WO2023053585A1 - 学習用データの取得方法、学習用データ取得システム、ソフトセンサの構築方法、ソフトセンサ、学習用データ - Google Patents
学習用データの取得方法、学習用データ取得システム、ソフトセンサの構築方法、ソフトセンサ、学習用データ Download PDFInfo
- Publication number
- WO2023053585A1 WO2023053585A1 PCT/JP2022/023317 JP2022023317W WO2023053585A1 WO 2023053585 A1 WO2023053585 A1 WO 2023053585A1 JP 2022023317 W JP2022023317 W JP 2022023317W WO 2023053585 A1 WO2023053585 A1 WO 2023053585A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- time
- data
- series data
- liquid
- sensor
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims description 98
- 238000012549 training Methods 0.000 title abstract description 6
- 238000010276 construction Methods 0.000 title description 6
- 239000007788 liquid Substances 0.000 claims abstract description 200
- 230000003595 spectral effect Effects 0.000 claims abstract description 65
- 239000003085 diluting agent Substances 0.000 claims abstract description 57
- 238000002156 mixing Methods 0.000 claims abstract description 26
- 230000007704 transition Effects 0.000 claims abstract description 21
- 239000000523 sample Substances 0.000 claims description 86
- 239000000356 contaminant Substances 0.000 claims description 39
- 239000000203 mixture Substances 0.000 claims description 37
- 102000004169 proteins and genes Human genes 0.000 claims description 34
- 108090000623 proteins and genes Proteins 0.000 claims description 34
- 230000008569 process Effects 0.000 claims description 31
- 238000002835 absorbance Methods 0.000 claims description 30
- 238000012545 processing Methods 0.000 claims description 30
- 238000001228 spectrum Methods 0.000 claims description 30
- 238000004587 chromatography analysis Methods 0.000 claims description 20
- 238000000926 separation method Methods 0.000 claims description 14
- 238000005259 measurement Methods 0.000 claims description 11
- 238000004113 cell culture Methods 0.000 claims description 8
- 238000010801 machine learning Methods 0.000 claims description 8
- 238000000149 argon plasma sintering Methods 0.000 claims description 7
- GPRLSGONYQIRFK-UHFFFAOYSA-N hydron Chemical compound [H+] GPRLSGONYQIRFK-UHFFFAOYSA-N 0.000 claims description 7
- 238000001237 Raman spectrum Methods 0.000 claims description 5
- 239000001963 growth medium Substances 0.000 claims description 5
- 239000011259 mixed solution Substances 0.000 claims description 5
- 239000012488 sample solution Substances 0.000 claims description 5
- 102000008394 Immunoglobulin Fragments Human genes 0.000 claims description 4
- 108010021625 Immunoglobulin Fragments Proteins 0.000 claims description 4
- 238000000862 absorption spectrum Methods 0.000 claims description 4
- 239000002609 medium Substances 0.000 claims description 4
- 238000002371 ultraviolet--visible spectrum Methods 0.000 claims description 4
- 238000002189 fluorescence spectrum Methods 0.000 claims description 3
- 238000001069 Raman spectroscopy Methods 0.000 description 26
- 210000004027 cell Anatomy 0.000 description 20
- 238000005516 engineering process Methods 0.000 description 20
- 238000010828 elution Methods 0.000 description 19
- 238000004458 analytical method Methods 0.000 description 18
- 230000010365 information processing Effects 0.000 description 18
- 239000000047 product Substances 0.000 description 16
- 230000000875 corresponding effect Effects 0.000 description 12
- 230000006870 function Effects 0.000 description 12
- 239000000243 solution Substances 0.000 description 12
- 238000005070 sampling Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 10
- QTBSBXVTEAMEQO-UHFFFAOYSA-N Acetic acid Chemical compound CC(O)=O QTBSBXVTEAMEQO-UHFFFAOYSA-N 0.000 description 9
- KRKNYBCHXYNGOX-UHFFFAOYSA-N citric acid Chemical compound OC(=O)CC(O)(C(O)=O)CC(O)=O KRKNYBCHXYNGOX-UHFFFAOYSA-N 0.000 description 9
- 238000011091 antibody purification Methods 0.000 description 8
- 230000000052 comparative effect Effects 0.000 description 8
- 238000004128 high performance liquid chromatography Methods 0.000 description 8
- 239000000126 substance Substances 0.000 description 8
- 229960000074 biopharmaceutical Drugs 0.000 description 7
- 239000000872 buffer Substances 0.000 description 7
- NBIIXXVUZAFLBC-UHFFFAOYSA-N Phosphoric acid Chemical compound OP(O)(O)=O NBIIXXVUZAFLBC-UHFFFAOYSA-N 0.000 description 6
- 230000008859 change Effects 0.000 description 6
- 239000012228 culture supernatant Substances 0.000 description 6
- 238000004519 manufacturing process Methods 0.000 description 6
- 238000000746 purification Methods 0.000 description 6
- 239000012149 elution buffer Substances 0.000 description 5
- 230000005284 excitation Effects 0.000 description 5
- 239000011534 wash buffer Substances 0.000 description 5
- 230000003068 static effect Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 239000007983 Tris buffer Substances 0.000 description 3
- 229910000147 aluminium phosphate Inorganic materials 0.000 description 3
- 239000012535 impurity Substances 0.000 description 3
- 238000011112 process operation Methods 0.000 description 3
- QGZKDVFQNNGYKY-UHFFFAOYSA-N Ammonia Chemical compound N QGZKDVFQNNGYKY-UHFFFAOYSA-N 0.000 description 2
- 239000012515 MabSelect SuRe Substances 0.000 description 2
- FAPWRFPIFSIZLT-UHFFFAOYSA-M Sodium chloride Chemical compound [Na+].[Cl-] FAPWRFPIFSIZLT-UHFFFAOYSA-M 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 2
- 230000002378 acidificating effect Effects 0.000 description 2
- 210000004978 chinese hamster ovary cell Anatomy 0.000 description 2
- 210000004748 cultured cell Anatomy 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 229940079593 drug Drugs 0.000 description 2
- 239000003814 drug Substances 0.000 description 2
- JVTAAEKCZFNVCJ-UHFFFAOYSA-N lactic acid Chemical compound CC(O)C(O)=O JVTAAEKCZFNVCJ-UHFFFAOYSA-N 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 238000007670 refining Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000005406 washing Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 235000005956 Cosmos caudatus Nutrition 0.000 description 1
- 241000699802 Cricetulus griseus Species 0.000 description 1
- WQZGKKKJIJFFOK-GASJEMHNSA-N Glucose Natural products OC[C@H]1OC(O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-GASJEMHNSA-N 0.000 description 1
- 238000012855 HCP-ELISA Methods 0.000 description 1
- 108060003951 Immunoglobulin Proteins 0.000 description 1
- 150000001413 amino acids Chemical class 0.000 description 1
- 229910021529 ammonia Inorganic materials 0.000 description 1
- 238000005571 anion exchange chromatography Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000011324 bead Substances 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- WQZGKKKJIJFFOK-VFUOTHLCSA-N beta-D-glucose Chemical compound OC[C@H]1O[C@@H](O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-VFUOTHLCSA-N 0.000 description 1
- 230000000975 bioactive effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000005277 cation exchange chromatography Methods 0.000 description 1
- 239000006143 cell culture medium Substances 0.000 description 1
- 238000013375 chromatographic separation Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000001641 gel filtration chromatography Methods 0.000 description 1
- 239000008103 glucose Substances 0.000 description 1
- 102000018358 immunoglobulin Human genes 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000001678 irradiating effect Effects 0.000 description 1
- 239000004310 lactic acid Substances 0.000 description 1
- 235000014655 lactic acid Nutrition 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 210000001672 ovary Anatomy 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000001742 protein purification Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 239000011347 resin Substances 0.000 description 1
- 229920005989 resin Polymers 0.000 description 1
- 238000007086 side reaction Methods 0.000 description 1
- 239000011780 sodium chloride Substances 0.000 description 1
- 239000001509 sodium citrate Substances 0.000 description 1
- NLJMYIDDQXHKNR-UHFFFAOYSA-K sodium citrate Chemical compound O.O.[Na+].[Na+].[Na+].[O-]C(=O)CC(O)(CC([O-])=O)C([O-])=O NLJMYIDDQXHKNR-UHFFFAOYSA-K 0.000 description 1
- 239000001488 sodium phosphate Substances 0.000 description 1
- 229910000162 sodium phosphate Inorganic materials 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000001179 sorption measurement Methods 0.000 description 1
- 238000004611 spectroscopical analysis Methods 0.000 description 1
- 239000011550 stock solution Substances 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- RYFMWSXOAZQYPI-UHFFFAOYSA-K trisodium phosphate Chemical compound [Na+].[Na+].[Na+].[O-]P([O-])([O-])=O RYFMWSXOAZQYPI-UHFFFAOYSA-K 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
- G01N21/65—Raman scattering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/85—Investigating moving fluids or granular solids
- G01N21/8507—Probe photometers, i.e. with optical measuring part dipped into fluid sample
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
- G16B40/10—Signal processing, e.g. from mass spectrometry [MS] or from PCR
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
- G01N21/64—Fluorescence; Phosphorescence
- G01N2021/6417—Spectrofluorimetric devices
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
- G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
- G01N21/3577—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light for analysing liquids, e.g. polluted water
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2201/00—Features of devices classified in G01N21/00
- G01N2201/12—Circuits of general importance; Signal processing
- G01N2201/129—Using chemometrical methods
- G01N2201/1296—Using chemometrical methods using neural networks
Definitions
- the disclosed technology relates to a learning data acquisition method, a learning data acquisition system, a soft sensor construction method, a soft sensor, and learning data.
- Japanese Patent Application Laid-Open No. 2020-101543 discloses that a plurality of waveforms obtained by adding together random noise and waveforms indicated by spectral information of a test substance are generated as spectral information of a virtual sample containing the test substance and contaminants. It is described that the peak height specified from the spectral information of the test substance, which is used as (learning spectral information) and is the basis of the generated spectral information, is used as correct data.
- proteins such as antibodies, which are bioactive substances produced from cultured cells, are purified and formulated.
- the protein purification process for example, by continuously performing purification treatment by different chromatographic techniques such as cation chromatography, anion chromatography, immunoaffinity chromatography and gel filtration chromatography, stepwise Efforts are being made to increase the purity of the target protein. It is preferable to monitor the refining state in order to verify whether the refining process is appropriately performed in each step. For example, by quantifying the target protein finally obtained by the purification process, it becomes possible to control the switching timing of the valve for recovering it. It is also important to know the concentration of contaminants separated from the protein of interest at each step.
- a soft sensor is a sensor that estimates a parameter that is difficult to measure based on input data that is relatively easy to measure.
- a soft sensor that can be used in the manufacturing process of biopharmaceuticals for example, the one that derives the concentration of a specific component contained in a liquid based on the spectrum data that indicates the intensity of each wavenumber or wavelength of the electromagnetic waves affected by the liquid. is assumed.
- the above soft sensor is constructed by machine learning using learning data in which spectral data and concentrations of specific components corresponding thereto are associated one-to-one.
- the disclosed technology has been made in view of the above points, and aims to efficiently acquire learning data used for machine learning of soft sensors.
- a learning data acquisition method is a software sensor that derives the concentration of a specific component contained in a liquid based on spectrum data indicating the intensity of each wavenumber or wavelength of an electromagnetic wave affected by the liquid.
- a method of acquiring learning data used for machine learning in which a sample solution with a known concentration of a specific component is prepared, and the sample solution and diluent are mixed while continuously changing their flow rate ratio. , while the sample liquid and the diluent are being mixed, the first time-series data indicating the transition of the mixing ratio and the second time-series data indicating the transition of the spectrum data are acquired for the mixed liquid obtained by mixing.
- the spectral data may be based on Raman spectrum, infrared absorption spectrum, fluorescence spectrum or UV-Vis absorption spectrum.
- the first time-series data may be obtained based on the optically detected absorbance, conductivity, hydrogen ion concentration, refractive index, or light scattering measured for the mixed liquid.
- the sample liquid may be a treated liquid that has undergone a separation process to separate specific components. Separation may be by chromatography.
- the specific component may be protein.
- the specific component may be a contaminant other than the antibody contained in the culture medium obtained by cell culture.
- the contaminants may include at least one of antibody aggregates, antibody fragments, charge isomers, immature sugar chains, cell-derived proteins, and cell-derived DNA.
- the diluent may contain specific components contained in the sample liquid.
- the diluent may contain only components other than the specific components contained in the sample liquid.
- the first time-series data may be acquired by a first sensor provided on the channel through which the mixed liquid flows, and the second time-series data may be acquired by a second sensor provided on the channel.
- Fourth time-series data indicating changes in at least one measurement value measured for the mixed liquid while the sample liquid and the diluent are being mixed; Learning data in which measured values, spectral data, and concentrations of specific components are associated with each other may be obtained from the time-series data of (1) and the time-series data of the fourth time-series data.
- a learning data acquisition system is a learning data acquisition system for implementing the acquisition method described above, and includes a first channel through which a sample liquid flows and a second channel through which a diluent flows. a flow path, a third flow path through which the mixed liquid flows, a first pump for feeding the sample liquid, a second pump for feeding the diluent, the first pump and the second pump.
- a control unit that controls the pump, a first sensor that is provided on the third flow path and acquires first time-series data, and a first sensor that is provided on the third flow path and acquires second time-series data a second sensor; and a recording processing unit that performs processing for recording outputs of the first sensor and the second sensor on a recording medium.
- the method of constructing a soft sensor according to the disclosed technique is to learn a soft sensor model using the learning data acquired by the acquisition method described above.
- a soft sensor according to the disclosed technique is a soft sensor learned using learning data acquired by the acquisition method described above.
- the learning data according to the disclosed technology is the learning data acquired by the acquisition method described above.
- FIG. 4 is a diagram illustrating an example of the functionality of a soft sensor according to an embodiment of the disclosed technology; It is a figure which shows an example of the method of acquiring the spectral data by Raman scattered light. It is a figure which shows an example of the data for learning concerning embodiment of the disclosed technique.
- 1 is a diagram showing an example of a configuration of a learning data acquisition system according to an embodiment of technology disclosed herein;
- FIG. 4 is a diagram showing an example of a method of sending a sample liquid and a diluent according to an embodiment of technology disclosed herein;
- 1 is a diagram illustrating an example of a hardware configuration of an information processing device according to an embodiment of technology disclosed herein;
- FIG. 4 is a flow chart showing an example of the flow of processing performed by executing a learning data generation program according to an embodiment of technology disclosed herein;
- FIG. 4 is a diagram illustrating an example of the structure of an estimation model according to an embodiment of technology disclosed; 4 is a flow chart showing an example of the flow of processing performed by executing a soft sensor construction program according to an embodiment of technology disclosed herein; 4 is a flowchart showing an example of the flow of processing performed by executing an estimation program according to an embodiment of technology disclosed herein;
- FIG. 10 is a graph comparing the antibody concentration indicated by the third time series data and the antibody concentration actually measured by off-line analysis of the sampled mixture.
- the disclosed technology relates to a method of acquiring learning data for soft sensors.
- the soft sensor 10 detects a specific component contained in the liquid based on spectral data indicating the intensity of the electromagnetic waves affected by the liquid for each wavenumber or wavelength. It has a function of deriving concentration data indicating an estimated value of concentration.
- the concentration data output from the soft sensor 10 has correlation with the spectral data. That is, spectrum data is input to the software sensor 10 as an explanatory variable, and concentration data is output from the software sensor 10 as an objective variable. Concentration data is not easily monitored in-line by actual measurement. By using the soft sensor 10, concentration data can be acquired in-line based on spectrum data, which is relatively easy to monitor in-line by actual measurement.
- the software sensor 10 applies an analysis method based on Raman spectroscopy. That is, in the present embodiment, spectral data by Raman scattered light is applied as the spectral data input to the soft sensor 10 .
- Raman spectroscopy is a spectroscopic method that evaluates substances using Raman scattered light. When a material is irradiated with light, the light interacts with the material to generate Raman scattered light with a wavelength different from that of the incident light. Since the wavelength difference between the incident light and the Raman scattered light corresponds to the molecular vibrational energy of the substance, Raman scattered light with different wavelengths (wavenumbers) can be obtained between substances with different molecular structures.
- Raman scattered light can be used to estimate various physical properties such as stress, temperature, electrical properties, orientation, and crystallinity. Of the Stokes line and the anti-Stokes line, the Raman scattered light preferably uses the Stokes line.
- the soft sensor 10 can be used, for example, to estimate the concentration of a specific component contained in the treatment liquid obtained in the manufacturing process of biopharmaceuticals.
- An example of a manufacturing process in which the soft sensor 10 is used is an antibody purification process.
- the “specific component” for which concentration data is to be derived by the soft sensor 10 may be protein, for example.
- the protein may be, for example, an immunoglobulin produced from cultured cells, ie an antibody.
- FIG. 2 is a diagram showing an example of a method of acquiring spectral data of the liquid 30 using Raman scattered light.
- Spectral data can be obtained using a known Raman spectrophotometer 20 .
- the Raman spectrophotometer includes a probe 21 and an analyzer 22.
- FIG. As shown in FIG. 2, the tip of probe 21 is immersed in liquid 30 contained in container 31 .
- the liquid 30 is irradiated with excitation light emitted from a light emitting portion (not shown) provided at the tip of the probe 21 .
- Raman scattered light generated by interaction between the excitation light and the liquid 30 is received by a light receiving section (not shown) provided at the tip of the probe 21 .
- the acquired Raman scattered light is decomposed for each wavenumber (reciprocal of wavelength) by the analyzer 22 to generate spectral data, which is a spectral intensity value for each wavenumber.
- spectral data may be spectral intensity values for each wavelength.
- Spectral data is input to the soft sensor 10 .
- the soft sensor 10 outputs concentration data indicating an estimated concentration of a specific component contained in the liquid 30 based on the spectrum data. It is also possible to obtain spectral data in-line by irradiating the liquid 30 with excitation light through a flow cell (not shown) provided on a channel through which the liquid 30 flows.
- the soft sensor 10 is constructed by machine learning using multiple combinations of spectral data and concentration data as learning data.
- FIG. 3 is a diagram showing an example of learning data.
- FIG. 3 exemplifies the case where the concentration data is antibody concentration.
- the soft sensor 10 receives spectral data from Raman scattered light and derives an estimated value of the antibody concentration contained in the liquid based on the spectral data.
- FIG. 4 is a diagram showing an example of the configuration of the learning data acquisition system 40 according to the embodiment of the technology disclosed.
- the learning data acquisition system 40 includes a first container 41, a second container 42, a first channel 43A, a second channel 43B, a third channel 43C, a first pump 44A, a second It includes a pump 44B, a control section 48, a first sensor 47A, a second sensor 47B, a collection container 49, and an information processing device 100.
- FIG. 4 is a diagram showing an example of the configuration of the learning data acquisition system 40 according to the embodiment of the technology disclosed.
- the learning data acquisition system 40 includes a first container 41, a second container 42, a first channel 43A, a second channel 43B, a third channel 43C, a first pump 44A, a second It includes a pump 44B, a control section 48, a first sensor 47A, a second sensor 47B, a collection container 49, and an information processing device 100.
- a sample liquid 50 is contained in the first container 41 .
- the sample liquid 50 is a liquid containing a "specific component” (here, an antibody) for which concentration data in learning data is to be obtained.
- a specific component here, an antibody
- concentration of a specific component (antibody) in sample liquid 50 is known.
- the antibody concentration in the sample liquid 50 may be measured, for example, by off-line analysis using HPLC (High Performance Liquid Chromatography).
- the sample liquid 50 preferably contains the same components as those contained in the liquid from which concentration data is derived by the soft sensor 10, in addition to the specific components (antibodies).
- the sample liquid 50 is subjected to a separation process to concentrate antibodies from the culture medium obtained by cell culture. It is preferable to use a processing liquid or a liquid simulating it.
- the sample liquid 50 preferably contains, in addition to the antibody, buffer components containing one or more of phosphoric acid, acetic acid, tris, and citric acid.
- the separation processing performed on the sample liquid 50 is preferably by chromatography.
- the chromatographic treatment described above is preferably an immunoaffinity chromatographic treatment.
- a diluent 51 is contained in the second container 42 .
- a diluent 51 is used to change the concentration of the antibody contained in the sample liquid 50 .
- the diluent 51 preferably contains components other than the “specific component” (ie, antibody) contained in the sample liquid 50 .
- the diluent 51 may contain no antibody and may contain the buffer components described above.
- the diluent 51 may contain a “specific component” (that is, an antibody) contained in the sample liquid 50 .
- the concentration of the antibody contained in the diluent 51 should be known.
- the diluent 51 may contain specific impurities assumed in the actual process.
- the diluent 51 contains an antibody or an impurity
- the first channel 43A is connected to the first container 41, and the sample liquid 50 flows through the first channel 43A.
- the second channel 43B is connected to the second container 42, and the diluent 51 flows through the second channel 43B.
- the first pump 44A is provided on the first flow path 43A and feeds the sample liquid 50 .
- the second pump 44B is provided on the second flow path 43B and feeds the diluent 51 .
- the control unit 48 controls the flow rate of the sample liquid 50 flowing through the first channel 43A and the flow rate of the diluent 51 flowing through the second channel 43B by controlling the first pump 44A and the second pump 44B. do.
- the controller 48 controls the first pump 44A and the second pump 44B so as to continuously change the flow rate ratio between the sample liquid 50 and the diluent 51 . More specifically, as shown in FIG.
- the control unit 48 keeps the total value of the flow rate A of the sample liquid 50 and the flow rate B of the diluent 51 constant (for example, 1 mL/min), and
- the liquid feed of the sample liquid 50 and the diluent 51 is controlled so that the flow rate ratio (A:B) changes linearly from 0:1 to 1:0 in (for example, 10 minutes).
- Such liquid transfer is called linear gradient liquid transfer.
- the liquid feeding of the sample liquid 50 and the diluted liquid 51 may be controlled so that the flow rate ratio (A:B) changes linearly from 1:0 to 0:1.
- the flow rate ratio of the sample liquid 50 and the diluent 52 may be changed stepwise.
- the range in which the flow path ratio (A:B) is changed is not limited to the entire range from 0:1 to 1:0, but a partial range (for example, 0.2:0.8 to 0.8: range up to 0.2). In this case, it is preferable to cover the concentration range handled in the actual process. However, from the viewpoint of comprehensively acquiring learning data, the range in which the flow rate ratio (A:B) is changed is preferably the entire range from 0:1 to 1:0. Moreover, it is preferable to perform channel replacement liquid feeding with the sample liquid 50 and the diluent 51 before starting the linear gradient liquid feeding.
- the first flow path 43A, the second flow path 43B and the third flow path 43C are connected at the connecting portion 55.
- the sample liquid 50 flowing through the first flow path 43A and the diluent liquid 51 flowing through the second flow path 43B join together at the connecting portion 55 and are mixed.
- a mixed liquid obtained by mixing the sample liquid 50 and the diluent 51 flows through the third channel 43C.
- a static mixer 45 is provided on the third flow path 43C.
- the sample liquid 50 and diluent 51 contained in the mixed liquid are stirred by passing through the inside of the static mixer 45 .
- the mixing ratio of the liquid mixture flowing through the third channel 43C changes continuously as the flow rate ratio (A:B) of the sample liquid 50 and the diluent 51 changes continuously. That is, the mixing ratio has a correlation with the flow ratio (A:B) between the sample liquid 50 and the diluent 51 .
- a first flow cell 46A and a second flow cell 46B are provided downstream of the static mixer 45 in the third channel 43C.
- the first sensor 47A acquires first time-series data indicating changes in the mixing ratio of the liquid mixture flowing through the first flow cell 46A.
- the first sensor 47A may be any sensor capable of outputting a measured value having a correlation with the mixing ratio of the sample liquid 50 and the diluent 51 .
- a UV-vis (ultraviolet visible light) spectrophotometer can be used as the first sensor 47A.
- the UV-vis spectrophotometer irradiates the liquid mixture flowing through the first flow cell 46A with light divided by wavelength, and detects the intensity of the light transmitted through the liquid mixture, thereby detecting a specific wavelength of the liquid mixture (for example, 280 nm) is output.
- the absorbance output from the first sensor 47A has a correlation with the mixing ratio of the sample liquid 50 and the diluent 51 in the liquid mixture. For example, as the ratio of the sample liquid 50 contained in the mixture decreases, the concentration of the antibody contained in the mixture decreases. This reduces the absorbance in the mixed liquid. If the absorbance exceeds the range depending on the concentration of the mixed solution or the linearity of the absorbance with respect to the mixture ratio is lost, the wavelength for measuring the absorbance may be changed.
- the first sensor 47A outputs, as first time-series data, the change in the absorbance of the mixed liquid that changes continuously with the continuous change in the flow rate ratio (A:B) of the sample liquid 50 and the diluent 51. do.
- the first time-series data output from the first sensor 47A is transmitted to the information processing device 100 .
- the second sensor 47B acquires second time-series data indicating changes in spectral data of the liquid mixture flowing through the second flow cell 46B.
- This spectral data is the same spectral data that is input to the soft sensor 10 .
- the Raman spectrophotometer 20 illustrated in FIG. 2 can be used as the second sensor 47B.
- the second sensor 47B captures the transition of the spectral data of the liquid mixture that changes continuously with the continuous change in the flow rate ratio (A:B) of the sample liquid 50 and the diluent 51 as second time-series data. output as The second time-series data output from the second sensor 47B is transmitted to the information processing device 100.
- the number of samplings in the first time-series data and the second time-series data is preferably 25 or more, more preferably 50 or more, and most preferably 100 or more.
- the liquid mixture that has passed through the first flow cell 46A and the second flow cell 46B is recovered in the recovery container 49.
- FIG. 6 is a diagram showing an example of the hardware configuration of the information processing device 100. As shown in FIG. FIG. 6 illustrates a configuration in which the information processing apparatus 100 has a function of generating learning data, a function of constructing the software sensor 10 using the learning data, and a function of operating as the software sensor 10. ing.
- the information processing device 100 is an example of a “recording processing unit” in technology disclosed herein.
- the information processing apparatus 100 includes a CPU (Central Processing Unit) 101, a RAM (Random Access Memory) 102 as a temporary storage area, a nonvolatile memory 103, a display 104, an input device 105 such as a keyboard and a mouse, and a first sensor 47A. and an external interface 106 to which the second sensor 47B is connected.
- CPU 101 , memory 102 , nonvolatile memory 103 , display 104 , input device 105 and external interface 106 are connected to bus 107 .
- the nonvolatile memory 103 is a nonvolatile recording medium such as an HDD (Hard Disk Drive), SSD (Solid State Drive), or flash memory.
- a learning data generation program 110 , an estimation model 111 , a software sensor construction program 112 and an estimation program 113 are stored in the nonvolatile memory 103 .
- the learning data generation program 110 relates to the function of generating learning data.
- Estimation model 111 and soft sensor construction program 112 are associated with the function of constructing soft sensor 10 .
- Estimation program 113 is associated with the function of operating as soft sensor 10 .
- a RAM 102 is a work memory for the CPU 101 to execute processing. The CPU 101 loads each program stored in the nonvolatile memory 103 to the RAM 102 and executes processing according to each program.
- the function of generating learning data, the function of constructing the software sensor 10 using the learning data, and the function of operating as the software sensor 10 may be realized by different information processing devices (hardware).
- FIG. 7 is a flowchart showing an example of the flow of processing executed by the CPU 101 executing the learning data generation program 110.
- the learning data generation program 110 is executed, for example, when the user operates the input device 105 to give an instruction to start processing.
- step S1 the CPU 101 transmits first time-series data indicating changes in the mixing ratio of the liquid mixture and second time-series data indicating changes in spectrum data of the liquid mixture to the first sensor 47A and the second sensor 47A, respectively. is acquired from the sensor 47B.
- step S2 the CPU 101 records the first time-series data and the second time-series data in the non-volatile memory 103 with time information indicating the time when they were acquired.
- the time information may be given by the first sensor 47A and the second sensor 47B.
- the time information indicates the measurement times of the first time-series data and the second time-series data.
- the first time-series data and the second time-series data are obtained based on the flow path length between the sensors and the flow rate of the mixed liquid. You may correct
- step S3 the CPU 101 normalizes the first time-series data acquired in step S1. Specifically, the CPU 101 normalizes the absorbance output from the UV-vis spectrophotometer, which is the first sensor 47A, so that its maximum value is "1". That is, "1" is assigned to the absorbance obtained when the flow rate ratio (A:B) of the sample liquid 50 and the diluent liquid 51 is 1:0, and when the flow rate ratio (A:B) is 0:1 A "0" is assigned to the absorbance obtained in .
- Absorbances obtained at intermediate stages of the flow ratio (A:B) from 0:1 to 1:0 are assigned numerical values greater than "0" and less than "1" according to the values.
- step S4 the CPU 101 derives the third time-series data showing the transition of the concentration of the antibody contained in the mixture based on the standardized first time-series data. Specifically, the CPU 101 determines the mixture ratio C (0 ⁇ C ⁇ 1) at each time indicated by the standardized first time-series data and the known concentration Q1 of the antibody in the sample liquid 50.
- a third time series data is derived by calculating the product (Q1*C).
- the product (Q1 ⁇ C) indicates the concentration of antibody at that point in the mixture.
- the third time series data can be obtained by arranging the products (Q1 ⁇ C) in time series. Time information is added to the third time series data based on the time information added to the first time series data.
- the first time-series data and the third time-series data may be subjected to preprocessing such as time averaging in order to smooth noise.
- step S5 the CPU 101 generates learning data based on the second time-series data acquired in step S1 and the third time-series data derived in step S4. Specifically, the CPU 101 generates a plurality of learning data in which spectrum data at a plurality of time points of the second time-series data are associated with antibody concentrations at corresponding time points of the third time-series data. do. The CPU 101 associates data at the same point in time of the second time-series data and the third time-series data with each other based on the time information given to them. As a result, a plurality of pieces of learning data can be obtained, each unit of which is a combination of spectrum data and antibody concentration that are in correspondence with each other.
- the second time-series data and the third time-series data if there is no data at the same point in time, at least one of the second time-series data and the third time-series data is obtained at the sampling point by linear interpolation or the like. may be interpolated.
- step S6 the CPU 101 records the plurality of learning data generated in step S5 in the nonvolatile memory 103.
- the learning data acquisition method continuously changes the flow rate ratio of the sample liquid 50 and the diluent liquid 51 in which the concentration of a specific component (antibody) is known. While the sample liquid 50 and the diluent 51 are being mixed, the first time-series data showing the transition of the mixing ratio and the second data showing the transition of the spectrum data of the mixed liquid obtained by mixing.
- time-series data deriving third time-series data showing changes in the concentration of a specific component (antibody) contained in the mixture based on the first time-series data, and generating second time-series data and acquiring learning data in which the spectrum data and the concentration of a specific component (antibody) are associated from the and third time-series data.
- the learning data generated by the information processing device 100 is used to construct the soft sensor 10.
- the construction of the soft sensor 10 is performed by the information processing device 100 .
- the construction of the soft sensor 10 may be performed by an information processing apparatus different from the information processing apparatus used to generate the learning data.
- the soft sensor 10 is constructed by making the estimation model 111 learn using learning data according to the soft sensor construction program 112 .
- FIG. 8 is a diagram showing an example of the structure of the estimation model 111. As shown in FIG.
- the estimation model 111 is assumed to be a neural network including an input layer, multiple intermediate layers, and an output layer.
- the input layer of the estimation model 111 receives spectral data from Raman scattered light.
- the output layer of the estimation model 111 outputs concentration data corresponding to the spectral data input to the input layer.
- FIG. 9 is a flowchart showing an example of the flow of processing executed by the CPU 101 executing the software sensor construction program 112.
- FIG. The soft sensor construction program 112 is executed, for example, when the user inputs a process execution instruction via the input device 105 after acquiring the learning data.
- step S11 the CPU 101 extracts one of the plurality of learning data recorded in the nonvolatile memory 103 in step S6 (see FIG. 7) of the learning data generation process.
- step S12 the CPU 101 inputs the learning spectrum data included in the learning data extracted in step S11 to the estimation model 111.
- the learning data may be preprocessed as described in Japanese Patent Application No. 2020-075480.
- step S13 the CPU 101 adjusts the estimation model so that the difference between the density data output from the estimation model 111 and the learning density data (that is, correct data) included in the learning data extracted in step S11 is small.
- the estimation model 111 is learned. For example, an error backpropagation method may be applied as a learning method for the estimation model 111 .
- step S14 the CPU 101 determines whether or not the processing from steps S11 to S13 has been completed for all learning data.
- the processing returns to step S11. This routine ends when the processing for all the learning data is completed.
- the soft sensor 10 is constructed by making the estimation model 111 learn using the learning data.
- the soft sensor 10 constructed as described above estimates the concentration of antibodies contained in a liquid based on spectral data obtained for a liquid with an unknown antibody concentration. It is possible to derive
- the information processing apparatus 100 functions as the software sensor 10 by the CPU 101 executing the estimation program 113 . Note that the function of operating as the software sensor 10 may be realized on an information processing apparatus different from the information processing apparatus used to generate the learning data and the information processing apparatus used to construct the software sensor 10 .
- FIG. 10 is a flowchart showing an example of the flow of processing performed by the CPU 101 executing the estimation program 113.
- the estimation program 113 is executed, for example, when the user inputs a process execution instruction via the input device 105 after the software sensor 10 is constructed.
- step S21 the CPU 101 acquires spectral data measured using a Raman spectrophotometer or the like for a liquid with an unknown antibody concentration.
- Spectral data may be obtained in-line, for example, in an antibody purification process.
- step S22 the CPU 101 inputs the spectral data acquired in step S21 to the learned estimation model 111. Based on the spectral data, the trained estimation model 111 derives concentration data indicating an estimated concentration of the antibody contained in the liquid.
- the CPU 101 outputs the density data derived at step S22.
- the CPU 101 may, for example, perform control to display density data on the display 104 .
- the following method can be considered as a method of acquiring learning data according to the comparative example.
- a process e.g., a cell culture process, a purification process, etc.
- spectral data is obtained for the treatment liquid in that process
- quality information corresponding to the spectral data i.e., antibody concentration
- acquisition of quality information requires multiple samplings from the processing liquid and off-line analysis of the sampled processing liquid, which requires enormous time and labor.
- the learning data acquisition method while the sample liquid 50 and the diluent 51 are being mixed, the first time indicating the transition of the mixing ratio of the mixed liquid Second time-series data indicating transitions of the series data and the spectral data are obtained by in-line measurement by the first sensor 47A and the second sensor 47B, respectively. That is, acquisition of the first time-series data and the second time-series data is performed automatically. Also, the third time-series data indicating the transition of the concentration of the antibody contained in the mixed solution is derived by calculation from the first time-series data. A plurality of pieces of learning data can be acquired by associating data at the same point in time of the second time-series data and the third time-series data.
- the offline analysis only needs to be performed to know the antibody concentration of the sample liquid 50 .
- the learning data is generated based on the time-series data obtained for the mixed liquid obtained by mixing the sample liquid 50 and the diluent 51 while continuously changing the flow rate ratio, the learning data diversity can be ensured. That is, according to the learning data acquisition method according to the embodiment of the disclosed technology, it is possible to efficiently acquire learning data used for machine learning of the soft sensor 10 .
- the number of offline analyzes can be reduced to 1/7 or less compared to the method according to the comparative example described above.
- the software sensor 10 is effective in, for example, the production of biopharmaceuticals in situations where estimating the quality state in real time is a great advantage.
- the soft sensor 10 can be used to control the timing of recovering the antibody eluted from the column in the antibody purification process by chromatography.
- the UV absorbance of the treatment solution is monitored by a UV sensor, and the valve is switched at the timing when the UV absorbance exceeds a predetermined value, and the treatment solution containing the antibody is separated. are collecting. That is, the valve switching timing is controlled based on the UV absorbance output from the UV sensor.
- UV absorbance cannot distinguish between antibodies and contaminants other than antibodies. Therefore, there is a risk that the valve will be switched at inappropriate timing, and the purity of the antibody in the recovered treatment liquid will not reach the target value.
- the soft sensor 10 By using the soft sensor 10 according to the present embodiment in place of the UV sensor, it is possible to independently estimate the concentration of antibodies even for a treatment liquid containing both antibodies and contaminants. This makes it possible to appropriately control the timing of switching the valves, making it possible to recover a treated liquid containing antibodies purified to a desired degree of purity.
- the software sensor 10 can quickly detect the abnormality. This can prevent significant rework of the process.
- the "specific component” from which the concentration data is to be derived is an antibody contained in the culture solution obtained by cell culture, but it is not limited to this aspect.
- the “specific component” from which concentration data is to be derived may be contaminants other than antibodies contained in the culture medium.
- the contaminants may be at least one of antibody aggregates, antibody fragments, charge isomers, immature sugar chains, host cell proteins (HCPs), or cell-derived DNA. If biopharmaceuticals are contaminated with such contaminants as described above, even a very small amount of such contaminants may affect the efficacy of the biopharmaceutical. Therefore, it is important to know the concentration of contaminants in the treated liquid obtained by purification treatment for purifying antibodies.
- the sample liquid 50 used to acquire learning data contains contaminants with known concentrations.
- the diluent 51 may be a liquid containing buffer components containing any one or more of phosphoric acid, acetic acid, tris, and citric acid and containing no contaminants.
- the procedure for acquiring learning data is the same as the above-described case of acquiring the estimated antibody concentration as concentration data. That is, for the liquid mixture obtained by mixing the sample liquid 50 and the diluent 51, the first time-series data indicating the transition of the mixing ratio and the second time-series data indicating the transition of the spectrum data of the mixture are generated. , from the first sensor 47A and the second sensor 47B.
- the first time-series data is normalized. Specifically, the absorbance output from the UV-vis spectrophotometer, which is the first sensor 47A, is normalized so that its maximum value is "1".
- the third time-series data showing the transition of the concentration of contaminants contained in the mixed liquid is derived, specifically, known in the sample liquid 50
- the product (Q2 ⁇ C) indicates the concentration of contaminants in the mixture at that time point.
- the third time series data can be obtained by arranging the products (Q2 ⁇ C) in time series. Time information is added to the third time series data based on the time information added to the first time series data.
- learning data is generated based on the second time-series data and the third time-series data. Specifically, a plurality of learning data are generated by associating spectral data at a plurality of time points of the second time-series data with contaminant concentrations at corresponding time points of the third time-series data. That is, the data at the same point in the second time-series data and the third time-series data are associated with each other based on the time information attached to them. As a result, a plurality of pieces of learning data can be obtained, each unit of which is a combination of spectrum data and contaminant concentrations that are in correspondence with each other.
- the "specific component" from which concentration data is derived may include both antibodies and contaminants other than antibodies contained in the culture medium. That is, it is possible to construct the soft sensor 10 such that it simultaneously derives an estimate of the concentration of the antibody and the concentration of at least one contaminant.
- the sample liquid 50 used to acquire learning data contains an antibody with a known concentration and at least one type of contaminant with a known concentration.
- the diluent 51 can be a liquid containing a buffer component containing any one or more of phosphoric acid, acetic acid, tris, and citric acid, and free of antibodies and contaminants.
- the procedure for acquiring learning data is the same as the above-described case of acquiring the estimated antibody concentration as concentration data.
- the first time-series data indicating the transition of the mixing ratio and the second time-series data indicating the transition of the spectrum data of the mixture are generated.
- the first sensor 47A and the second sensor 47B are normalized.
- the absorbance output from the UV-vis spectrophotometer, which is the first sensor 47A is normalized so that its maximum value is "1".
- the third time-series data are derived, which respectively show changes in the concentration of the antibody and the concentration of contaminants contained in the mixture, specifically , the product of the antibody concentration Q1 and the contaminant concentration Q2, which are known in the sample liquid 50, and the mixing ratio C (0 ⁇ C ⁇ 1) at each time indicated by the normalized first time-series data
- a third time series is derived for each antibody and contaminant by calculating (Q1 ⁇ C, Q2 ⁇ C).
- the product (Q1 ⁇ C) indicates the concentration of antibody at that point in the mixture.
- the product (Q2 ⁇ C) indicates the concentration of contaminants in the mixture at that time point.
- third time series data can be obtained for each of the antibody and contaminants.
- Time information is added to the third time series data based on the time information added to the first time series data.
- learning data is generated based on the second time-series data and the third time-series data.
- a plurality of learning data in which the spectrum data at a plurality of time points of the second time-series data and the concentrations of antibodies and contaminants at each corresponding time point of the third time-series data are associated Generate. That is, the data at the same point in the second time-series data and the third time-series data are associated with each other based on the time information attached to them.
- a plurality of pieces of learning data can be obtained, each of which is a combination of spectral data having a corresponding relationship with each other, antibody concentration, and contaminant concentration.
- multiple soft sensors 10 can be configured to derive concentration estimates for different components (eg, antibodies and contaminants).
- the spectrum of Raman scattered light is used as an example, but the present invention is not limited to this mode.
- the absorption spectrum of infrared rays irradiated to the liquid infrared absorption spectrum
- fluorescence fluorescence spectrum
- UV-Vis absorption spectrum UV-Vis absorption spectrum
- Raman scattered light spectrum which has a higher correlation to the concentration data.
- Measured values other than spectral data include, for example, optically detected values of temperature, absorbance, conductivity, hydrogen ion concentration, refractive index, or light scattering measured for the liquid from which concentration data is derived by the software sensor 10.
- the fourth time-series data may be time-series data of two or more types of measured values of the liquid mixture.
- the case of using a UV-vis spectrophotometer that outputs the absorbance of the liquid mixture as the first sensor 47A was exemplified, but it is not limited to this aspect.
- the first sensor 47A it is also possible to use a sensor that outputs optically detected values of electrical conductivity, hydrogen ion concentration, refractive index, or light scattering for the liquid mixture.
- the first time-series data may be obtained based on optically detected values of absorbance, conductivity, hydrogen ion concentration, refractive index, or light scattering measured for the liquid mixture.
- the electrical conductivity, hydrogen ion concentration, refractive index, or optically detected value of light scattering measured for the mixed liquid are all correlated with the mixing ratio of the mixed liquid, and can be used instead of the absorbance.
- the present invention is limited to this aspect. not.
- the first sensor 47A and the second sensor 47B are of the same type, it is possible to make the correspondence without time information.
- both the first sensor 47A and the second sensor 47B are UV-vis spectrophotometers that output the absorbance at a specific wavelength of the mixture, at a common wavelength (eg, 280 nm) in the output of both sensors , data showing the same absorbance may be associated with each other. This is because the absorbances output from both sensors at the same time and at the same wavelength are the same.
- the learning data acquisition system 40 shown in FIG. 4 can be modified in various ways.
- the probes of the first sensor 47A and the second sensor 47B may be arranged inside the static mixer 45 or inside a chamber (not shown) provided in the flow path. good.
- the soft sensor 10 is constructed by machine learning was exemplified. It may be constructed using a technique.
- Example 1 Learning data for constructing the soft sensor 10 that derives the estimated value of the concentration of the antibody contained in the liquid as concentration data was acquired. Moreover, the software sensor 10 was constructed using the obtained learning data. The details are described below.
- sample solution As a stock solution of the sample solution, a solution obtained by removing cells from a Chinese Hamster Ovary (CHO) cell culture medium was used. In addition to antibody proteins produced by cells, this solution is contaminated with glucose, lactic acid, amino acids, ammonia, antibody aggregates, antibody fragments, charge isomers, immature sugar chains, cell-derived proteins, and cell-derived DNA. included as objects. In this example, this solution will be referred to as "culture supernatant". Note that the technique described in WO2019/117136 can be applied to establish CHO cells.
- a protein A column is a column filled with a protein-A-presented resin that specifically adsorbs an antibody on the surface of porous beads, and is a kind of column commonly used in antibody purification processes.
- a Protein A column (Cytiva, MabSelect SuRe) was used in conjunction with a chromatography system (Cytiva, AKTA pure 25).
- the culture supernatant was introduced from the input line of the chromatography device, and the antibody contained in the liquid was specifically adsorbed onto the protein A column.
- a liquid containing contaminants that was not adsorbed to the protein A column was collected from the discharge line as a "flow-through liquid".
- a washing buffer (20 mM sodium phosphate, 150 mM sodium chloride, pH 7.2) was introduced from the buffer line of the chromatography device to wash away contaminants nonspecifically adsorbed to the protein A column. The solution discharged from the protein A column at this time was collected as a "wash fraction”.
- an elution buffer (0.1 M sodium citrate, pH 3.0) was introduced from the buffer line of the chromatography device to desorb the antibody specifically adsorbed on the protein A column. The solution eluted from the protein A column at this time was collected as an "eluted fraction".
- sample liquids 50 The samples taken from the culture supernatant, the flow-through fraction, the washing fraction, and the elution fraction were each designated as 50 sample liquids.
- the antibody concentration was measured by off-line analysis by HPLC. Thus, four sample liquids 50 with known antibody concentrations were obtained.
- a connection portion 55 to which the first flow path 43A, the second flow path 43B, and the third flow path 43C are connected is configured by a T-shaped tube.
- Plunger pumps were used as the first pump 44A and the second pump 44B.
- a UV-vis spectrophotometer was used as the first sensor 47A.
- a Raman spectrophotometer (Kaiser optical systems, Kaiser Raman RXN2 Analyzer) was used as the second sensor 47B.
- One of the four sample liquids 50 described above was contained in the first container 41, and the diluent 51 was contained in the second container.
- the diluent 51 the above-described washing buffer, elution buffer, and liquid medium for cell culture were used.
- the first pump 44A and the second pump 44B were controlled so that the total flow rate of the sample liquid 50 flowing through the first channel 43A and the diluent 51 flowing through the second channel 43B was maintained at 1 mL/min. .
- the flow rate ratio (A:B) which is the ratio of the flow rate A of the sample liquid 50 and the flow rate B of the diluent 51, was controlled to be 0:1.
- the first pump 44A and the second pump 44B were controlled so that the flow rate ratio (A:B) changed linearly from 0:1 to 1:0 during 20 minutes.
- the absorbance at a wavelength of 280 nm is measured by the UV-vis spectrophotometer, which is the first sensor 47A, for the mixed liquid in which the sample liquid 50 and the diluent 51 flowing through the third flow path 43C are mixed. It was measured. Changes in absorbance that change with changes in the flow rate ratio (A:B) were obtained as first time-series data. Further, while the liquid was being sent, the Raman spectrophotometer, which was the second sensor 47B, acquired spectral data of Raman scattered light for the liquid mixture flowing through the third channel 43C.
- the laser output was 200 mW
- the excitation wavelength was 785 nm
- the exposure time was 1 second
- the number of times of integration was 15 times.
- the transition of spectral data that changes with changes in the flow rate ratio (A:B) was obtained as second time-series data.
- the first time-series data and the second time-series data are recorded in the non-volatile memory 103 included in the information processing apparatus 100 with time information indicating the time when they were acquired.
- the mixed solution was sampled from the discharge line at a time of 1 mL each time while the solution was being sent.
- Antibody concentrations were determined for each sample by off-line analysis by HPLC. This verification may be performed as necessary and can be omitted.
- the first time-series data and the second time-series data were obtained for the remaining three kinds of sample liquids 50 according to the same procedure as described above.
- the absorbance at each time point indicated by the first time-series data was normalized so that its maximum value was "1".
- the product (Q1 ⁇ C) of the antibody concentration Q1 which is known in the sample liquid 50, and the mixture ratio C (0 ⁇ C ⁇ 1) indicated by the normalized first time-series data is calculated.
- the third time-series data was derived by That is, the third time-series data was obtained by arranging the products (Q1 ⁇ C) in time series.
- FIG. 11 is a graph comparing the antibody concentration indicated by the third time-series data and the antibody concentration actually measured by offline analysis of the sampled mixture. As shown in FIG. 11 , the two results were approximately the same, so it was verified that the antibody concentration can be estimated from the first time-series data.
- the spectral data at multiple time points of the second time-series data were associated with the antibody concentration at each corresponding time point of the third time-series data.
- the data at the same point in the second time-series data and the third time-series data are associated with each other based on the time information attached to them.
- the sampling points were interpolated by linear interpolation and associated with each other.
- a plurality of pieces of learning data were generated, each unit of which is a combination of spectrum data and antibody concentration that are in correspondence with each other.
- a culture lot different from the culture lot of the sample liquid 50 was subjected to separation processing by immunoaffinity chromatography using a protein A column.
- a gradient elution was performed in the range of 10 CV with a continuous switch from wash buffer to elution buffer during elution.
- CV indicates the volume of the protein A column.
- the Raman spectrum was collected using the flow cell installed in the channel, and at the same time, 0.5 CV sampling was performed using the fraction collector.
- An estimate of the antibody concentration by the soft sensor 10 was derived using the collected spectra as input. In addition, the antibody concentration was measured for each collected elution fraction by off-line analysis by HPLC.
- the coefficient of determination R2 and Root Mean Square Error (RMSE) were obtained.
- the coefficient of determination R2 was 0.99 and the RMSE was 0.39 for the soft sensor 10 constructed by the technique according to this example.
- Example 2 Learning data for constructing a soft sensor 10 that derives an estimated value of the concentration of a host cell-derived protein (HCP), which is a type of contaminant contained in a liquid, as concentration data was obtained. Moreover, the software sensor 10 was constructed using the obtained learning data. The details are described below.
- HCP host cell-derived protein
- sample liquid 50 was obtained by sampling the liquid.
- concentration of HCP was measured by off-line analysis by HPLC.
- concentration of HCP was measured using the 360-HCP ELISA kit (Cosmo Bio).
- a culture lot different from the culture lot of the sample liquid 50 was subjected to separation processing by immunoaffinity chromatography using a protein A column.
- a gradient elution was performed in the range of 10 CV with a continuous switch from wash buffer to elution buffer during elution.
- the Raman spectrum was collected using the flow cell installed in the channel, and at the same time, 0.5 CV sampling was performed using the fraction collector.
- An estimate of the HPC concentration by the soft sensor 10 was derived using the collected spectra as input.
- the HCP concentration of the sampled elution fraction was measured by off-line analysis by HPLC.
- the coefficient of determination R2 and RMSE were obtained to assess the accuracy of the HCP concentration estimates in the soft sensor 10 relative to the actual measurements.
- the coefficient of determination R2 was 0.96 and the RMSE was 34.11 for the soft sensor 10 constructed by the technique according to this example.
- the CHO cell culture supernatant was separated by immunoaffinity chromatography using a protein A column.
- a protein A column (Cytiva, MabSelect SuRe) was used in conjunction with a chromatography system (Cytiva, AKTA pure 25).
- an acidic eluate was introduced from the buffer line of the chromatography device to desorb the antibody specifically adsorbed on the protein A column.
- it was introduced in a gradient manner while creating a concentration gradient (gradient elution).
- Three patterns of gradient elution conditions were used as follows. Gradient elution was performed using 5 CV, 10 CV and 15 CV volumes of the eluate, where CV is the volume of the protein A column.
- Raman scattered light spectra were measured at multiple points in time during process operation.
- the elution fraction was sampled at multiple points during process operation, and the antibody concentration was obtained by off-line analysis of the sampled elution fraction.
- a plurality of data for learning was acquired by associating spectra at a plurality of points in time during process operation with antibody concentrations at the corresponding points.
- the software sensor 10 was constructed by making the estimation model 111 learn using the plurality of learning data acquired as described above. Using the soft sensor 10, a culture lot different from the culture lot used to acquire the learning data was subjected to separation processing by immunoaffinity chromatography using protein A. Here, a gradient elution was performed in the range of 10 CV with a continuous switch from wash buffer to elution buffer during elution. During the elution process, the Raman spectrum was collected using the flow cell installed in the channel, and at the same time, 0.5 CV sampling was performed using the fraction collector. An estimate of the antibody concentration by the soft sensor 10 was derived using the collected spectra as input.
- the sampled elution fraction was subjected to off-line analysis by HPLC to measure the antibody concentration.
- Coefficient of determination R2 and RMSE were obtained to assess the accuracy of the estimated antibody concentration in the soft sensor 10 relative to the measured values.
- the coefficient of determination R2 was 0.98 and the RMSE was 0.53.
- the accuracy of the soft sensor 10 constructed using the learning data acquired by the method according to the disclosed technique is higher than that of the learning data acquired by the method according to the comparative example, which mainly consists of offline analysis. It can be said that it is equivalent to the soft sensor 10 constructed using.
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biochemistry (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Bioethics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Public Health (AREA)
- Epidemiology (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biotechnology (AREA)
- Signal Processing (AREA)
- Evolutionary Biology (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
Description
ータに対応する品質情報(ここでは特定の成分の濃度)を取得することが行われている。しかしながら、多様な品質情報の取得には処理液中から複数回に亘ってサンプリングを行い、サンプリングされた処理液についてオフライン分析を行う必要があり、膨大な時間と労力を要する。更に、学習用データの多様性を確保するためには、処理条件を作為的に変更した多数のバッチについて、スペクトルデータ及び品質情報を取得する必要がある。この場合、バッチ数とサンプリング数との積に相当する回数のオフライン分析が必要となる。また、処理液の品質は、プロセスに依存するため、学習用データにおける品質の分布が制限されやすい。すなわち、実プロセスにおける処理液のサンプリングによって学習用データを取得する場合、多様な品質状態を網羅することは困難であった。
宜省略する。
状態を網羅した学習用データを用意する必要がある。以下において、開示の技術の実施形態に係る学習用データの取得方法について説明する。以下の説明においては、ソフトセンサ10が、ラマン散乱光によるスペクトルデータを入力とし、スペクトルデータに基づいて、液体に含まれる抗体の濃度の推定値を導出する場合を例示する。
させるように第1のポンプ44A及び第2のポンプ44Bを制御する。より具体的には、制御部48は、図5に示すように、サンプル液50の流量Aと希釈液51の流量Bとの合計値を一定(例えば1mL/min)に保ちながら、所定期間内(例えば10分間)に、流量比(A:B)が0:1から1:0に直線的に変化するようにサンプル液50及び希釈液51の送液を制御する。このような送液をリニアグラジエント送液という。なお、流量比(A:B)が、1:0から0:1に直線的に変化するようにサンプル液50及び希釈液51の送液を制御してもよい。また、サンプル液50と希釈液52の流量比を、ステップ状(段階的)に変化させてもよい。また、流路比(A:B)を変化させる範囲は、0:1から1:0までの全範囲に限らず、一部の範囲(例えば、0.2:0.8から0.8:0.2までの範囲)であってもよい。この場合、実プロセスで扱う濃度範囲をカバーしていることが好ましい。しかしながら、学習用データを網羅的に取得する観点から、流量比(A:B)を変化させる範囲は0:1から1:0までの全範囲であることが好ましい。また、リニアグラジエント送液を開始する前に、サンプル液50及び希釈液51による流路置換送液を行うことが好ましい。
センサ47Bから出力される第2の時系列データは情報処理装置100に送信される。なお、第1の時系列データ及び第2の時系列データにおけるサンプリング数は、25以上が好ましく、50以上が更に好ましく、100以上が最も好ましい。第1のフローセル46A及び第2のフローセル46Bを通過した混合液は、回収容器49に回収される。
してのRAM(Random Access Memory)102、及び不揮発性メモリ103、ディスプレイ104、キーボード及びマウス等の入力装置105、第1のセンサ47A及び第2のセンサ47Bが接続される外部インターフェース106を含む。CPU101、メモリ102、不揮発性メモリ103、ディスプレイ104、入力装置105及び外部インターフェース106は、バス107に接続される。
、UVセンサにより処理液のUV吸光度をモニタリングし、UV吸光度が所定値以上になるタイミングでバルブを切替え、抗体が含まれる処理液を回収している。すなわち、バルブを切り替えるタイミングをUVセンサから出力されるUV吸光度に基づいて制御している。しかしながら、UV吸光度では抗体と抗体以外の夾雑物とを区別することができない。このため、バルブを不適切なタイミングで切り替えてしまい、回収した処理液において、抗体の純度が目標値に達しないおそれがある。
わせを1単位とする複数の学習用データを得ることができる。
率、水素イオン濃度、屈折率、又は光散乱の光学的検出値)の推移を示す第4の時系列データを更に取得し、第2の時系列データと第3の時系列データと第4の時系列データから、上記の測定値とスペクトルデータと特定の成分の濃度とを対応付けた学習用データを取得する。なお、上記において例示したものの2つ以上の組み合わせをソフトセンサ10の入力データとして用いることも可能である。すなわち、第4の時系列データは、混合液について測定された2種類以上の測定値の時系列データであってもよい。
もよい。
(実施例1)
液体に含まれる抗体の濃度の推定値を濃度データとして導出するソフトセンサ10を構築するための学習用データを取得した。また、取得した学習用データを用いてソフトセンサ10を構築した。以下にその詳細について説明する。
サンプル液の原液として、チャイニーズハムスター卵巣(CHO:Chinese Hamster Ovary)細胞培養液から除細胞した溶液を使用した。この溶液には細胞が産生した抗体タンパク質の他に、グルコース、乳酸、アミノ酸、アンモニア、抗体の凝集物、抗体の断片、電荷異性体、未成熟糖鎖、細胞由来タンパク質、細胞由来DNAなどが夾雑物として含まれる。本実施例では、この溶液を「培養上清液」と呼ぶことにする。なお、CHO細胞の樹立については、WO2019/117136に記載の技術を適用することが可能である。
上記した4種類のサンプル液50のそれぞれについて、図4に示す学習用データ取得システム40を用いてソフトセンサ10を構築するための学習用データを取得した。第1の流路43A、第2の流路43B及び第3の流路43Cが接続される接続部55をT字管によって構成した。第1のポンプ44A及び第2のポンプ44Bとしてプランジャーポンプを用いた。第1のセンサ47AとしてUV-vis分光光度計を用いた。第2のセンサ47Bとしてラマン分光光度計(Kaiser optical systems、Kaiser Raman RXN2 Analyzer)を用いた。
ポンプ44A及び第2のポンプ44Bを制御した。初めに、サンプル液50の流量Aと希釈液51の流量Bとの比である流量比(A:B)が0:1になるように制御した。送液が安定した後、流量比(A:B)が20分間の間に0:1から1:0に直線的に変化するように第1のポンプ44A及び第2のポンプ44Bを制御した。
示す時刻情報を付与して情報処理装置100が備える不揮発性メモリ103に記録した。また、第1の時系列データから混合液における抗体の濃度の推移を推定できることを検証するために、送液を行っている間、複数回に亘り、混合液を排出ラインから1mLずつサンプリングし、各サンプルについてHPLCによるオフライン分析によって抗体濃度を測定した。この検証は、必要に応じて行えばよく、省略することが可能である。残りの3種類のサンプル液50についても、上記と同様の手順に従って第1の時系列データ及び第2の時系列データを取得した。
以上のようにして生成した複数の学習用データを用いて、図9に示すフローチャートによって示される手順に従い推定モデル111を学習させることによりソフトセンサ10を構築した。
液体に含まれる、夾雑物の一種である宿主細胞由来タンパク質(HCP)の濃度の推定値を濃度データとして導出するソフトセンサ10を構築するための学習用データを取得した。また、取得した学習用データを用いてソフトセンサ10を構築した。以下にその詳細について説明する。
上記した実施例1と同様、培養上清液、及びプロテインAカラムを用いたイムノアフィニティクロマトグラフィーによる分離処理によって得られる素通り画分液、洗浄画分液、溶出画分液をサンプリングしたものを、それぞれサンプル液50とした。これらのサンプル液50のそれぞれについて、HPLCによるオフライン分析によってHCPの濃度を測定した。HCPの濃度は、360-HCP ELISAキット(コスモバイオ)を用いて測定した。こ
のようにして、HCPの濃度が既知とされた4種類のサンプル液50を得た。
上記した実施例1と同様、4種類のサンプル液50のそれぞれについて、図4に示す学習用データ取得システム40を用いて学習用データを取得した。すなわち、混合液について第1の時系列データ及び第2の時系列データを取得し、規格化された第1の時系列データによって示される各時点における混合比C(0≦C≦1)と、サンプル液50において既知とされたHCPの濃度Q2との積(Q2×C)を時系列に沿って並べることで、第3の時系列データを取得した。その後、第2の時系列データの複数の時点におけるスペクトルデータと、第3の時系列データの対応する各時点におけるHCPの濃度とを対応付けた。これにより、互いに対応関係にあるスペクトルデータとHCPの濃度との組み合わせを1単位とする複数の学習用データを生成した。
以上のようにして生成した複数の学習用データを用いて、図9に示すフローチャートによって示される手順に従い推定モデル111を学習させることによりソフトセンサ10を構築した。
本比較例では、抗体の精製工程においてサンプリングされた処理液のオフライン分析を主体とする手法によって学習用データを取得し、取得した学習用データを用いてソフトセンサ10を構築した。
に導入した(グラジエント溶出)。グラジエント溶出の条件を以下の通り3パターンとした。プロテインAカラムのボリュームをCVとしたとき、5CV、10CV、15CVの各分量の溶出液を用いてグラジエント溶出を実施した。上記したパターンのそれぞれについて、プロセス運転中の複数の時点においてラマン散乱光によるスペクトルを測定した。また、プロセス運転中の複数の時点において溶出画分液をサンプリングし、サンプリングした溶出画分液について、オフライン分析により抗体濃度を取得した。プロセス運転中の複数の時点におけるスペクトルと、対応する時点における抗体の濃度とを対応付けることより複数の学習用データを取得した。
Claims (17)
- 液体による作用を受けた電磁波の、波数または波長毎の強度を示すスペクトルデータに基づいて前記液体中に含まれる特定の成分の濃度を導出するソフトセンサの機械学習に用いる学習用データの取得方法であって、
前記特定の成分の濃度が既知であるサンプル液を用意し、
前記サンプル液と希釈液とをこれらの流量比を連続的に変化させながら混合し、
前記サンプル液と前記希釈液との混合を行っている間、前記混合によって得られる混合液について、混合比の推移を示す第1の時系列データ及び前記スペクトルデータの推移を示す第2の時系列データを取得し、
前記第1の時系列データに基づいて、前記混合液に含まれる前記特定の成分の濃度の推移を示す第3の時系列データを導出し、
前記第2の時系列データと前記第3の時系列データから、前記スペクトルデータと前記特定の成分の濃度とを対応付けた学習用データを取得する
取得方法。 - 前記第2の時系列データの複数の時点におけるスペクトルデータと、前記第3の時系列データの前記複数の時点に対応する各時点における前記特定の成分の濃度と、を対応付けた複数の学習用データを取得する
請求項1に記載の取得方法。 - 前記スペクトルデータがラマンスペクトル、赤外線吸収スペクトル、蛍光スペクトル又はUV-Vis吸収スペクトルによるものである
請求項1又は請求項2に記載の取得方法。 - 前記混合液について測定した吸光度、導電率、水素イオン濃度、屈折率、又は光散乱の光学的検出値に基づいて前記第1の時系列データを取得する
請求項1から請求項3のいずれか1項に記載の取得方法。 - 前記サンプル液は、前記特定の成分を分離する分離処理を行った処理液である
請求項1から請求項4のいずれか1項に記載の取得方法。 - 前記分離処理は、クロマトグラフィーによるものである
請求項5に記載の取得方法。 - 前記特定の成分は、タンパク質である
請求項1から請求項6のいずれか1項に記載の取得方法。 - 前記特定の成分は、細胞培養によって得られる培養液に含まれる抗体以外の夾雑物である
請求項1から請求項7のいずれか1項に記載の取得方法。 - 前記夾雑物は、抗体の凝集物、抗体の断片、電荷異性体、未成熟糖鎖、細胞由来タンパク質、細胞由来DNAのうちの少なくとも1つを含む
請求項8に記載の取得方法。 - 前記希釈液は、前記サンプル液に含まれる前記特定の成分を含む
請求項1から請求項9のいずれか1項に記載の取得方法。 - 前記希釈液は、前記サンプル液に含まれる前記特定の成分以外の成分のみを含む
請求項1から請求項9のいずれか1項に記載の取得方法。 - 前記混合液が流れる流路上に設けられた第1のセンサによって前記第1の時系列データを取得し、
前記流路上に設けられた第2のセンサによって前記第2の時系列データを取得する
請求項1から請求項11のいずれか1項に記載の取得方法。 - 前記サンプル液と前記希釈液との混合を行っている間、前記混合液について測定された少なくとも1種類の測定値の推移を示す第4の時系列データを更に取得し、
前記第2の時系列データと前記第3の時系列データと前記第4の時系列データから、前記測定値と前記スペクトルデータと前記特定の成分の濃度とを対応付けた学習用データを取得する
請求項1から請求項12のいずれか1項に記載の取得方法。 - 請求項1から請求項13のいずれか1項に記載の取得方法を実施するための学習用データ取得システムであって、
前記サンプル液が流通する第1の流路と、
前記希釈液が流通する第2の流路と、
前記混合液が流通する第3の流路と、
前記サンプル液の送液を行う第1のポンプと、
前記希釈液の送液を行う第2のポンプと、
前記第1のポンプ及び前記第2のポンプを制御する制御部と、
前記第3の流路上に設けられ、前記第1の時系列データを取得する第1のセンサと、
前記第3の流路上に設けられ、前記第2の時系列データを取得する第2のセンサと、
前記第1のセンサ及び前記第2のセンサの出力を記録媒体に記録する処理を行う記録処理部と、
を含む学習用データ取得システム。 - 請求項1から請求項13のいずれか1項に記載の取得方法によって取得された学習用データを用いて、前記ソフトセンサのモデルを学習させる
ソフトセンサの構築方法。 - 請求項1から請求項13のいずれか1項に記載の取得方法によって取得された学習用データを用いて学習されたソフトセンサ。
- 請求項1から請求項13のいずれか1項に記載の取得方法によって取得された学習用データ。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2023551066A JPWO2023053585A1 (ja) | 2021-09-30 | 2022-06-09 | |
EP22875464.4A EP4390379A1 (en) | 2021-09-30 | 2022-06-09 | Training data acquisition method, training data acquisition system, soft sensor construction method, soft sensor, and training data |
CN202280061879.3A CN117980998A (zh) | 2021-09-30 | 2022-06-09 | 学习用数据的获取方法、学习用数据获取系统、软传感器的构建方法、软传感器、学习用数据 |
US18/612,135 US20240232723A1 (en) | 2021-09-30 | 2024-03-21 | Method for acquiring learning data, learning data acquisition system, method for constructing soft sensor, soft sensor, and learning data |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2021-162035 | 2021-09-30 | ||
JP2021162035 | 2021-09-30 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US18/612,135 Continuation US20240232723A1 (en) | 2021-09-30 | 2024-03-21 | Method for acquiring learning data, learning data acquisition system, method for constructing soft sensor, soft sensor, and learning data |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023053585A1 true WO2023053585A1 (ja) | 2023-04-06 |
Family
ID=85782215
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2022/023317 WO2023053585A1 (ja) | 2021-09-30 | 2022-06-09 | 学習用データの取得方法、学習用データ取得システム、ソフトセンサの構築方法、ソフトセンサ、学習用データ |
Country Status (5)
Country | Link |
---|---|
US (1) | US20240232723A1 (ja) |
EP (1) | EP4390379A1 (ja) |
JP (1) | JPWO2023053585A1 (ja) |
CN (1) | CN117980998A (ja) |
WO (1) | WO2023053585A1 (ja) |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH1137935A (ja) * | 1997-07-18 | 1999-02-12 | Fuji Electric Co Ltd | 下水中の複数成分定量方法とその装置 |
JP2004198327A (ja) * | 2002-12-19 | 2004-07-15 | Japan Science & Technology Agency | 複数の化学物質の濃度の測定方法 |
JP2006234524A (ja) * | 2005-02-24 | 2006-09-07 | Shimadzu Corp | 液体クロマトグラフ用検出器 |
JP2011158340A (ja) * | 2010-01-29 | 2011-08-18 | Nippon Steel Corp | 排水中の特定化学物質又は特定排水の濃度測定方法及び検知方法並びに装置 |
WO2019117136A1 (ja) | 2017-12-11 | 2019-06-20 | 富士フイルム株式会社 | 動物細胞、動物細胞の製造方法および目的タンパク質の製造方法 |
JP2019530860A (ja) * | 2016-09-01 | 2019-10-24 | ジーイー・ヘルスケア・バイオサイエンス・アクチボラグ | 表面プラズモン共鳴アッセイのための段階的組み合わせ注入 |
JP2020075480A (ja) | 2018-11-09 | 2020-05-21 | 大日本印刷株式会社 | バリアフィルム |
JP2020101543A (ja) | 2018-12-20 | 2020-07-02 | キヤノン株式会社 | 情報処理装置、情報処理装置の制御方法、及びプログラム |
JP2021162035A (ja) | 2020-03-30 | 2021-10-11 | 日本電産シンポ株式会社 | 遊星減速機 |
-
2022
- 2022-06-09 EP EP22875464.4A patent/EP4390379A1/en active Pending
- 2022-06-09 CN CN202280061879.3A patent/CN117980998A/zh active Pending
- 2022-06-09 JP JP2023551066A patent/JPWO2023053585A1/ja active Pending
- 2022-06-09 WO PCT/JP2022/023317 patent/WO2023053585A1/ja active Application Filing
-
2024
- 2024-03-21 US US18/612,135 patent/US20240232723A1/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH1137935A (ja) * | 1997-07-18 | 1999-02-12 | Fuji Electric Co Ltd | 下水中の複数成分定量方法とその装置 |
JP2004198327A (ja) * | 2002-12-19 | 2004-07-15 | Japan Science & Technology Agency | 複数の化学物質の濃度の測定方法 |
JP2006234524A (ja) * | 2005-02-24 | 2006-09-07 | Shimadzu Corp | 液体クロマトグラフ用検出器 |
JP2011158340A (ja) * | 2010-01-29 | 2011-08-18 | Nippon Steel Corp | 排水中の特定化学物質又は特定排水の濃度測定方法及び検知方法並びに装置 |
JP2019530860A (ja) * | 2016-09-01 | 2019-10-24 | ジーイー・ヘルスケア・バイオサイエンス・アクチボラグ | 表面プラズモン共鳴アッセイのための段階的組み合わせ注入 |
WO2019117136A1 (ja) | 2017-12-11 | 2019-06-20 | 富士フイルム株式会社 | 動物細胞、動物細胞の製造方法および目的タンパク質の製造方法 |
JP2020075480A (ja) | 2018-11-09 | 2020-05-21 | 大日本印刷株式会社 | バリアフィルム |
JP2020101543A (ja) | 2018-12-20 | 2020-07-02 | キヤノン株式会社 | 情報処理装置、情報処理装置の制御方法、及びプログラム |
JP2021162035A (ja) | 2020-03-30 | 2021-10-11 | 日本電産シンポ株式会社 | 遊星減速機 |
Also Published As
Publication number | Publication date |
---|---|
JPWO2023053585A1 (ja) | 2023-04-06 |
CN117980998A (zh) | 2024-05-03 |
EP4390379A1 (en) | 2024-06-26 |
US20240232723A1 (en) | 2024-07-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6953508B2 (ja) | 製剤精製のリアルタイムモニタリング | |
Rüdt et al. | Real‐time monitoring and control of the load phase of a protein A capture step | |
Thakur et al. | An NIR‐based PAT approach for real‐time control of loading in protein A chromatography in continuous manufacturing of monoclonal antibodies | |
Brestrich et al. | Selective protein quantification for preparative chromatography using variable pathlength UV/Vis spectroscopy and partial least squares regression | |
Read et al. | Process analytical technology (PAT) for biopharmaceutical products: Part I. Concepts and applications | |
Yilmaz et al. | Application of Raman spectroscopy in monoclonal antibody producing continuous systems for downstream process intensification | |
US20180339244A1 (en) | Method for controlling continuous chromatography and multi-column chromatography arrangement | |
Rathore et al. | Chemometrics applications in biotech processes: a review | |
Rathore et al. | Application of process analytical technology for downstream purification of biotherapeutics | |
Xue et al. | Application of particle swarm optimization (PSO) algorithm to determine dichlorvos residue on the surface of navel orange with Vis-NIR spectroscopy | |
JP6196220B2 (ja) | 生体試料分析のための核磁気共鳴および近赤外線の使用 | |
CN112041663A (zh) | 用于生物制造的多元谱分析与监测 | |
Armstrong et al. | Advanced control strategies for bioprocess chromatography: Challenges and opportunities for intensified processes and next generation products | |
KR101832917B1 (ko) | 근적외선 분광분석기를 이용한 아미노산 발효공정의 실시간관리방법 | |
US20210269888A1 (en) | Method in Bioprocess Purification System | |
Ramakrishna et al. | Multi-wavelength UV-based PAT tool for measuring protein concentration | |
Hamla et al. | A new alternative tool to analyse glycosylation in pharmaceutical proteins based on infrared spectroscopy combined with nonlinear support vector regression | |
WO2023053585A1 (ja) | 学習用データの取得方法、学習用データ取得システム、ソフトセンサの構築方法、ソフトセンサ、学習用データ | |
US11867673B2 (en) | NIR based real-time control of loading in protein a chromatography | |
Wasalathanthri et al. | Paving the way for real time process monitoring in biomanufacturing | |
Chen et al. | Application of Raman spectroscopy during pharmaceutical process development for determination of critical quality attributes in Protein A chromatography | |
Wang et al. | Simultaneous prediction of 16 quality attributes during protein A chromatography using machine learning based Raman spectroscopy models | |
WO2023090015A1 (ja) | 情報処理装置、情報処理装置の作動方法、情報処理装置の作動プログラム、校正済み状態予測モデルの生成方法、並びに校正済み状態予測モデル | |
Wasalathanthri et al. | The Role of Process Analytical Technology (PAT) in Biologics Development | |
Rüdt | Spectroscopy as process analytical technology for preparative protein purification |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22875464 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 202280061879.3 Country of ref document: CN |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2023551066 Country of ref document: JP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2022875464 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 2022875464 Country of ref document: EP Effective date: 20240320 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |