetic waves emitted from the suspension in which biological molecules are dispersed as components in a liquid, and includes using a state prediction model calibrated with reference spectral data obtained from a reference suspension in a reference measurement environment, acquiring target spectral data obtained from a target suspension in which the state of the target component is unknown in a target measurement environment different from the reference measurement environment, acquiring a correction coefficient derived by comparing the reference spectral data with reference spectral data corresponding to the target spectral data, which is obtained from the reference suspension in the target measurement environment, correcting the target spectral data with the correction coefficient to generate corrected target spectral data, and applying the corrected target spectral data to the state prediction model to cause the state prediction model to predict the state of the target component.

[0024] The operating program of an information processing device disclosed herein is an operating program of an information processing device that predicts the state of a target component in a suspension based on spectral data obtained by measuring electromagnetic waves emitted from the suspension in which biological molecules are dispersed as components in a liquid, and causes a computer to execute processes including: using a state prediction model calibrated with reference spectral data obtained from a reference suspension in a reference measurement environment; acquiring target spectral data obtained from a target suspension in which the state of the target component is unknown in a target measurement environment different from the reference measurement environment; acquiring a correction coefficient derived by comparing the reference spectral data with reference spectral data corresponding to the target spectral data, which is obtained from the reference suspension in the target measurement environment; correcting the target spectral data with the correction coefficient to generate corrected target spectral data; and applying the corrected target spectral data to the state prediction model to cause the state prediction model to predict the state of the target component.

[0025] According to the technology disclosed herein, it is possible to provide an information processing device, an operating method for an information processing device, and an operating program for an information processing device that are capable of obtaining highly reliable prediction results for the state of a target component in a suspension even when the electromagnetic wave measurement environment changes.

[0026] 1 is a diagram illustrating a concentration unit, a measurement system, and an information processing device; FIG. 2 is a diagram illustrating excitation light and Raman scattered light; FIG. 3 is a diagram illustrating Raman spectral data; FIG. 4 is a cross-sectional view of a flow cell; FIG. 5 is a block diagram of a computer constituting an information processing device; FIG. 6 is a block diagram of a processing unit of a CPU of the information processing device; FIG. 7 is a diagram illustrating correction coefficient information; FIG. 8 is a diagram illustrating measurement of reference Raman spectral data and reference Raman spectral data; FIG. 9 is a table illustrating the contents of a reference suspension; FIG. 10 is a diagram illustrating correction coefficient derivation processing; FIG. 11 is a diagram illustrating a flow cell to be used selection screen; FIG. 12 is a diagram illustrating reading and acquiring from storage a correction coefficient according to the designation of the type of flow cell used; FIG. 13 is a diagram illustrating processing by the correction unit; FIG. 14 is a diagram illustrating processing by the prediction unit; FIG. 15 is a diagram illustrating a neural network constituting an antibody concentration prediction model; FIG. 16 is a diagram illustrating processing in the learning phase of the antibody concentration prediction model; FIG. 17 is a diagram illustrating a prediction result display screen; FIG. 18 is a flowchart illustrating the processing procedure of an information processing device; FIG. 19 is a diagram illustrating another example of correction coefficient information; FIG. 20 is a diagram illustrating yet another example of correction coefficient information; FIG. 21 is a diagram illustrating a second embodiment in which an antibody concentration prediction model is trained using Raman spectral data to be corrected; FIG. 22 is a diagram illustrating predicted values ​​and actual measured values ​​of antibody concentrations in examples, and their average errors; and FIG. 23 is a table illustrating the contents of a comparative example. 1 is a diagram showing predicted and measured antibody concentration values ​​and their average errors in Comparative Example 1. 2 is a diagram showing predicted and measured antibody concentration values ​​and their average errors in Comparative Example 2. 3 is a diagram showing predicted and measured antibody concentration values ​​and their average errors in Comparative Example 3.

[0027] 1, a measurement system 10 includes a flow cell 11 and a Raman spectrometer 12. The measurement system 10 is incorporated into a concentration unit 15 subsequent to a culture unit 13 and a purification unit 14 in a manufacturing system for an active pharmaceutical ingredient of a biopharmaceutical, for example.

[0028] The culture unit 13 includes a culture tank and a cell-removing filter. The culture tank stores a cell culture medium (medium). Antibody-producing cells are seeded into the cell culture medium and cultured in the cell culture medium. The culture method may be either perfusion culture or fed-batch culture. The antibody-producing cells are established by incorporating an antibody gene into host cells such as Chinese hamster ovary cells (CHO cells). The antibody-producing cells produce immunoglobulins, i.e., antibodies 16, during the culture process. Therefore, not only antibody-producing cells but also antibodies 16 are present in the cell culture medium. The antibodies 16 are, for example, monoclonal antibodies, and serve as active ingredients in biopharmaceuticals. The antibodies 16 are an example of a "component," a "target component," and a "protein" according to the technology of the present disclosure.

[0029] The cell removal filter captures antibody-producing cells in the cell culture solution with a filter membrane, for example, by tangential flow filtration (TFF) or alternating tangential flow filtration (ATF), and removes the antibody-producing cells from the cell culture solution. The cell removal filter also allows antibodies 16 to pass through. Therefore, the cell culture solution flowing from the culture section 13 to the purification section 14 mainly contains antibodies 16. The cell culture solution from which antibody-producing cells have been removed by the cell removal filter is called a culture supernatant. In addition to antibodies 16, the culture supernatant also contains cell-derived proteins, cell-derived DNA (deoxyribonucleic acid), aggregates 17 of antibodies 16, viruses, and the like. These cell-derived proteins, cell-derived DNA, aggregates 17 of antibodies 16, viruses, etc. are also examples of "components" according to the technology of the present disclosure.

[0030] The purification unit 14 sequentially performs component separation processes using a plurality of chromatography devices on the culture supernatant from the culture unit 13, thereby gradually removing contaminants and viruses and gradually increasing the purity of the antibody 16 in the culture supernatant. The plurality of chromatography devices may include, for example, an immunoaffinity chromatography device, a size exclusion chromatography device, a cation chromatography device, an anion chromatography device, a hydrophobic interaction chromatography device, etc.

[0031] The purifying unit 14 also performs a process to inactivate viruses in the culture supernatant and a process to remove viruses in the culture supernatant using a filter. A purified solution 18 obtained through these various processes is sent from the purifying unit 14 to the concentrating unit 15. Note that a filter using a single pass tangential flow filtration (SPTFF) method may be provided upstream of each of the chromatography devices.

[0032] The concentration section 15 is provided with a primary tank 20, a pump 21, a concentration filter 22, a secondary tank 23, a pressure control valve 24, etc. An inlet line 25, an outlet line 26, and a return line 27 are connected to the primary tank 20. The purified liquid 18 from the purification section 14 flows into the primary tank 20 through the inlet line 25. The primary tank 20 stores the purified liquid 18.

[0033] The pump 21 is provided in the outlet path 26. When the pump 21 is driven, the purified liquid 18 stored in the primary tank 20 is sent to the outlet path 26 and introduced into the concentration filter 22 through the outlet path 26. The concentration filter 22 subjects the introduced purified liquid 18 to concentration and filtration processes, for example, by ultrafiltration (UF) and diafiltration (DF).

[0034] Through the concentration and filtration process of the concentration filter 22, the purified liquid 18 is separated into a concentrate in which the purity of the antibody 16 has been further increased, and a waste liquid 28 containing water, solvent, etc. The concentrate is returned to the primary tank 20 through the return flow path 27. Therefore, the liquid stored in the primary tank 20 is a mixed liquid 18M of the purified liquid 18 from the purification unit 14 and the concentrated liquid from the concentration filter 22. The purified liquid 18 and the mixed liquid 18M are examples of the "suspension" and "target suspension" according to the technology of the present disclosure. Hereinafter, unless a particular distinction is required, the purified liquid 18 and the mixed liquid 18M will be collectively referred to as the target suspension 18T.

[0035] The waste liquid 28 is introduced into the secondary tank 23 through a waste liquid passage 29. The secondary tank 23 stores the waste liquid 28.

[0036] A pressure control valve 24 is provided in the return flow path 27. By controlling the opening and closing of this pressure control valve 24, the liquid pressure of the concentrated liquid passing through the return flow path 27 is maintained constant. Although not shown, pressure gauges are provided downstream of the pump 21 in the outflow path 26, upstream of the pressure control valve 24 in the return flow path 27, and in the waste liquid path 29. In addition, the primary tank 20 and the secondary tank 23 are placed on a weighing scale. The operation of the pump 21 and the pressure control valve 24 is controlled according to the pressure of each liquid measured by the pressure gauge and the weight of the primary tank 20 and the secondary tank 23 measured by the weighing scale.

[0037] The flow cell 11 is connected to the outlet channel 26 on the upstream side of the pump 21. The target suspension 18T from the primary tank 20 flows through the flow cell 11 at a preset flow rate.

[0038] As an example, as shown in FIG. 2 , the Raman spectrometer 12 is an instrument that evaluates a substance M by utilizing the characteristics of Raman scattered light RSL. When excitation light EL is irradiated onto the substance M, the excitation light EL interacts with the substance M, generating Raman scattered light RSL having a different wavelength from that of the excitation light EL. The wavelength difference between the excitation light EL and the Raman scattered light RSL corresponds to the energy of the molecular vibrations of the substance M. Therefore, Raman scattered light RSL with different wavenumbers can be obtained between substances M with different molecular structures. The Raman scattered light RSL is an example of an "electromagnetic wave" according to the technology of the present disclosure. Of the Stokes line and the anti-Stokes line, it is preferable to use the Stokes line for the Raman scattered light RSL.

[0039] The Raman spectrometer 12 is composed of a sensor unit 35 and an analyzer 36. The sensor unit 35 is connected to the flow cell 11. The sensor unit 35 emits excitation light EL from its tip. The excitation light EL is irradiated onto the target suspension 18T flowing through the flow cell 11. Raman scattered light RSL is generated by the interaction between this excitation light EL and components such as the antibodies 16 in the target suspension 18T. The sensor unit 35 receives the Raman scattered light RSL and outputs the received Raman scattered light RSL to the analyzer 36.

[0040] The analyzer 36 resolves the Raman scattered light RSL into wavenumbers and derives the intensity value of the Raman scattered light RSL for each wavenumber, thereby generating Raman spectrum data 37 representing the spectrum of the Raman scattered light RSL, i.e., the Raman spectrum, as shown in Fig. 3 as an example. The Raman spectrum data 37 is data in which the intensity value of the Raman scattered light RSL for each wavenumber is registered. In this example, the Raman spectrum data 37 is －１ ~1800cm －１ The intensity values ​​of the Raman scattered light RSL in the range from 1 cm －１ The graph shown below the Raman spectrum data 37 in Fig. 3 is obtained by plotting the intensity values ​​of the Raman spectrum data 37 for each wavenumber and connecting them with a line. The Raman spectrum data 37 is an example of "spectral data" according to the technology of the present disclosure.

[0041] In this way, the measurement system 10 flows the target suspension 18T through the flow cell 11. Then, by irradiating the target suspension 18T flowing through this flow cell 11 with excitation light EL via the sensor unit 35, the Raman scattered light RSL of components such as the antibody 16 in the target suspension 18T is measured, and Raman spectrum data 37 is obtained. Note that the measurement environment for the Raman scattered light RSL using the flow cell 11 shown in FIG. 1 is an example of a "target measurement environment" according to the technology of the present disclosure.

[0042] An information processing device 40 is connected to the Raman spectrometer 12 via a computer network such as a local area network (LAN). The Raman spectrometer 12 transmits measured Raman spectrum data 37 to the information processing device 40. The Raman spectrum data 37 transmitted from the Raman spectrometer 12 to the information processing device 40 is data for predicting the concentration of the antibody 16. The concentration of the antibody 16 is an example of the "state of the target component" according to the technology of the present disclosure. In the following description, the Raman spectrum data 37 transmitted from the Raman spectrometer 12 to the information processing device 40 will be referred to as target Raman spectrum data 37T. The target Raman spectrum data 37T is an example of the "target spectrum data" according to the technology of the present disclosure.

[0043] The information processing device 40 is, for example, a desktop personal computer, and includes a display 41 that displays various screens, and input devices 42 such as a keyboard, a mouse, a touch panel, and / or a microphone for voice input. The information processing device 40 is installed, for example, in a pharmaceutical company that develops biopharmaceuticals, or in an organization that is contracted by a pharmaceutical company to develop biopharmaceuticals, i.e., a contract research organization (CRO). The information processing device 40 is operated by a user U who is involved in the development of biopharmaceuticals at the pharmaceutical company or the contract research organization (hereinafter collectively referred to as a pharmaceutical facility).

[0044] As an example, as shown in Figure 4, the flow cell 11 is composed of a main body 50 and a sensor unit connector 51. The main body 50 is a cylindrical member having a linear flow path 52 with a circular cross section at its center. The main body 50 is made of metal, such as Hastelloy. Alternatively, the main body 50 may be made of resin, such as polyolefin resin.

[0045] A cylindrical boss-shaped first connector 53 and a second connector 54 are provided at the center of both end faces of the main body 50. The first connector 53 has an inlet 55 for the flow path 52, and the second connector 54 has an outlet 56 for the flow path 52. The first connector 53 and the second connector 54 are, for example, parallel threads or tapered threads. A sterile connector provided in the outlet path 26 is attached to the first connector 53 and the second connector 54 in a liquid-tight manner.

[0046] A mounting hole 57 is formed in the center of the circumferential surface of the main body 50 for detachably mounting the sensor unit connector 51 to the main body 50. A thread is formed on the inner circumferential surface of the mounting hole 57. The mounting hole 57 penetrates into the flow path 52. A thread is formed on the tip of the sensor unit connector 51 to screw into the thread on the inner circumferential surface of the mounting hole 57. By attaching the sensor unit connector 51 to the mounting hole 57 by screwing the threads together, the tip of the sensor unit connector 51 is accommodated in the mounting hole 57. The tip of the sensor unit 35 is detachably connected to the sensor unit connector 51.

[0047] The sensor unit connector 51 is a cylindrical member with an optical system 58 disposed at its tip. The optical system 58 is composed of a ball lens 59 and a transparent plate 60. The optical axes of the ball lens 59 and the transparent plate 60 are aligned. The ball lens 59 is literally a spherical lens and is made of, for example, sapphire glass or quartz glass. The ball lens 59 focuses the excitation light EL from the sensor unit 35 and also captures, into the sensor unit 35, Raman scattered light RSL generated by the interaction between the excitation light EL and the antibodies 16 and the like in the target suspension 18T.

[0048] Like ball lens 59, transparent plate 60 is made of, for example, sapphire glass or quartz glass. Transparent plate 60 is a circular plate in which a first surface 61 on the ball lens 59 side and a second surface 62 on the flow path 52 side are parallel to each other. Second surface 62 contacts target suspension 18T flowing through flow path 52. Second surface 62 is an example of an "incident surface" according to the technology of the present disclosure. Here, "parallel" refers to parallelism in the sense of including, in addition to perfect parallelism, an error that is generally acceptable in the technical field to which the technology of the present disclosure belongs and that does not contradict the spirit of the technology of the present disclosure.

[0049] The focusing position of the ball lens 59 is located, for example, on the second surface 62 of the transparent plate 60. The focusing position of the ball lens 59 is determined by the diameter, refractive index, focal length FL, and the like of the ball lens 59. In this case, the focusing position is point-like. By setting the focusing position of the ball lens 59 to a position on the second surface 62 of the transparent plate 60, it is possible to reduce the risk of the excitation light EL being attenuated by the target suspension 18T. Consequently, it is possible to maintain a high S / N ratio for the Raman spectrum data 37.

[0050] The wall surface 63 forming the flow path 52 is a smooth surface without irregularities of, for example, 1 mm or more. The wall surface 63A facing the second surface 62 of the transparent plate 60 has a reflectance R. The distance D between the second surface 62 and the wall surface 63A is approximately the same as the diameter of the flow path 52. The diameter of the flow path 52 is set depending on the flow rate and viscosity of the target suspension 18T flowing through the flow path 52, the size of the concentration filter 22, and the like. For example, the diameter of the flow path 52 is within a range such that the Reynolds number Re in the flow path 52 is 2300 or greater (Re≧2300) under the conditions that the flow rate of the target suspension 18T flowing through the flow path 52 is 200 cc / min or greater and the viscosity of the target suspension 18T is 0.001 Pa·s, the same as that of water. If the Reynolds number Re is 2300 or greater, turbulence occurs in the target suspension 18T flowing through the flow path 52. This reduces the bias of components in the target suspension 18T, and improves the measurement stability of the Raman spectrum data 37.

[0051] 5, the computer constituting the information processing device 40 includes, in addition to the display 41 and input device 42, a storage 70, a memory 71, a CPU (Central Processing Unit) 72, and a communication unit 73. These are interconnected via a bus line 74.

[0052] The storage 70 is a hard disk drive built into a computer constituting the information processing device 40 or connected via a cable or network. Alternatively, the storage 70 is a disk array consisting of multiple hard disk drives. The storage 70 stores control programs such as an operating system, various application programs, and various data associated with these programs. The storage 70 is an example of a "storage unit" according to the technology of the present disclosure. Note that a solid state drive may be used instead of a hard disk drive.

[0053] The memory 71 is a work memory for the CPU 72 to execute processing. The CPU 72 loads programs stored in the storage 70 into the memory 71 and executes processing in accordance with the programs. In this way, the CPU 72 comprehensively controls each part of the computer. The CPU 72 is an example of a "processor" according to the technology of the present disclosure. The memory 71 may be built into the CPU 72. The communication unit 73 controls the transmission of various information to and from external devices such as the Raman spectrometer 12.

[0054] As an example, as shown in Fig. 6, an operating program 80 is stored in the storage 70. The operating program 80 is an application program for causing a computer to function as the information processing device 40. In addition to the operating program 80, the storage 70 also stores correction coefficient information 81 and an antibody concentration prediction model 82. The antibody concentration prediction model 82 is an example of a "state prediction model" according to the technology of the present disclosure.

[0055] When the operating program 80 is started, the CPU 72 of the computer constituting the information processing device 40 works in cooperation with the memory 71 and the like to function as an acquisition unit 85, a read / write (hereinafter abbreviated as RW (Read Write)) control unit 86, an instruction receiving unit 87, a correction unit 88, a prediction unit 89, and a display control unit 90.

[0056] The acquisition unit 85 acquires the target Raman spectrum data 37T from the Raman spectrometer 12. The acquisition unit 85 outputs the target Raman spectrum data 37T to the RW control unit 86.

[0057] The RW control unit 86 controls the storage of various data in the storage 70 and the reading of various data stored in the storage 70. The RW control unit 86 stores the target Raman spectrum data 37T from the acquisition unit 85 in the storage 70. The RW control unit 86 also reads the target Raman spectrum data 37T and the correction coefficients CF (see FIG. 7 ) registered in the correction coefficient information 81 from the storage 70, and outputs the read target Raman spectrum data 37T and correction coefficients CF to the correction unit 88. The RW control unit 86 also reads the antibody concentration prediction model 82 from the storage 70 and outputs the read antibody concentration prediction model 82 to the prediction unit 89.

[0058] The instruction receiving unit 87 receives various operation instructions input by the user U via the input device 42 .

[0059] The corrector 88 corrects the target Raman spectrum data 37T using the correction coefficient CF to generate corrected target Raman spectrum data 37TC. The corrector 88 outputs the corrected target Raman spectrum data 37TC to the predictor 89.

[0060] The prediction unit 89 applies the corrected Raman spectral data 37TC to the antibody concentration prediction model 82, causing the antibody concentration prediction model 82 to predict the concentration of the antibody 16 in the target suspension 18T, and causes the antibody concentration prediction model 82 to output the prediction result 95. The prediction unit 89 outputs the prediction result 95 to the display control unit 90.

[0061] The display control unit 90 controls the display of various screens on the display 41. For example, the display control unit 90 controls the display 41 to display a used flow cell selection screen 115 (see FIG. 11 ) that allows the user U to select the flow cell 11 that has been used, a prediction result display screen 135 (see FIG. 17 ) that displays the prediction result 95, and the like.

[0062] 7 , the correction coefficient information 81 is information in which the correction coefficients CF of multiple types of flow cells 11 (product names FC1, FC2, FC3, ...) that can be used in a biopharmaceutical manufacturing system are comprehensively registered. The multiple types of flow cells 11 differ in at least one of the distance D between the second surface 62 of the transparent plate 60 and the wall surface 63A of the flow channel 52 facing the second surface 62, the reflectance R of the wall surface 63A, and the focal length FL of the ball lens 59. Note that the correction coefficient information 81 is not limited to cases in which a flow cell 11 is used; it is also possible to register a correction coefficient CF for cases in which the sensor unit 35 is directly immersed in a tank (chamber) that stores the target suspension 18T for measurement without using a flow cell 11.

[0063] In a biopharmaceutical manufacturing system, for example, multiple different types of flow cells 11 are selectively used depending on the scale of the equipment. Specifically, conditions are set using a relatively small-scale equipment, and when the setting is completed and the equipment is switched to a relatively large-scale equipment, the flow cell is changed to suit the large-scale equipment. Multiple measurement environments in which multiple different types of flow cells 11 are selectively used in this way are an example of "multiple target measurement environments" according to the technology of the present disclosure.

[0064] The correction coefficient CF is derived as follows. First, as shown in FIG. 8, reference suspensions 18ST are prepared in two tanks 100A and 100B. There are five types of reference suspensions 18ST, as shown in Table 105 in FIG. 9. Specifically, the reference suspensions 18ST include Reference Suspension 1, which is a 78 g / L solution of monoclonal antibody (mAb), Reference Suspension 2, which is a 50 g / L solution of bovine serum albumin (BSA), and Reference Suspension 3, which is a 25 g / L solution of bovine serum albumin. Furthermore, the reference suspensions 18ST include Reference Suspension 4, which is a 5 g / L solution of phenylalanine (Phe), and Reference Suspension 5, which is a 5 g / L solution of tryptophan (Trp).

[0065] The sensor unit 35 of the Raman spectrometer 12 is directly immersed in the reference suspension 18ST in the tank 100A, and the Raman scattered light RSL from the reference suspension 18ST is measured, thereby obtaining reference Raman spectrum data 37ST. This measurement environment for the Raman scattered light RSL without using the flow cell 11 is an example of a "reference measurement environment" according to the technology of the present disclosure.

[0066] Meanwhile, the reference suspension 18ST in the tank 100B is flowed through the flow cell 11, and the sensor unit 35 is connected to the flow cell 11 to measure the Raman scattered light RSL from the reference suspension 18ST flowing through the flow cell 11, thereby obtaining reference Raman spectral data 37R. The reference Raman spectral data 37R corresponds to the target Raman spectral data 37T. The measurement environment of this reference Raman spectral data 37R uses the flow cell 11, and thus, like the measurement environment shown in FIG. 1, is an example of a "target measurement environment" according to the technology of the present disclosure. That is, the reference measurement environment and the target measurement environment in this example differ in whether or not the flow cell 11 is used. Note that the reference measurement environment and the target measurement environment may use different types of flow cell 11. That is, a flow cell 11 with preset specifications may also be used in the reference measurement environment.

[0067] The standard Raman spectrum data 37ST and the reference Raman spectrum data 37R thus obtained are subjected to a correction coefficient derivation process 101.

[0068] As shown in graph 110 of FIG. 10 as an example, the correction coefficient derivation process 101 begins by calculating the ratio (peak ratio) between the intensity value of the reference Raman spectrum data 37ST corresponding to the component in the reference suspension 18ST and the intensity value of the reference Raman spectrum data 37R corresponding to the component in the reference suspension 18ST for each of the reference suspensions 1 to 5. The component in the reference suspension 18ST is a monoclonal antibody for the reference suspension 1, and phenylalanine for the reference suspension 4. For example, if the intensity value of the reference Raman spectrum data 37ST corresponding to the component in the reference suspension 18ST is 100 and the intensity value of the reference Raman spectrum data 37R corresponding to the component in the reference suspension 18ST is 95, the ratio is 100 / 95, which is approximately 1.05. After calculating the intensity value ratios for each of the reference suspensions 1 to 5, the average value of these intensity value ratios is calculated. This average value is used as the correction coefficient CF.

[0069] When the flow cell 11 is used, the Raman spectrum data 37 not only contains peaks corresponding to the components in the target suspension 18T or the reference suspension 18ST, but also peaks originating from the optical system 58 of the flow cell 11, such as peaks originating from the sapphire glass used to form the ball lens 59. These peaks originating from the optical system 58 are not observed in the reference measurement environment of this example, which does not use the flow cell 11, or the peak positions and / or intensity values ​​are significantly different. Therefore, as shown in graph 111, the ratio of the intensity value of the reference Raman spectrum data 37ST corresponding to the optical system 58 to the intensity value of the reference Raman spectrum data 37R corresponding to the optical system 58 deviates from the ratio of the intensity value of the reference Raman spectrum data 37ST corresponding to the components in the reference suspension 18ST to the intensity value of the reference Raman spectrum data 37R corresponding to the components in the reference suspension 18ST. Therefore, in this example, the correction coefficient CF is derived excluding the intensity value ratio corresponding to the optical system 58, as indicated by the cross.

[0070] When the instruction receiving unit 87 receives an instruction from the user U to predict the concentration of the antibody 16 using the target Raman spectral data 37T, the display control unit 90 controls the display 41 to display a used flow cell selection screen 115, as shown in FIG. 11 , for example. The used flow cell selection screen 115 includes a pull-down menu 116 for alternatively selecting the flow cell 11 used when the target Raman spectral data 37T was obtained. The pull-down menu 116 lists selectable characters that can uniquely identify the flow cell 11, such as the product name and model number of the flow cell 11 (here, the product name is used as an example). The user U selects the used flow cell 11 from the pull-down menu 116 and presses the OK button 117.

[0071] When the user U selects the flow cell 11 used from the pull-down menu 116 on the used flow cell selection screen 115 and presses the OK button 117, the instruction receiving unit 87 receives specification information 120 for the flow cell 11 used, as shown in FIG. 12 as an example. The specification information 120 includes the product name, model number, etc. of the flow cell 11 (here, the product name is used as an example). The instruction receiving unit 87 outputs the specification information 120 to the RW control unit 86.

[0072] The RW control unit 86 obtains the correction coefficient CF corresponding to the flow cell 11 specified in the specification information 120 by reading it from the correction coefficient information 81 in the storage 70. The RW control unit 86 outputs the read correction coefficient CF to the correction unit 88. The correction unit 88 corrects the target Raman spectrum data 37T using the correction coefficient CF from the RW control unit 86. Note that FIG. 12 illustrates an example in which the flow cell 11 with the product name "FC1" is selected as the flow cell 11 used, and "1.05" is read from the correction coefficient information 81 as the correction coefficient CF and output to the correction unit 88.

[0073] As an example, as shown in FIG. 13, the correction unit 88 converts the target Raman spectrum data 37T into corrected target Raman spectrum data 37TC by uniformly multiplying all intensity values ​​of the target Raman spectrum data 37T by the correction coefficient CF.

[0074] 14 , the prediction unit 89 inputs the correction target Raman spectrum data 37TC into the antibody concentration prediction model 82, and causes the antibody concentration prediction model 82 to output a prediction result 95. The prediction result 95 includes a predicted value of the concentration of the antibody 16 in the target suspension 18T.

[0075] As an example, as shown in FIG. 15 , the antibody concentration prediction model 82 is constructed using a neural network 125. Therefore, the antibody concentration prediction model 82 is also an example of a "machine learning model" according to the technology of the present disclosure. As is well known, the neural network 125 has an input layer 126, an intermediate layer (also called a hidden layer) 127, and an output layer 128. The input layer 126, the intermediate layer 127, and the output layer 128 each have a plurality of nodes ND. Coefficients indicating the strength of connections between the nodes ND in the input layer 126 and the nodes ND in the intermediate layer 127, between the nodes ND in the intermediate layer 127, and between the nodes ND in the intermediate layer 127 and the nodes ND in the output layer 128 are set. An appropriate activation function, such as a linear function or a ReLu (Rectified Linear Unit) function, is set for the nodes ND in the output layer 128.

[0076] The intensity value of each wavenumber of the Raman spectrum data 37TC to be corrected is input to each node ND of the input layer 126. A predicted value of the concentration of the antibody 16 is output from each node ND of the output layer 128. Note that the antibody concentration prediction model 82 is not limited to the exemplified neural network 125, and may be other machine learning models such as a decision tree, a gradient boosting decision tree, a random forest, a support vector machine, or a naive Bayes model.

[0077] As an example, as shown in FIG. 16 , teacher data (also referred to as learning data or training data) 130 is used in the learning phase of the antibody concentration prediction model 82. The teacher data 130 is a set of learning reference Raman spectral data 37STL and a correct antibody concentration 95CA. The learning reference Raman spectral data 37STL was obtained by measuring Raman scattered light RSL emitted from a reference suspension 18ST containing at least antibody 16 in the reference measurement environment shown in FIG. 8. The correct antibody concentration 95CA is the concentration of antibody 16 in the reference suspension 18ST, which was the source of the learning reference Raman spectral data 37STL, introduced into a high performance liquid chromatography (hereinafter referred to as HPLC) apparatus and measured using the mass spectrometry function of the HPLC apparatus.

[0078] In the learning phase, the learning reference Raman spectrum data 37STL is input to the antibody concentration prediction model 82, and the learning prediction result 95L is output from the antibody concentration prediction model 82. Next, a loss calculation is performed on the antibody concentration prediction model 82 using a loss function based on the results of comparing the learning prediction result 95L with the correct antibody concentration 95CA. Then, the coefficients of the antibody concentration prediction model 82 (coefficients of node ND) are updated according to the results of the loss calculation, and the antibody concentration prediction model 82 is updated according to the updated settings.

[0079] In the learning phase, the above-described series of processes, including inputting the learning reference Raman spectral data 37STL into the antibody concentration prediction model 82, outputting the learning prediction result 95L from the antibody concentration prediction model 82, calculating the loss, updating the settings, and updating the antibody concentration prediction model 82, are repeatedly performed while changing the teacher data 130. The repetition of the above-described series of processes is terminated when the prediction accuracy of the learning prediction result 95L for the correct antibody concentration 95CA reaches a predetermined set level. The antibody concentration prediction model 82 whose prediction accuracy has thus reached the set level is stored in the storage 70 and used by the prediction unit 89. The repetition of the above-described series of processes is an example of "calibration" according to the technology of the present disclosure. Note that learning may be terminated when the above-described series of processes has been repeated a set number of times, regardless of the prediction accuracy of the learning prediction result 95L for the correct antibody concentration 95CA.

[0080] Upon receiving the prediction result 95 from the prediction unit 89, the display control unit 90 controls the display 41 to display a prediction result display screen 135, as shown in FIG. 17 , for example. The prediction result display screen 135 displays the predicted value of the antibody 16 concentration in the prediction result 95. A save button 136 and an OK button 137 are provided at the bottom of the prediction result display screen 135. When the save button 136 is selected, the predicted value is stored in the storage 70 in association with the target Raman spectrum data 37T. When the OK button 137 is selected, the display of the prediction result display screen 135 is cleared. For example, the antibody 16 concentration may be predicted at multiple time points at regular intervals using the antibody concentration prediction model 82 to obtain predicted values ​​of the antibody 16 concentration at the multiple time points. The predicted values ​​of the antibody 16 concentration at the multiple time points may then be plotted on a graph with time on the horizontal axis, and the time series changes in the predicted values ​​of the antibody 16 concentration may be displayed on the prediction result display screen 135.

[0081] Next, the operation of the above configuration will be described with reference to the flowchart shown in Fig. 18. As shown in Fig. 6, the CPU 72 of the information processing device 40 functions as an acquisition unit 85, a RW control unit 86, an instruction acceptance unit 87, a correction unit 88, a prediction unit 89, and a display control unit 90 when the operating program 80 is started.

[0082] First, the acquiring unit 85 acquires the target Raman spectrum data 37T from the Raman spectrometer 12 (step ST100). The target Raman spectrum data 37T is output from the acquiring unit 85 to the RW control unit 86 and stored in the storage 70 under the control of the RW control unit 86 (step ST110).

[0083] Under the control of the display control unit 90, the flow cell selection screen 115 shown in Fig. 11 is displayed on the display 41 (step ST120). The user U selects the flow cell 11 used to obtain the target Raman spectrum data 37T from the pull-down menu 116 and presses the OK button 117. As a result, the instruction receiving unit 87 receives designation information 120 as shown in Fig. 12 (step ST130). The designation information 120 is output from the instruction receiving unit 87 to the RW control unit 86.

[0084] Under the control of the RW control unit 86, the target Raman spectrum data 37T is read from the storage 70. Furthermore, the correction coefficient CF corresponding to the flow cell 11 designated by the designation information 120 is read from the correction coefficient information 81 in the storage 70 and acquired (step ST140). The target Raman spectrum data 37T and the correction coefficient CF are output from the RW control unit 86 to the correction unit 88.

[0085] 13, the corrector 88 corrects the target Raman spectrum data 37T using the correction coefficient CF to generate the corrected Raman spectrum data 37TC (step ST150). The corrected Raman spectrum data 37TC is output from the corrector 88 to the predictor 89.

[0086] 14 , in the prediction unit 89, the correction target Raman spectrum data 37TC is input to the antibody concentration prediction model 82, and a prediction result 95 is output from the antibody concentration prediction model 82 (step ST160). The prediction result 95 is output from the prediction unit 89 to the display control unit 90.

[0087] As shown in Figure 17, under the control of the display control unit 90, the prediction result display screen 135 is displayed on the display 41 (step ST170), and the predicted value of the antibody 16 concentration in the prediction result 95 is made available for viewing by the user U.

[0088] The user U makes various decisions based on the predicted values ​​displayed on the prediction result display screen 135. For example, consider a case where the user U is determining conditions for culturing antibody-producing cells using small-scale equipment. In this case, if the predicted values ​​are worse than the target values, the user U decides to stop the current experiment and move on to an experiment using new conditions. Also, consider a case where the determination of conditions has been completed and mass production is being carried out using large-scale equipment. In this case, if the predicted values ​​are worse than the target values, the user U decides to stop mass production and perform maintenance on the culture tanks in the culture unit 13 or the various chromatography devices in the purification unit 14.

[0089] As described above, the information processing device 40 includes an acquisition unit 85, a RW control unit 86, a correction unit 88, and a prediction unit 89. The prediction unit 89 uses the antibody concentration prediction model 82 calibrated using the reference Raman spectral data 37ST obtained from the reference suspension 18ST in the reference measurement environment (trained using the training reference Raman spectral data 37ST). The acquisition unit 85 acquires target Raman spectral data 37T obtained from a target suspension 18T with an unknown antibody 16 concentration in a target measurement environment different from the reference measurement environment. The RW control unit 86 acquires a correction coefficient CF derived by comparing the reference Raman spectral data 37ST with reference Raman spectral data 37R corresponding to the target Raman spectral data 37T, which is obtained from the reference suspension 18ST in the target measurement environment. The correction unit 88 corrects the target Raman spectral data 37T using the correction coefficient CF to generate corrected target Raman spectral data 37TC. The prediction unit 89 applies the correction target Raman spectrum data 37TC to the antibody concentration prediction model 82, causing the antibody concentration prediction model 82 to predict the concentration of the antibody 16.

[0090] Differences in the Raman spectrum data 37 resulting from the measurement environment can be eliminated by correcting using the correction coefficient CF. Therefore, even if the measurement environment for the Raman scattered light RSL changes, it is possible to obtain a highly reliable prediction result 95 of the concentration of the antibody 16 in the target suspension 18T.

[0091] 10, the correction coefficient CF is derived based on the ratio of the intensity values ​​of the reference Raman spectral data 37ST and the reference Raman spectral data 37R corresponding to the components in the reference suspension 18ST. Therefore, a valid correction coefficient CF can be easily derived. Alternatively, a machine learning model may be used that outputs the correction coefficient CF in response to input of the reference Raman spectral data 37ST and the reference Raman spectral data 37R.

[0092] As shown in FIG. 10, the correction coefficient CF is derived by excluding the ratio of the intensity values ​​corresponding to the measuring instrument for the Raman scattered light RSL, in this case, the optical system 58 of the flow cell 11.

[0093] As described above, when the flow cell 11 is used, not only peaks corresponding to the components in the target suspension 18T but also peaks originating from the optical system 58 of the flow cell 11 are observed in the Raman spectrum data 37. The peaks originating from the optical system 58 appear differently from the peaks corresponding to the components in the target suspension 18T. Therefore, as shown in graph 111 of FIG. 10 , the ratio of the intensity value of the reference Raman spectrum data 37ST corresponding to the optical system 58 to the intensity value of the reference Raman spectrum data 37R corresponding to the optical system 58 is different from the ratio of the intensity value of the reference Raman spectrum data 37ST corresponding to the components in the reference suspension 18ST to the intensity value of the reference Raman spectrum data 37R corresponding to the components in the reference suspension 18ST. Therefore, if the correction coefficient CF is derived without taking into account the ratio of the intensity values ​​corresponding to the optical system 58 of the flow cell 11, a more valid correction coefficient CF can be derived.

[0094] 8, the reference measurement environment and the target measurement environment differ in whether or not a measurement device for Raman scattered light RSL, in this case the flow cell 11, is used. Therefore, a more valid correction coefficient CF can be derived.

[0095] As shown in Fig. 7, there are multiple target measurement environments, and the multiple target measurement environments use different types of flow cells 11. A correction coefficient CF is pre-stored in the storage 70 for each of the multiple target measurement environments. As shown in Figs. 11 and 12, the instruction receiving unit 87 receives a designation of the type of flow cell 11 used. The RW control unit 86 obtains the correction coefficient CF corresponding to the designation by reading it from the storage 70.

[0096] This allows for easy support for multiple target measurement environments. It also eliminates the need to prepare antibody concentration prediction models 82 for multiple target measurement environments, and therefore for multiple types of flow cells 11. All that is required as training data 130 for the antibody concentration prediction model 82 is at least the training reference Raman spectral data 37STL and the correct antibody concentration 95CA obtained in the reference measurement environment. This allows for easy preparation of training data 130.

[0097] The flow cell 11 differs in at least one of the distance D between the second surface 62 of the transparent plate 60, which is the incident surface of the Raman scattered light RSL of the optical system 58, and the wall surface 63A of the flow path 52 facing the second surface 62, the reflectance R of the wall surface 63A, and the focal length FL of the optical system 58. The differences in the distance D, reflectance R, and focal length FL allow the flow cell 11 to easily accommodate changes in the target measurement environment.

[0098] Concentration is the most popular index for knowing the physicochemical characteristics of antibody 16. Therefore, by predicting the concentration as the state of antibody 16 as in this example, user U can easily understand the physicochemical characteristics of antibody 16.

[0099] Biopharmaceuticals containing antibody 16 are called antibody drugs and are widely used to treat chronic diseases such as cancer, diabetes, and rheumatoid arthritis, as well as rare diseases such as hemophilia and Crohn's disease. Therefore, this example, in which the protein is antibody 16, can promote the development of antibody drugs that are widely used to treat various diseases.

[0100] Raman scattered light RSL is likely to reflect information derived from the functional groups of amino acids in proteins. Therefore, by using electromagnetic waves as Raman scattered light RSL as in this example, it is possible to obtain Raman spectrum data 37 that accurately reflects the physical properties of antibody 16, which is a protein, such as its concentration.

[0101] 15 and 16 , the antibody concentration prediction model 82 is a machine learning model trained using the training reference Raman spectral data 37STL as training data. Machine learning models are generally used to predict unknown parameters, and can improve prediction accuracy to a certain level through training. Therefore, it is possible to easily generate an antibody concentration prediction model 82 with relatively high prediction accuracy.

[0102] (Modification) In the correction coefficient information 81 of the first embodiment, an example has been shown in which one correction coefficient CF is registered for one flow cell 11, but this is not limiting. As an example, as in correction coefficient information 140 shown in Fig. 19 , a correction coefficient CF may be registered for one flow cell 11 for each wavenumber range of the Raman spectrum data 37. Furthermore, as an example, as in correction coefficient information 142 shown in Fig. 20 , a correction coefficient CF may be registered for one flow cell 11 for each wavenumber. In this way, by setting the correction coefficient CF in detail for each wavenumber range, for each wavenumber, etc., more precise correction can be applied to the target Raman spectrum data 37T.

[0103] Second Embodiment In the first embodiment, as shown in Fig. 16, the teacher data 130 includes the training reference Raman spectral data 37STL, but this is not limiting. As an example, as shown in Fig. 21, teacher data 130 including training correction target Raman spectral data 37TCL may be used in addition to the teacher data 130 including the training reference Raman spectral data 37STL. The training correction target Raman spectral data 37TCL is generated by correcting previously measured target Raman spectral data 37T using a correction coefficient CF.

[0104] In this way, in the second embodiment, the antibody concentration prediction model 82 is trained using the training correction target Raman spectral data 37TCL. This makes it possible to secure a large amount of training data 130. As a result, the prediction accuracy of the antibody concentration prediction model 82 can be improved.

[0105] First, in the reference measurement environment shown in Fig. 8 , in which the sensor unit 35 of the Raman spectrometer 12 was directly immersed in the reference suspension 18ST in the tank 100A, Raman scattered light RSL from the reference suspension 18ST was measured. Reference Raman spectral data 37ST was thus obtained. Using training data 130 including the thus obtained reference Raman spectral data 37ST as training reference Raman spectral data 37STL, an antibody concentration prediction model 82 was trained as shown in Fig. 16 , and an antibody concentration prediction model 82 with prediction accuracy reaching a set level was obtained.

[0106] 1, a flow cell 11 was provided in the outlet 26 of the concentrating section 15, and the sensor section 35 of the Raman spectrometer 12 was connected to the flow cell 11 to measure the Raman scattered light RSL of the target suspension 18T, thereby obtaining target Raman spectral data 37T. The correction coefficient CF derived in the correction coefficient derivation process 101 shown in FIG. 10 was applied to this target Raman spectral data 37T as shown in FIG. 13 to generate corrected target Raman spectral data 37TC.

[0107] 14 , the corrected Raman spectral data 37TC was input into an antibody concentration prediction model 82, and a prediction result 95 was output from the antibody concentration prediction model 82. Meanwhile, the concentration of the antibody 16 in the target suspension 18T was actually measured using the mass analysis function of the HPLC device. The prediction of the concentration of the antibody 16 using this antibody concentration prediction model 82 and the actual measurement of the concentration of the antibody 16 using the HPLC device were performed at the start of concentration in the concentrating section 15, intermediate 1, intermediate 2, and final stages, and when the buffer solution used for washing was recovered (referred to as recovery).

[0108] 22 shows a comparison graph 150 of the predicted values ​​of antibody 16 concentration by antibody concentration prediction model 82 of the Example at each time point with the actual measured values ​​of antibody 16 concentration by the HPLC device (referred to as offline analysis values ​​in the figure; the same applies to FIGS. 24 to 26). The average error of the predicted values ​​from the actual measured values ​​at each time point (hereinafter referred to as average error) was 3.7%.

[0109] Comparative Examples Comparative Examples 1, 2, and 3 were prepared as shown in Table 152 in FIG. 23 as an example. Comparative Example 1 is a case where the target Raman spectral data 37T was not corrected using the correction coefficient CF, and the target Raman spectral data 37T was directly input to the antibody concentration prediction model 82. Comparative Example 2 is a case where the derivative transformation (including 15-point data smoothing) described in JP-T-2022-552876 was performed on the target Raman spectral data 37T. Comparative Example 3 is a case where the derivative transformation (including 15-point data smoothing) and standard normalization described in JP-T-2022-552876 were performed on the target Raman spectral data 37T.

[0110] Figure 24 shows a comparison graph 160 of the predicted values ​​of antibody 16 concentration by antibody concentration prediction model 82 at each time point in Comparative Example 1 with the actual measured values ​​of antibody 16 concentration by the HPLC device. At each time point, the predicted values ​​derived are lower than the actual measured values. This is consistent with the case where the correction coefficient CF is 1 or greater, as shown in Figure 10. The average error in this case was 6.0%, which was worse than that of the Example.

[0111] 25 shows a comparison graph 161 of the predicted values ​​of antibody 16 concentration by antibody concentration prediction model 82 at each time point in Comparative Example 2 and the actual measured values ​​of antibody 16 concentration by the HPLC device. The average error in this case was 16.9%, which was significantly worse than that of the Example.

[0112] 26 shows a comparison graph 162 of the predicted values ​​of antibody 16 concentration by antibody concentration prediction model 82 at each time point in Comparative Example 3 and the actual measured values ​​of antibody 16 concentration by the HPLC device. The average error in this case was 12.9%, which, like Comparative Example 1, was significantly worse than in the Examples.

[0113] From the above, it was confirmed that the examples to which the technology of the present disclosure was applied had more useful effects than comparative examples 1 to 3.

[0114] Although the concentration of antibody 16 is predicted as a state, this is not limiting. The concentration of aggregates 17 of antibody 16 may also be predicted. Furthermore, instead of or in addition to concentration, purity, density, etc. may be predicted. Purity is calculated, for example, by using the sum of the amount of antibody 16 and the amount of impurities as the denominator and the amount of antibody 16 as the numerator. For example, two or more states, such as concentration and density, may be predicted. Furthermore, in addition to quantitative indicators such as concentration and purity, qualitative indicators such as the quality level of antibody 16 (two levels of good or bad, or five levels from 1 to 5, etc.) may also be used.

[0115] Other conditions may include viable cell density, glucose concentration, glutamic acid concentration, amino acid concentration, solvent component concentration, additive concentration, lactic acid concentration, ammonia concentration, cell-derived protein / cell-derived DNA concentration, other cellular metabolite concentrations, antibody fragment concentration, charge isomer concentration, etc.

[0116] The substance for which the Raman spectrum data 37 is measured is not limited to the antibody 16. Other substances than the antibody 16, such as proteins, peptides, nucleic acids (DNA, RNA (Ribonucleic Acid)), lipids, viruses, virus subunits, and virus-like particles, may also be used.

[0117] The component is not limited to antibody 16. It may also be a cytokine (interferon, interleukin, etc.), a hormone (insulin, glucagon, follicle-stimulating hormone, erythropoietin, etc.), a growth factor (IGF (insulin-like growth factor)-1, bFGF (basic fibroblast growth factor), etc.), a blood coagulation factor (factor 7, factor 8, factor 9, etc.), an enzyme (lysosomal enzyme, DNA (deoxyribonucleic acid) degrading enzyme, etc.), an Fc (fragment crystalline) fusion protein, a receptor, albumin, or a protein vaccine. The antibody 16 also includes bispecific antibodies, antibody-drug conjugates, low molecular weight antibodies, sugar chain modified antibodies, and the like.

[0118] The electromagnetic wave is not limited to Raman scattered light RSL, and therefore the spectrum data is not limited to Raman spectrum data 37. It may be infrared absorption spectrum data, near-infrared absorption spectrum data, nuclear magnetic resonance spectrum data, ultraviolet-visible absorption spectroscopy (UV-Vis) spectrum data, or fluorescence spectrum data.

[0119] The antibody concentration prediction model 82 is not limited to a machine learning model. It may also be a model generated by multivariate analysis or statistical analysis. Examples of multivariate analysis and statistical analysis include linear regression, multiple regression, principal component regression, partial least squares regression, logistic regression, Lasso regression, ridge regression, support vector regression, and Gaussian process regression. In such models generated by multivariate analysis or statistical analysis, determining the coefficients of the regression equation based on at least two sets of training data 130 is an example of "calibration" according to the technology of the present disclosure.

[0120] Note that different measurement environments include, instead of or in addition to the use or non-use of the flow cell 11 or different types, cases where the age of the optical components, the shape of the flow path 52, temperature, humidity, the flow rate and flow velocity of the target suspension 18T, etc. Specifically, the optical components are optical fibers that guide the excitation light EL and the Raman scattered light RSL, or a light source for the excitation light EL, an optical system 58, etc., that are arranged in the sensor unit 35 of the Raman spectrometer 12, etc. Specifically, the shape of the flow path 52 refers to the cross-sectional shape of the flow path 52, such as a circular cross-sectional shape or an elliptical cross-sectional shape, or the overall shape of the flow path 52, such as a straight, U-shaped, or L-shaped shape.

[0121] The flow path 52 of the flow cell 11 may be U-shaped. The shape of the flow cell 11 is not limited to a cylindrical shape, but may be a rectangular tube. The cross-sectional shape of the flow path 52 is not limited to a circular shape, but may be an elliptical shape or a rectangular shape. The flow cell 11 may be made of a composite material such as carbon fiber reinforced resin.

[0122] The fluid is not limited to purified liquid 18 or a mixture 18M of purified liquid 18 and a concentrated liquid. It may be a cell culture solution before cells are removed by a cell removal filter in the culture unit 13, or a culture supernatant after cells are removed. It may be a cell culture solution (culture medium) that does not contain biological molecules before being supplied to the culture unit 13. It may be a purified liquid introduced into various chromatography devices or a purified liquid sent out from various chromatography devices. It may also be a purified liquid after virus inactivation processing.

[0123] The fluid is not limited to that related to the manufacture of biopharmaceuticals, but may be, for example, river water collected to check for water pollution, etc. Furthermore, the fluid is not limited to a liquid, but may be a gas.

[0124] In the above-described embodiments, so-called in-line sensing has been exemplified, in which a measurement sensor, in this case the sensor unit 35 of the Raman spectrometer 12, is installed within the manufacturing process to perform measurements within the manufacturing process, but the present disclosure is not limited to this. The technology of the present disclosure may also be applied to offline sensing, in which measurements are performed outside the manufacturing process using a measurement sensor that is separated from the manufacturing process.

[0125] The optical element constituting the optical system 58 is not limited to the illustrated ball lens 59. It may also be a hemispherical lens, a plano-convex lens, a biconvex lens, or a cylindrical lens. Furthermore, the optical element is not limited to the illustrated disc-shaped transparent plate 60 having a first surface 61 and a second surface 62 that are parallel to each other. It may also be a transparent plate having a first surface shaped to match the shape of the exit surface of a ball lens, a plano-convex lens, a biconvex lens, or the like, and a flat second surface. It is noted that the transparent plate need not be used, and the optical system 58 may be composed solely of the ball lens 59, a hemispherical lens, or the like.

[0126] The information processing device 40 may be a personal computer installed in the pharmaceutical facility as shown in FIG. 1, or may be a server computer installed in a data center independent of the pharmaceutical facility.

[0127] When the information processing device 40 is configured as a server computer, the target Raman spectrum data 37T is transmitted from a personal computer installed in each pharmaceutical facility to the server computer via a network such as the Internet. The server computer distributes various screens, such as the flow cell selection screen 115 and the prediction result display screen 135, to the personal computer in the form of screen data for web distribution created using a markup language such as XML (Extensible Markup Language). The personal computer reproduces the screen to be displayed on the web browser based on the screen data and displays it on the display. Note that other data description languages, such as JSON (Javascript Object Notation), may be used instead of XML.

[0128] The hardware configuration of the computer constituting the information processing device 40 according to the technology of the present disclosure can be modified in various ways. For example, the information processing device 40 can be configured with multiple computers separated as hardware in order to improve processing power and reliability. For example, the functions of the correction unit 88 and the prediction unit 89 can be distributed and performed by two computers. In this case, the information processing device 40 is configured with two computers.

[0129] In this way, the hardware configuration of the computer of the information processing device 40 can be changed as appropriate depending on the required performance such as processing power, safety, reliability, etc. Furthermore, not only the hardware, but also application programs such as the operating program 80 can be duplicated or stored in a distributed manner across multiple storage devices in order to ensure safety and reliability.

[0130] In each of the above embodiments, the following various processors can be used as the hardware structure of processing units that perform various processes, such as the acquisition unit 85, RW control unit 86, instruction reception unit 87, correction unit 88, prediction unit 89, and display control unit 90. The various processors include the CPU 72, which is a general-purpose processor that executes software (operation program 80) and functions as various processing units, as described above, as well as dedicated electrical circuits that are processors having a circuit configuration specifically designed to perform specific processes, such as a programmable logic device (PLD) that is a processor whose circuit configuration can be changed after manufacture, such as an FPGA (Field Programmable Gate Array), and an ASIC (Application Specific Integrated Circuit).

[0131] A single processing unit may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs and / or a combination of a CPU and an FPGA).Furthermore, multiple processing units may be configured with a single processor.

[0132] Examples of configuring multiple processing units with a single processor include, first, a form in which one processor is configured with a combination of one or more CPUs and software, as typified by computers such as client and server, and this processor functions as multiple processing units. Second, a form in which a processor is used to realize the functions of the entire system including multiple processing units with a single IC (Integrated Circuit) chip, as typified by systems on chips (SoCs). In this way, various processing units are configured using one or more of the above-mentioned various processors as a hardware structure.

[0133] Furthermore, more specifically, the hardware structure of these various processors can be an electric circuit (circuitry) that combines circuit elements such as semiconductor elements.

[0134] From the above description, the technology described in the following supplementary paragraphs can be understood.

[0135] [Supplementary Item 1] An information processing device that predicts a state of a target component in a suspension in which biological molecules are dispersed as components in a liquid, based on spectral data obtained by measuring electromagnetic waves emitted from the suspension, comprising a processor, wherein the processor: uses a state prediction model calibrated with reference spectral data obtained from a reference suspension in a reference measurement environment, acquires target spectral data obtained from a target suspension in which the state of the target component is unknown, in a target measurement environment different from the reference measurement environment, acquires a correction coefficient derived by comparing the reference spectral data with reference spectral data corresponding to the target spectral data, the reference spectral data being obtained from the reference suspension in the target measurement environment, corrects the target spectral data with the correction coefficient to generate corrected target spectral data, and applies the corrected target spectral data to the state prediction model, thereby causing the state prediction model to predict the state of the target component. [Supplementary Item 2] The information processing device according to Supplementary Item 1, wherein the correction coefficient is derived based on a ratio of intensity values ​​of the reference spectral data and the reference spectral data corresponding to the component in the reference suspension. [Supplementary Item 3] The information processing device according to Supplementary Item 1 or Supplementary Item 2, wherein the correction coefficient is derived by excluding a ratio of intensity values ​​corresponding to a measuring instrument for the electromagnetic waves. [Supplementary Item 4] The information processing device according to Supplementary Item 3, wherein the measuring instrument is a flow cell having a flow path through which the suspension flows and an optical system for capturing the electromagnetic waves, and the intensity value corresponding to the measuring instrument is an intensity value corresponding to the optical system. [Supplementary Item 5] The information processing device according to any one of Supplementary Item 1 to Supplementary Item 4, wherein the reference measurement environment and the target measurement environment differ in either the presence or absence of an electromagnetic wave measuring instrument or the type of instrument. [Supplementary Item 6] The information processing device according to Supplementary Item 5, wherein the measuring instrument is a flow cell having a flow path through which the suspension flows and an optical system for capturing the electromagnetic waves. [Supplementary Item 7] The information processing device according to Supplementary Item 6, wherein there are a plurality of target measurement environments, and the plurality of target measurement environments use different types of flow cells.[Supplementary Item 8] The information processing device according to Supplementary Item 7, wherein the correction coefficient is pre-stored in a storage unit for each of the plurality of target measurement environments, and the processor receives a designation of the type of the flow cell used, and obtains the correction coefficient according to the designation by reading it from the storage unit. [Supplementary Item 9] The information processing device according to Supplementary Item 7 or Supplementary Item 8, wherein the flow cells differ in at least one of the distance between an incident surface of the optical system for the electromagnetic wave and a wall surface of the flow channel facing the incident surface, the reflectivity of the wall surface, and the focal length of the optical system. [Supplementary Item 10] The information processing device according to any one of Supplementary Item 1 to Supplementary Item 9, wherein the state prediction model is calibrated using the spectral data to be corrected. [Supplementary Item 11] The information processing device according to any one of Supplementary Item 1 to Supplementary Item 10, wherein the state of the target component is the concentration of the target component. [Supplementary Item 12] The information processing device according to any one of Supplementary Item 1 to Supplementary Item 11, wherein the target component is a protein. [Supplementary Item 13] The information processing device according to Supplementary Item 12, wherein the protein is an antibody. [Supplementary Item 14] The information processing device according to any one of Supplementary Items 1 to 13, wherein the electromagnetic wave is Raman scattered light. [Supplementary Item 15] The information processing device according to any one of Supplementary Items 1 to 14, wherein the state prediction model is a machine learning model trained using the reference spectrum data as training data.

[0136] The technology of the present disclosure can be appropriately combined with the various embodiments and / or various modified examples described above. Furthermore, it is not limited to the above-described embodiments, and various configurations can be adopted without departing from the spirit of the present disclosure. Furthermore, the technology of the present disclosure extends not only to programs, but also to storage media that non-temporarily store programs, and computer program products that include programs.

[0137] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[0138] In this specification, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed by connecting them with "and / or."

[0139] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

Claims

1. An information processing device that predicts a state of a target component in a suspension in which biological molecules are dispersed as components in a liquid, based on spectral data obtained by measuring electromagnetic waves emitted from the suspension, comprising a processor, wherein the processor uses a state prediction model calibrated with reference spectral data obtained from a reference suspension in a reference measurement environment, acquires target spectral data obtained from a target suspension in which the state of the target component is unknown, in a target measurement environment different from the reference measurement environment, acquires a correction coefficient derived by comparing the reference spectral data with reference spectral data corresponding to the target spectral data, which is obtained from the reference spectral data obtained from the reference suspension in the target measurement environment, corrects the target spectral data with the correction coefficient to generate corrected target spectral data, and applies the corrected target spectral data to the state prediction model, thereby causing the state prediction model to predict the state of the target component.

2. An information processing device according to claim 1, wherein the correction coefficient is derived based on a ratio of intensity values ​​of the standard spectrum data and the reference spectrum data corresponding to the component in the standard suspension.

3. An information processing device according to claim 1, wherein the correction coefficient is derived by excluding a ratio of intensity values ​​corresponding to an instrument for measuring the electromagnetic waves.

4. The information processing device according to claim 3, wherein the measuring instrument is a flow cell having a flow path through which the suspension flows and an optical system for capturing the electromagnetic waves, and the intensity value corresponding to the measuring instrument is an intensity value corresponding to the optical system.

5. The information processing device according to claim 1, wherein the reference measurement environment and the target measurement environment differ in either the presence or absence of an electromagnetic wave measuring instrument and the type of the instrument.

6. The information processing device according to claim 5, wherein the measuring device is a flow cell having a flow path through which the suspension flows and an optical system for capturing the electromagnetic waves.

7. The information processing device according to claim 6, wherein there are a plurality of target measurement environments, and the plurality of target measurement environments use different types of flow cells.

8. An information processing device as described in claim 7, wherein the correction coefficient is pre-stored in a memory unit for each of a plurality of target measurement environments, and the processor receives a designation of the type of the flow cell used, and obtains the correction coefficient corresponding to the designation by reading it from the memory unit.

9. An information processing device as described in claim 7, wherein the flow cell differs in at least one of the distance between the incident surface of the optical system for the electromagnetic wave and the wall surface of the flow path facing the incident surface, the reflectivity of the wall surface, and the focal length of the optical system.

10. The information processing device according to claim 1, wherein the state prediction model is calibrated by the correction target spectrum data.

11. The information processing device according to claim 1, wherein the state of the target component is the concentration of the target component.

12. The information processing device according to claim 1, wherein the target component is a protein.

13. The information processing device according to claim 12, wherein the protein is an antibody.

14. The information processing device according to claim 1, wherein the electromagnetic wave is Raman scattered light.

15. The information processing device according to claim 1, wherein the state prediction model is a machine learning model trained using the reference spectrum data as training data.

16. A method for operating an information processing device for predicting a state of a target component in a suspension in which biological molecules are dispersed as components in a liquid, based on spectral data obtained by measuring electromagnetic waves emitted from the suspension, the method comprising: using a state prediction model calibrated with reference spectral data obtained from a reference suspension in a reference measurement environment; acquiring target spectral data obtained from a target suspension in which the state of the target component is unknown, in a target measurement environment different from the reference measurement environment; acquiring a correction coefficient derived by comparing the reference spectral data with reference spectral data corresponding to the target spectral data, which is obtained from the reference suspension in the target measurement environment; correcting the target spectral data with the correction coefficient to generate corrected target spectral data; and applying the corrected target spectral data to the state prediction model, thereby causing the state prediction model to predict the state of the target component.

17. An operating program for an information processing device that predicts a state of a target component in a suspension in which biological molecules are dispersed as components in a liquid, based on spectral data obtained by measuring electromagnetic waves emitted from the suspension, the operating program for an information processing device that causes a computer to execute processes including: using a state prediction model calibrated with reference spectral data obtained from a reference suspension in a reference measurement environment; acquiring target spectral data obtained from a target suspension in which the state of the target component is unknown, in a target measurement environment different from the reference measurement environment; acquiring a correction coefficient derived by comparing the reference spectral data with reference spectral data corresponding to the target spectral data, which is obtained from the reference suspension in the target measurement environment; correcting the target spectral data with the correction coefficient to generate corrected target spectral data; and applying the corrected target spectral data to the state prediction model, thereby causing the state prediction model to predict the state of the target component.

Images