WO2024111540A1

WO2024111540A1 - Protein collection determination system, protein collection determination method, and recording medium

Info

Publication number: WO2024111540A1
Application number: PCT/JP2023/041605
Authority: WO
Inventors: 大地末政
Original assignee: Jsr株式会社; 株式会社医学生物学研究所
Priority date: 2022-11-21
Filing date: 2023-11-20
Publication date: 2024-05-30

Abstract

This protein collection determination system comprises: a first determination unit that uses first image data obtained by capturing a time series of images of a culture of a cell population infected with a virus as first input data and determines, as first output data, whether a cell population infected with a virus is in a suitable state for collecting a protein; and an acquisition unit that acquires image data obtained by capturing a time series of images of the inside of a culture container. The first determination unit performs determination by means of a first model, and the first model is a trained model obtained by performing deep learning of a first neural network using the actual values of the first input data and the actual values of the first output data as first teacher data.

Description

Protein recovery determination system, protein recovery determination method, and recording medium

The present invention relates to a protein recovery determination system, a protein recovery determination method, and a recording medium.

In the production of proteins such as antigens, there is a method of multiplying cells into which the amino acid sequence of the protein has been introduced using a specific virus having the amino acid sequence of the target protein (see, for example, Non-Patent Document 1). In this method, for example, as shown in Figure 17, a portion of the cultured host cells is grown, and after a specific time has passed, the host cells are infected with the specific virus. After viral infection, when it is determined that the virus-infected host cells have produced a sufficient amount of protein, the protein is recovered. Figure 17 is a diagram for explaining the protein production process of the conventional method. Conventionally, workers have implicitly determined the time to recover the protein based on the results of observations made using a microscope (see Non-Patent Documents 1 and 2).

However, with conventional technology, the timing of recovery was determined by the operator by observing the images, which could vary from operator to operator. For example, when cell death occurs due to viral infection, problems can arise, such as degradation of the target protein by proteasomes contained within the cells, cell death due to abnormal infection caused by the release of viruses within the cells, and difficulty in purifying the target protein due to contamination of the culture medium by components contained in the cells other than the target protein. For this reason, there is a demand for proteins to be recovered at the appropriate time (when there is little cell death and the protein has been sufficiently produced by the cells).

The present invention was made in consideration of the above problems, and aims to provide a protein recovery determination system that can appropriately determine the timing for recovering a protein, a protein recovery method, and a recording medium that records a trained model.

The present invention includes the following embodiments.
(1) In order to achieve the above-mentioned object, a protein recovery determination system according to one embodiment of the present invention includes a first determination unit that inputs, as first input data, first image data obtained by photographing a culture of a virus-infected cell population in a time series, and determines, as first output data, whether the virus-infected cell population is in a state suitable for recovering a protein, and an acquisition unit that acquires image data by photographing the inside of a culture vessel in a time series, wherein the first determination unit makes a determination using a first model, and the first model is a trained model obtained by deep learning a first neural network using actual values of the first input data and actual values of the first output data as first teacher data.

(2) The protein recovery determination system described in (1) above further includes a second determination unit that inputs, as second input data, second image data obtained by photographing the expansion culture of a virus-uninfected cell population in a time series, and determines, as second output data, whether the virus-uninfected cell population is in a suitable state for infecting the virus, and the second determination unit makes a determination using a second model, which is a trained model obtained by deep learning a second neural network using the actual values of the second input data and the actual values of the second output data as second teacher data, and the first determination unit makes a determination after the second determination unit has made a determination.

(3) In the protein recovery determination system described in (1) or (2) above, the cells of the cell population are adherent cells.

(4) In the protein recovery determination system described in (3) above, the adherent cells are insect cells.

(5) In the protein recovery determination system described in (4) above, the insect cells are one of Sf (Spodoptera frugiperda) 9 cells, Sf21 cells, Tni (Trichoplusia ni) cells, and High Five cells.

(6) The protein recovery determination system described in (1) or (2) above further includes a first learning unit that trains the first neural network.

(7) The protein recovery determination system described in (2) above further includes a second learning unit that trains the second neural network.

(8) In the protein recovery determination system described in (2) above, the acquisition unit acquires information indicating whether the cell population is in a virus-uninfected state or a virus-infected state, and selects one of the second determination unit and the first determination unit to be used based on the information acquired by the acquisition unit.

(9) A recording medium according to another aspect of the present invention records a trained model obtained by deep learning a first neural network using the first input data and the first output data as first training data, the first input data being first image data obtained by photographing the culture of a virus-infected cell population in a time series, and the first output data being data determining whether the virus-infected cell population is in a suitable state for recovering a protein.

(10) In another aspect of the present invention, the recording medium records a trained model obtained by deep learning a second neural network using the second input data and the second output data as second training data, the second input data being second image data obtained by photographing the expansion culture of a virus-uninfected cell population in a time series, and the second output data being data determining whether the virus-uninfected cell population is in a suitable state for infecting the virus.

(11) A method for determining protein recovery according to another aspect of the present invention includes photographing a culture of a virus-infected cell population in a time series to obtain first image data, inputting the first image data as first input data to a first determination unit, and determining whether the virus-infected cell population is in a suitable state for recovering a protein as first output data from the first determination unit, the first determination unit making a determination using a first model, and the first model being a trained model obtained by deep learning a first neural network using actual values of the first input data and actual values of the first output data as first teacher data.

(12) The protein recovery determination method described in (11) above includes, before determining whether the cell population is in a suitable state for recovering a protein, acquiring second image data obtained by photographing an expansion culture of a virus-uninfected cell population in a time series, inputting the second image data as second input data to a second determination unit, and determining whether the virus-uninfected cell population is in a suitable state for infecting the virus as second output data from the second determination unit, the second determination unit making the determination using a second model, and the second model being a trained model obtained by deep learning a second neural network using actual values of the second input data and actual values of the second output data as second teacher data.

The above (1) to (12) allow the timing for recovering the protein to be appropriately determined.

FIG. 2 is a diagram illustrating an example of a functional configuration of a protein recovery determination system according to an embodiment. FIG. 13 is a diagram showing examples of images (taken by bright-field observation) taken by the imaging device at predetermined time intervals. FIG. 1 is an explanatory diagram of an example of a protein production process by cell engineering. 11 is an example of creating a first model by a first learning unit. FIG. 11 is a diagram illustrating an example of a determination made by a first determination unit; The host cell is a High Five cell, and the image was taken (by bright-field observation) at the early stage of viral infection (for example, the period from time t14 to t15 in Figure 3). This is an image (taken using bright-field observation) taken when the host cells are High Five cells and in a state suitable for recovery (for example, the period between times t16 and t17 in Figure 3). The host cells are Sf9 cells, and the images are example images (taken by bright-field observation) taken at the early stage of viral infection (for example, the period from time t14 to t15 in FIG. 3). This is an example of an image (taken by bright-field observation) taken when the host cells are Sf9 cells and in a state suitable for recovery (for example, the period from time t16 to t17 in FIG. 3). 1 is a flowchart of a processing procedure of the protein recovery determination system. FIG. 2 is a diagram showing an example of a functional configuration of a protein recovery determination system according to an embodiment, when the system has a protein recovery determination device. 13 is an example of creating a second model by a second learning unit. 11 is a diagram showing a determination example of a second determination circuit according to the embodiment; FIG. The host cells are High Five cells, and this is an example of an image (taken using bright-field observation) taken during the infection period (for example, the period from time t13 to t14 in Figure 4). The host cells are Sf9 cells, and the images are example images (taken by bright-field observation) taken during the infection period (for example, the period from time t13 to t14 in FIG. 4). 13 is a flowchart of a processing procedure performed by a protein recovery determination system according to a second embodiment. FIG. 1 is a diagram for explaining an antigen production process according to a conventional method. FIG. 1 shows the average correct answer rate and the average antibody yield in the determinations using the first model and the second model.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the drawings used in the following description, the scale of each component is appropriately changed so that each component can be recognized.
In addition, in all the drawings for explaining the embodiments, the same reference numerals are used for the parts having the same functions, and the repeated explanation is omitted.
In addition, "based on XX" in this specification means "based on at least XX" and includes cases where it is based on other elements in addition to XX. Furthermore, "based on XX" is not limited to cases where XX is used directly, but also includes cases where it is based on XX that has been calculated or processed. "XX" is any element (for example, any information).
Additionally, the term "host cell" herein is synonymous with "cells of a cell population."
In addition, in this specification, "image data" refers to a two-dimensional image.
Additionally, in this specification, "model" is synonymous with "trained model."

Fig. 1 is a diagram showing an example of the configuration of a protein recovery determination system 5 in this embodiment. As shown in Fig. 1, the protein recovery determination system 5 includes, for example, a protein recovery determination device 1, a photographing device 2, and an external device 3. The protein recovery determination device 1, the photographing device 2, and the external device 3 are connected by wired or wireless lines.
The protein recovery determination device 1 includes, for example, an acquisition unit 101, a first determination unit 102, an optional first learning unit 106, an optional storage unit 108, and an output unit 109. The first determination unit 102 executes a first model 104. The first determination unit 102 may have the first model 104. The protein recovery determination device 1 is configured by hardware using an information processing device such as a PC having a circuit such as an IC. The protein recovery determination device 1 is realized by reading a program that realizes a model having a specific function stored in the storage unit 108, etc., and executing the program in each unit such as the acquisition unit 101, the first determination unit 102, and the first learning unit 106.

The acquisition unit 101, the first judgment unit 102, the first learning unit 106, and the output unit 109 may be processors such as a CPU, MPU, SoC, and dedicated circuits that execute programs and perform various controls, as well as ICs having a processor. The memory unit 108 may be storage such as a RAM, ROM, HDD, and SSD. The memory unit 108 may be an external memory of the protein recovery judgment system 5.

The photographing device 2 photographs the virus-infected cell population in the culture vessel at predetermined times. The photographing device 2 and the protein recovery determination device 1 are connected to each other by wired or wireless lines. The photographed image is also added with the photographing time. The photographed image may be associated with identification information for identifying the subject being photographed. The photographing device 2 includes a photographing means 21 and a culture vessel 22. An example of the photographing device 2 is a device described in JP 2020-156419 A. An example of the photographing means 21 is a digital camera having an imaging element such as a CCD or CMOS, an imaging lens, etc.

The inner surface of the culture vessel 22 may be of any shape, but is preferably flat, in which case the cell population can be cultured adherently on the flat inner surface. The inner surface of the culture vessel 22 can also be coated with a surface suitable for adherent culture (e.g., coated with an extracellular matrix that serves as a scaffold for the cells).

The image capturing device 2 is preferably fixed in position relative to the inner surface of the culture vessel 22. If the inner surface of the culture vessel 22 is flat, the image capturing direction of the cell population in the culture vessel 22 by the image capturing device 2 is preferably approximately perpendicular to the flat surface.

The acquisition unit 101 acquires the image (image data a1) captured by the imaging device 2 at predetermined time intervals (e.g., once every 0.5 to 10 hours). The acquisition unit 101 may store the acquired image in the storage unit 108. The images acquired by the acquisition unit 101 can be used as first teacher data and second teacher data when creating a first model 104 and a second model 105 (described below).

The first judgment unit 102 executes the first model 104, inputs the first image data (image data a11) to the first model 104 (trained model a12) acquired from the memory unit 108, and judges whether the virus-infected cell population, which is the first output data, is in a suitable state for recovering proteins.

The first image data preferably includes 10% or more, 20% or more, 30% or more, 40% or more, or 50% or more of the culture surface of the cell population, with the upper limit being 60% or less, 70% or less, 80% or less, 90% or less, or 100% or less. In order to include the number of cells in the above-mentioned cell population in the first image data, the first image data obtained at the same time can be divided into multiple pieces. In order to divide into multiple pieces, the number of images taken at the same time may be multiple (e.g., 2 to 100).

Examples of image data include image data taken by bright-field observation, dark-field observation, phase contrast observation, differential interference contrast observation, polarized light observation, relief contrast observation, and fluorescence observation, as well as MIX observation. Of these image data, image data taken by bright-field observation, dark-field observation, phase contrast observation, differential interference contrast observation, and relief contrast observation are preferred, because it is preferable not to stain the host cells, the external shape of the host cells changes significantly due to viral infection of the host cells, and this is easily extracted as an inherent feature of the image data in deep learning, and there are also colorless and transparent host cells.

The first model 104 is a model that determines a state suitable for protein recovery from image data of a cell population after virus infection in a culture vessel by performing deep learning using a neural network (first neural network) in the first learning unit 106. Since the first input data is unstructured data, the accuracy rate of the first model can be improved by deep learning. The first model 104 is created by the first learning unit 106 through prior learning. Examples of learning methods using deep learning include DNN (Deep Neural Network), RNN (Recurrent Neural Network), CNN (Convolutional Neural Network), LSTM (Long Short Term Memory), and GAN (Generative Adversarial Networks), as well as combinations of two or more of these.

Host cells often undergo local morphological changes over time due to viral infection, and images of host cells contain limited areas (local receptive fields). In addition, the first input data is image data of a cell population containing multiple host cells. Therefore, as a learning method using deep learning, CNN, which is a learning method that can share weights for local receptive fields, is preferable.

The first model 104 may be a model obtained by learning in the first learning unit 106 of the protein recovery determination device 1, or may be a model obtained by learning in another device similar to the protein recovery determination device 1.

The first model 104 can be re-learned by the first learning unit 106 using teacher data in which the first image data (image data a14) acquired by the acquisition unit 101 is used as input data, and data in which an evaluation is added to the judgment of the first judgment unit 102 is used as output data, and can be updated by a first update unit (not shown) included in the first learning unit 106. Re-learning can be performed every time the acquisition unit 101 acquires the first image data, or it can be recorded in the storage unit 108 after the acquisition unit 101 acquires the first image data and performed as needed (for example, once every January to June).

The storage unit 108 stores, for example, programs used by the protein recovery determination device 1 and the protein recovery determination device 1A described below for various processes, thresholds, feature values, image data acquired by the acquisition unit 101, and the trained model a15 acquired by training in the first training unit 106.

The output unit 109 outputs information indicating the result of the judgment made by the first judgment unit 102 (data a13 indicating the result of the judgment) to the external device 3 as data a2 indicating the result of the judgment. The output unit 109 may also output image data determined to be in a state suitable for protein recovery, an estimated amount of recovered protein, or an estimated amount of host cell protein (HCP) to the external device 3.

The external device 3 presents the information output by the protein recovery determination device 1. The external device 3 is, for example, an image display device, a printing device, a tablet terminal, a smartphone, a personal computer, etc. The external device 3 and the protein recovery determination device 1 are connected to each other via a wired or wireless line.

FIG. 2 is a diagram showing an example of images (taken by bright field observation) taken by the image taking device 2 at each predetermined time. The host cell in FIG. 2 is High Five. Image g11 is an image taken at time t11 in FIG. 3. Image g12 is an image taken at a time between time t12 and time t13. Image g13 is an image taken at a time between time t12 and time t13. Image g14 is an image taken at time t14. Image g15 is an image taken at a time between time t15 and time t16. Image g16 is an image taken at a time between time t16 and time t17.
Note that the host cells and the photographed images shown in FIG. 2 are merely examples, and the photographed images, host cells, etc. are not limited to these.

Next, we will explain the host cells in the protein expression system and the infectious virus used to infect the host cells. Examples of host cells include bacterial cells, yeast cells, fungal cells, insect cells, and mammalian cells. The infectious virus can be selected appropriately depending on the host cell. For example, when the host cell is an insect cell, it is preferable to select a nuclear polyhedrosis virus.

The host cells are preferably adherent cells, and more preferably insect cells. Adherent culture using adherent cells allows the morphology of the host cells to be fixed before and after viral infection, making it possible to make accurate judgments. In addition, the morphology of the virus-infected host cells does not become complicated, as occurs with image data of suspension cultures, resulting in excellent learning efficiency when creating a trained model.

When the host cells are insect cells, many of them are suitable for adherent culture, and the host cells undergo large morphological changes before and after virus infection. For this reason, when the host cells are insect cells, deep learning is excellent for noise removal and feature selection.

Examples of insect cells include Sf9 cells, Sf21 cells, Tni cells, and High Five cells. Sf9 cells are a cell line derived from the armyworm moth (Spodoptera frugiperda). Tni cells are a cell line derived from the nettle moth (Trichoplusia ni).

3 is an explanatory diagram of an example of a protein production process by cell engineering, in which the horizontal axis represents time.
Time t11 is the time when the proliferation of cells (e.g., High Five cells) that are to be infected with the virus in the host cells begins. The period from time t12 to t13 is the period during which the host cells are expanded. The period from time t13 to t14 is the period during which the confluence rate increases and the host cells are in a suitable state for infecting the virus.
Time t14 is the start of viral infection. The period from time t14 onwards is the post-infection period. The period from time t15 to t16 is the period during which viral infection of the host cell progresses. The period from time t16 to t17 is a period suitable for recovering the protein.

FIG. 4 shows an example of the first model 104 created by the first learning unit 106. In FIG.
The first learning unit 106 inputs the first training data (actual values) into a first neural network to create a first model 104 that judges whether or not a state is suitable for recovery. The first learning unit 106 preferably performs learning for each pair of a host cell and an infectious virus. An example of a pair of a host cell and a virus is a pair of a host cell being an Sf9 cell and a virus being a nuclear polyhedrosis virus. The training data for whether or not a state is suitable for recovery uses the results of a judgment made by an experienced person.

The first model 104 may be a different model for each host cell. In this case, even if the infecting virus is different, the same model can be used as long as the host cell is the same.

FIG. 5 is a diagram showing an example of a judgment made by the first judgment unit 102. The first judgment unit 102 executes the first model 104, inputs the captured image to the first model 104 at each predetermined time, and judges whether or not the state is suitable for recovery. If the first model 104 has a model for each host cell, the first judgment unit 102 may select the first model 104 based on information indicating the host cell when making the judgment. The information indicating the host cell may be input by an operator operating the external device 3, for example, and information indicating the host cell may be added to the first image data acquired by the acquisition unit 101 as identification information for identifying the captured subject.

Next, examples of images taken before viral infection and at the time of recovery are shown.
Fig. 6 is an image (taken by bright-field observation) of a host cell that is a High Five cell at the early stage of viral infection (e.g., the period from time t14 to t15 in Fig. 3), and Fig. 7 is an image (taken by bright-field observation) of a host cell that is a High Five cell in a state suitable for recovery (e.g., the period from time t16 to t17 in Fig. 3).
As shown in Figures 6 and 7, when the host cell is High Five, in the early stage of viral infection, the virus extends its tail like a tadpole, and as the viral infection progresses, it shrinks into a circular shape and its outline becomes darker.

Fig. 8 shows an example of an image (taken by bright-field observation) of an Sf9 host cell at the early stage of viral infection (e.g., the period from time t14 to t15 in Fig. 3), while Fig. 9 shows an example of an image (taken by bright-field observation) of an Sf9 host cell in a state suitable for recovery (e.g., the period from time t16 to t17 in Fig. 3).
As shown in Figures 8 and 9, when the host cell is an Sf9 cell, the virus is spherical at the early stage of infection, and as the infection progresses, the outline becomes darker (the virus detaches from the adhesion surface), and as the virus infection progresses, the outline becomes darker still further.

Figure 10 is a flowchart of the processing procedure of the protein recovery determination system.

(Step S1) The imaging device 2 captures images of the virus-infected cell population at predetermined times.

(Step S2) The acquisition unit 101 acquires images captured by the image capture device 2 at predetermined times.

(Step S3) The first judgment unit 102 inputs first image data obtained by photographing the virus-infected cell population in a time series into the trained first model 104, and judges whether the virus-infected cell population is in a suitable state for recovering proteins.

(Step S4) If the first determination unit 102 determines that the state is suitable for collection (Step S4; YES), the process proceeds to Step S5. If the first determination unit 102 determines that it is not time to collect the waste (Step S4; NO), the process returns to Step S1.

(Step S5) The output unit 109 outputs information indicating that the state determined by the first determination unit 102 is suitable for recovering the protein to the external device 3.

As a result, according to this embodiment, it is possible to determine whether or not the state is suitable for recovering the protein. Furthermore, according to this embodiment, the protein can be recovered at the appropriate time (at a stage when there is little cell death and sufficient protein is produced by the cells), so the amount of protein produced by cell engineering can be increased.

The first model 104 uses image data captured in a time series, eliminating fluctuations in time caused by the worker and fluctuations caused by the worker's photography skills (such as light intensity and focus). This makes it possible to identify features using deep learning, improving the accuracy rate of the output data.

In addition, the first model 104 uses time-series image data, so even minute changes that the worker would not notice can be identified as features through deep learning, improving the accuracy rate of the output data.

FIG. 11 is a diagram showing a configuration example of a protein recovery determination system according to this embodiment having a protein recovery determination device 1A.
The protein recovery determination apparatus 1A further includes, for example, a second determination unit 103, an optional second learning unit 107, and an optional selection unit 110 in addition to the configuration of the protein recovery determination apparatus 1. The second determination unit 103 executes a second model 105. The second determination unit 103 may have the second model 105.

The second determination unit 103, the second learning unit 107, and the selection unit 110 may be processors such as a CPU, MPU, SoC, and dedicated circuits that execute programs and perform various controls, as well as ICs having a processor.

The imaging device 2 photographs the host cell culture in the culture vessel before it is infected with the virus at predetermined time intervals (e.g., once every 0.5 to 10 hours).

The second judgment unit 103 inputs second image data, which is second input data obtained by photographing a virus-uninfected cell population in time series during expansion culture, into the second model 105 (trained model a22) acquired from the memory unit 108, and judges whether the virus-uninfected cell population, which is second output data, is in a suitable state for infecting the virus.

The second image data preferably includes 10% or more, 20% or more, 30% or more, 40% or more, or 50% or more of the culture surface of the cell population, with the upper limit being 100%. In order to include the number of cells in the above-mentioned cell population in the second image data, the second image data acquired at the same time can be divided into multiple pieces. In order to divide it into multiple pieces, the number of images taken at the same time may be multiple (e.g., 2 to 100).

Examples of image data include image data taken by bright-field observation, dark-field observation, phase contrast observation, differential interference contrast observation, polarized light observation, relief contrast observation, and fluorescence observation, as well as MIX observation. Of these image data, image data taken by bright-field observation, dark-field observation, phase contrast observation, differential interference contrast observation, and relief contrast observation are preferred, because it is preferable not to stain the host cells, the proliferation of host cells causes large changes in their external shape due to the formation of colonies, which makes them easy to extract as features inherent in the image data in deep learning, and there are also colorless and transparent host cells.

The second model 105 is a model constructed by deep learning using a neural network (second neural network) that determines the time of viral infection from an image of a cell population being expanded and cultured in a culture vessel. Because the second input data is unstructured data, the accuracy rate of the second model can be improved by deep learning. Examples of learning methods for deep learning include DNN, RNN, CNN, LSTM, and GAN, as well as combinations of two or more of these.

Since the proliferation of host cells progresses through colony formation, and the morphology of the colony changes over time, the image of the host cell colony contains a limited area (local receptive field). In addition, the second input data is image data containing multiple colonies. Therefore, as a learning method using deep learning, CNN, which is a learning method that can share weights for local receptive fields, is preferable.

The second model 105 may be a model obtained by learning using a second learning unit 107 included in the protein recovery determination device 1A, or a model obtained by learning using another device similar to the protein recovery determination device 1.

The second learning unit 107 uses the second image data (image data a24) acquired by the acquisition unit 101 as input data and data with an evaluation added to the judgment of the second judgment unit 103 as output data to perform re-learning using teacher data, and can update the second model 105 by a second update unit (not shown) included in the second learning unit 107. Re-learning can be performed every time the acquisition unit 101 acquires the second image data, or can be recorded in the storage unit 108 after the acquisition unit 101 acquires the second image data and performed as needed (for example, once every 1-6 months). The second learning unit 107 stores the generated trained model a25 in the storage unit 108.

The output unit 109 outputs information indicating the result of the judgment made by the first judgment unit 102 (data a13 indicating the result of the judgment) as well as information indicating the result of the judgment made by the second judgment unit 103 (data a23 indicating the result of the judgment) to the external device 3 as data a2 indicating the result of the judgment. The output unit 109 may also output the captured image and the confluence rate that have been determined to be in a suitable state for infection to the external device 3.

The selection unit 110 determines whether the image data a3 acquired by the acquisition unit 101 is the first image data or the second image data, and selects whether to use the first determination unit 102 to determine whether the state is suitable for protein recovery, or the second determination unit 103 to determine whether the state is suitable for infection. If the selection unit 110 selects to make a determination using the first determination unit 102, it outputs the image data a3 acquired by the acquisition unit 101 to the first determination unit 102 as image data a11. On the other hand, if the selection unit 110 selects to make a determination using the second determination unit 103, it outputs the image data a3 acquired by the acquisition unit 101 to the second determination unit 103 as image data a21.

Next, an example of creating the second model 105 by the second learning unit 107 will be described.
Fig. 12 is an example of the creation of a second model by the second learning unit 107. As shown in Fig. 12, the second learning unit 107 creates a second model 105 by inputting second teacher data (actual values) to the second learning unit 107. The second learning unit 107 can perform learning for each pair of a host cell and an infectious virus. The determination of whether or not a state is suitable for infection is the result of a judgment made by an experienced person.

The second model 105 may include multiple models created for each host cell. In this case, even if the infecting virus is different, the same model may be used as long as the host cell is the same. For example, the host cell of the second model 105-1 may be an Sf9 cell, the host cell of the second model 105-2 may be a High Five cell, the host cell of the second model 105-3 may be an Sf21 cell, and the host cell of the second model 105-4 may be a Tni cell.

FIG. 13 is a diagram showing an example of a judgment made by the second judgment unit 103 according to this embodiment. The second judgment unit 103 executes the second model, inputs the captured image to the second model 105 at each predetermined time, and judges whether or not the state is suitable for infection. If the second model 105 has a model for each host cell, the second judgment unit 103 may select the second model 105 based on information indicating the host cell when making the judgment. Note that the information indicating the host cell may be input, for example, by an operator operating the external device 3, and information indicating the host cell may be added to the second image data acquired by the acquisition unit 101 as identification information for identifying the captured subject.

Next, examples of images captured during the infection period will be described.
FIG. 14 shows an example of an image (taken by bright-field observation) taken during the infection period (for example, the period from time t13 to t14 in FIG. 4) when the host cells are High Five cells.

Figure 15 shows an example of an image (taken using bright-field observation) taken during the infection period (e.g., the period from time t13 to t14 in Figure 4) when the host cell was Sf9 cell.

Next, an example of a processing procedure performed by the protein recovery determination system 5A will be described.
FIG. 16 is a flowchart of the processing procedure performed by the protein recovery determination system according to this embodiment.

(Step S21) The imaging device 2 captures images of the culture of the virus-uninfected cell population at predetermined times.

(Step S22) The acquisition unit 101 acquires images captured by the image capture device 2 at predetermined times.

(Step S23) The second judgment unit 103 inputs second image data obtained by photographing the culture of a cell population not infected with a virus in a time series into the trained second model 105, and judges the time to collect the cells based on information indicating whether the cells are in a suitable state for infecting the virus.

(Step S24) If the second judgment unit 103 judges that it is the time of infection (Step S24; YES), the process proceeds to step S25. If the second judgment unit 103 judges that it is not the time of infection (Step S24; NO), the process returns to step S21.

(Step S25) The output unit 109 outputs information indicating that it is a suitable time to infect the virus determined by the second determination unit 103 to the external device 3. Next, after the virus is introduced into the culture vessel, the processing procedure shown in the flowchart of FIG. 10 is executed.

In the protein recovery determination device 1A, the second determination unit 103 performs a determination, and then the first determination unit 102 performs a determination. That is, the second determination unit 103 and the first determination unit 102 perform their determinations in this order.

If the timing of viral infection is appropriate, the accuracy rate of the first learning model obtained by learning the first neural network can be improved. If there are many gaps between cells, there is a problem that not only is the number of cells small, but also the amount of protein recovered is reduced due to cell death caused by weak intercellular response. If there are no gaps between cells, there is a problem that the amount of protein recovered is reduced due to cell death caused by nutrient deficiency caused by insufficient absorption of nutrients from the culture medium. According to this embodiment, the virus can be infected at an appropriate time when the cells can withstand viral infection, so it is possible to obtain the effect of solving these problems.

Since the operator knows the timing of the virus infection, for example, the operator may operate the external device 3 to input or select to determine the time of infection, and the acquisition unit 101 may acquire the input or selection result. Note that the virus infection may be performed automatically by another external device 3 based on the result output by the protein recovery determination device 1A. In this case, the protein recovery determination device 1A can determine and acquire whether it is before or after infection. For example, the acquisition unit 101 may acquire information indicating whether it is in a virus-free state or a virus-infected state, and the selection unit 110 may select whether to determine the protein recovery time by the first judgment unit 102 or to determine the virus infection time by the second judgment unit 103 based on the acquired result. The selection circuit may select based on a learning model that distinguishes information indicating whether it is in a virus-free state or a virus-infected state.

In this way, the protein recovery determination device 1A can obtain whether to determine the time of viral infection or the time of protein recovery, and select the determination unit, model, and learning unit based on the obtained result.

The protein

recovery determination systems

5 and 5A are used in the production of proteins using cells (cell engineering).

　A program for implementing all or part of the functions of the protein recovery determination device 1 (or 1A) of the present invention may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed to perform all or part of the processes performed by the protein recovery determination device 1 (or 1A). Note that the term "computer system" here includes hardware such as an OS and peripheral devices. The term "computer system" also includes a WWW system equipped with a homepage providing environment (or display environment). The term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. The term "computer-readable recording medium" also includes those that hold a program for a certain period of time, such as volatile memory (RAM) inside a computer system that becomes a server or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line.

The above program may also be transmitted from a computer system in which the program is stored in a storage device or the like to another computer system via a transmission medium, or by transmission waves in the transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has the function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The above program may also be one that realizes part of the above-mentioned functions. Furthermore, it may be one that can realize the above-mentioned functions in combination with a program already recorded in the computer system, a so-called difference file (difference program).

Figure 18 shows the average accuracy rate and average antibody yield of the judgments made by the first and second models. Figure 18 shows the results of judging whether a virus-infected cell population is in a suitable state for recovering proteins (first judgment) and whether a virus-uninfected cell population is in a suitable state for infecting the virus (second judgment) for each of three images (n=3) taken by the image capture device 2. For comparison, Figure 18 also shows the results of the first and second judgments made by person A (a highly skilled person with over 10 years of experience) and person B (a less skilled person with less than 1 year of experience). σ indicates the standard deviation of the antibody yield.

As shown in FIG. 18, in the results for the combination of the cell tumor "Sf9" and the virus species "Tif1-r", when both the first and second judgments were performed by person A, the average correct answer rate was "100%," the average antibody yield was "15.4", and the standard deviation σ was "2.16". On the other hand, when the first and second judgments were performed using the first and second models, the average correct answer rate was "93%,", the average antibody yield was "14.1", and the standard deviation σ was "2.54". From this, it was confirmed that even when the first and second models were used, highly accurate judgments could be made, and thus a high antibody yield could be achieved. Similarly, it was confirmed that highly accurate judgments could be made, and thus a high antibody yield could be achieved, for the combination of the cell tumor "Sf9" and the virus species "MD5", and for the combination of the cell tumor "HF" and the virus species "Mi-2".

　Although the above describes the form for carrying out the present invention using an embodiment, the present invention is in no way limited to such an embodiment, and various modifications and substitutions can be made without departing from the spirit of the present invention.

5, 5A...Protein recovery determination system, 1, 1A...Protein recovery determination device, 2...Photographing device, 3...External device, 101...Acquisition unit, 102...First determination unit, 103...Second determination unit, 104, 104-1, 104-2, 104-3, 104-4...First model, 105, 105-1, 105-2, 105-3, 105-4...Second model, 106...First learning unit, 107...Second learning unit, 108...Storage unit, 109...Output unit, 110...Selection unit, 21...Photographing means, 22...Culture vessel, a1, a3...Image data, a2...Data showing the result of the determination, a11, a21...Image data, a12, a22...Trained model, a13, a23...Data showing the result of the determination, a14, a24...Image data, a15, a25...Trained model

Claims

a first determination unit that receives as input data first image data obtained by photographing the culture of a virus-infected cell population in a time series, and determines as first output data whether the virus-infected cell population is in a suitable state for recovering a protein;
An acquisition unit that acquires image data by photographing the inside of the culture vessel in a time series,
The first determination unit performs determination using a first model,
The first model is a trained model obtained by deep learning a first neural network using the actual value of the first input data and the actual value of the first output data as first teacher data.
Protein recovery determination system.
a second determination unit that receives as second input data second image data obtained by photographing the expansion culture of a virus-uninfected cell population in a time series, and determines as second output data whether the virus-uninfected cell population is in a suitable state for infecting the virus,
The second determination unit performs determination using a second model,
the second model is a trained model obtained by deep learning a second neural network using actual values of the second input data and actual values of the second output data as second teacher data;
The first determination unit makes a determination after the second determination unit makes a determination.
The protein recovery determination system according to claim 1 .
The protein recovery determination system according to claim 1 or 2, wherein the cells of the cell population are adherent cells.
The protein recovery determination system according to claim 3, wherein the adherent cells are insect cells.
The protein recovery determination system according to claim 4, wherein the insect cells are one of Sf (Spodoptera frugiperda) 9 cells, Sf21 cells, Tni (Trichoplusia ni) cells, and High Five cells.
The protein recovery determination system according to claim 1 or 2, further comprising a first learning unit that trains the first neural network.
The protein recovery determination system according to claim 2, further comprising a second learning unit that trains the second neural network.
the acquiring unit acquires information indicating whether the cell population is in a virus-uninfected state or a virus-infected state,
The protein recovery determination system according to claim 2 , further comprising: a determination unit for use selected from the first determination unit and the second determination unit based on the information acquired by the acquisition unit.
the first input data is first image data obtained by photographing a culture of a virus-infected cell population in a time series;
The first output data is data obtained by determining whether the virus-infected cell population is in a state suitable for recovering a protein;
A recording medium on which a trained model obtained by deep learning of a first neural network using the first input data and the first output data as first teacher data is recorded.
The second input data is second image data obtained by photographing an expansion culture of a virus-uninfected cell population in a time series,
The second output data is data obtained by determining whether the state of the virus-uninfected cell population is suitable for infecting the virus,
A recording medium on which a trained model obtained by deep learning a second neural network using the second input data and the second output data as second teacher data is recorded.
acquiring first image data by photographing a culture of the virus-infected cell population over time;
inputting the first image data as first input data to a first determination unit;
determining whether the virus-infected cell population is in a state suitable for recovering a protein as first output data from the first determination unit;
The first determination unit performs determination using a first model,
the first model is a trained model obtained by deep learning of a first neural network using actual values of the first input data and actual values of the first output data as first teacher data.
Prior to said determination of whether said cell population is suitable for recovering a protein,
acquiring second image data obtained by photographing the expansion culture of the virus-uninfected cell population in a time series;
inputting the second image data as second input data to a second determination unit;
determining whether the state is suitable for infecting the virus-uninfected cell population with a virus as second output data from the second determination unit;
The second determination unit performs determination using a second model,
The protein recovery determination method according to claim 11, wherein the second model is a trained model obtained by deep learning a second neural network using an actual value of the second input data and an actual value of the second output data as second teacher data.