Apparatus for inferring cell state, apparatus for inferring cell phenotype, apparatus for inferring cell's differentiation state, method for inferring cell state, method for inferring cell phenotype, and method for inferring cell's differentiation state

Abstract

To provide an apparatus and a method for inferring changes in a cell state, such as polarization that changes cell phenotype, differentiation that changes cell type, cell transformation, and cell death.SOLUTION: Provided an apparatus for inferring a cell state that infers the state of a cell based on parameters indicating changes over time in the morphology or position of a cell whose state is unknown, comprising: an acquisition unit that acquires parameters indicating changes over time in the morphology or position of a cell whose state is unknown; an inference unit that infers the state of a cell about which the acquisition unit acquired information based on the results of learning the association between the parameters indicating changes over time in the morphology or position of a cell whose state is known and the state of the cell; and an output unit that outputs the state of the cell inferred by the inference unit.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a device for inferring a cell state, a device for inferring a cell phenotype, a device for inferring a cell differentiation state, a method for inferring a cell ... phenotype, and a method for inferring a cell differentiation state. [Background technology]

[0002] Some cells undergo state changes. Changes in state include polarization (change in phenotype), differentiation (change in cell type), epithelial-mesenchymal transition (epithelial cells acquire mesenchymal traits and become cancerous), and cell death (cell death). As an example of polarization, macrophages can reversibly polarize from non-polarized M0 macrophages to two types, M1 and M2. The state of cells is an important indicator in evaluating the compatibility of biomaterials.

[0003] Non-Patent Document 1 describes the determination of macrophage phenotype by immunostaining.

[0004] Non-Patent Document 2 describes that in neuroinflammatory diseases, M1 macrophages damage nerve cells, whereas M2 macrophages regenerate and repair nerve cells. It also describes that there is a difference in the intracellular distribution of actin filaments stained with phalloidin, that M0 macrophages and M2 macrophages can form filopodia, whereas M1 macrophages only have a spherical morphology, and that M2 macrophages migrate significantly longer distances than M1 macrophages.

[0005] Non-Patent Document 3 describes that a correlation between the polarization state of macrophages and changes in cell shape was confirmed by immunostaining and Western blotting. Specifically, it describes that induction of M1 polarization results in a flat, circular cell shape, whereas induction of M2 polarization results in cell elongation.

[0006] Non-Patent Document 4 describes a method for identifying the phenotype of macrophages by analyzing a composite image of macrophages in which f-actin has been stained with phalloidin and the nuclei have been stained with DAPI, and data showing the phenotype of the cells in the composite image, and by analyzing the area, direction, range, shape, intensity, etc. of each cell and performing machine learning to identify the phenotype.

[0007] US Patent No. 5,999,943 describes a machine learning method for classifying different subsets or types of cells in a mixed population of cells. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Special Publication No. 2022-546396 [Non-patent literature]

[0009] [Non-Patent Document 1] Bai et al., Oncotarget, “Contact-dependent carcinoma aggregate dispersion by M2a macrophages via ICAM-1 and β2 integrin interactions”, July 30, 2015, Vol.6, No.28, p.25295-25307 [Non-Patent Document 2] Vogel et al., “Macrophages migrate in an activation-dependent manner to chemokines involved in neuroinflammation”, Journal of Neuroinflammation, February 1, 2014, Vo.11, No.23 [Non-Patent Document 3] McWhorter et al., “Modulation of macrophage phenotype by cell shape”, PNAS, October 7, 2013, Vol.110, No.43, p.17253-17258 [Non-Patent Document 4] Rostam et al., “Image based Machine Learning for identification of macrophage subsets”, Scientific Reports, June 14, 2017, Vol. 7, Article number: 3521 Summary of the Invention [Problem to be solved by the invention]

[0010] However, methods for identifying phenotypes by immunostaining or the like have not been able to non-invasively and over time identify the phenotype of living cells whose phenotypes change.

[0011] An object of the present invention is to provide an apparatus and method for identifying the phenotype of a cell undergoing phenotypic change while the cell is alive, as well as for inferring changes in the state of a cell, such as polarization, which changes the phenotype, differentiation, which changes the type of cell, epithelial-mesenchymal transition, in which an epithelial cell acquires mesenchymal characteristics and becomes cancerous, and cell death, which causes a cell to die. [Means for solving the problem]

[0012] In order to solve the above problems, a first aspect of the present disclosure is a cell state inference device that infers a state of a cell, the state of which is unknown, based on a parameter indicating a time-dependent change in a morphology or position of the cell, the device comprising: an acquisition unit for acquiring parameters indicative of changes over time in morphology or position of a cell whose state is unknown; an inference unit that infers the state of the cell about which the acquisition unit acquired the information, based on a result of learning the association between a parameter indicating a time-dependent change in the morphology or position of the cell, the state of which is known, and the state of the cell; and An output unit that outputs the state of the cell inferred by the inference unit.

[0013] A second aspect of the present disclosure is a cell phenotype inference device that infers a phenotype of a cell, the phenotype of which is unknown, based on a parameter indicating a time-dependent change in a morphology or position of the cell, the device comprising: an acquisition unit for acquiring parameters indicative of changes in morphology or position of cells of unknown phenotype over time; an inference unit that infers the phenotype of the cell about which the acquisition unit acquired the information, based on a result of learning the association between a parameter indicating a time-dependent change in the morphology or position of a cell whose phenotype is known and the cell phenotype; and An output unit that outputs the cell phenotype inferred by the inference unit.

[0014] A third aspect of the present disclosure is an apparatus for inferring a differentiation state of a cell, the apparatus inferring the differentiation state of the cell based on a parameter indicating a time-dependent change in a morphology or a position of the cell, the differentiation state of which is unknown, the apparatus comprising: an acquisition unit for acquiring a parameter indicative of a change over time in the morphology or position of a cell whose differentiation state is unknown; an inference unit that infers the differentiation state of the cell about which the acquisition unit has acquired the information, based on a result of learning the association between a parameter indicating a time-dependent change in the morphology or position of a cell whose differentiation state is known and the cell's differentiation state; and An output unit outputs the differentiation state of the cell inferred by the inference unit.

[0015] A fourth aspect of the present disclosure is a method for inferring a state of a cell, comprising the steps of: inputting parameters indicating a change over time in a morphology or position of a cell whose state is known and the state of the cell into a machine learning model as learning data; and causing the machine learning model to learn an association between the parameters indicating a change over time in the morphology or position of a cell whose state is known and the state of the cell; obtaining a parameter indicative of a change in morphology or position of a cell, the state of which is unknown, over time; inputting parameters indicative of changes in morphology or position of the unknown cell over time into the trained machine learning model; and The method includes a step of having the trained machine learning model infer the state of a cell whose state is unknown.

[0016] A fifth aspect of the present disclosure is a method for inferring a phenotype of a cell, comprising: inputting parameters indicating a change over time in the morphology or position of a cell whose phenotype is known and the phenotype of the cell into an inference model as learning data, and causing the inference model to learn the association between the parameters indicating a change over time in the morphology or position of a cell whose phenotype is known and the phenotype of the cell; obtaining a parameter indicative of a change in morphology or location of a cell of unknown phenotype over time; inputting parameters representing changes in morphology or location of the cells with unknown phenotype over time into the trained inference model; and The method includes a step of having the trained inference model infer the phenotype of a cell whose phenotype is unknown.

[0017] A sixth aspect of the present disclosure is a method for inferring the differentiation state of a cell, comprising: inputting parameters indicating changes over time in the morphology or position of a cell whose differentiation state is known and the differentiation state of the cell into an inference model as learning data, and causing the inference model to learn the association between the parameters indicating changes over time in the morphology or position of a cell whose differentiation state is known and the differentiation state of the cell; obtaining a parameter indicative of the change over time in morphology or position of cells whose differentiation state is unknown; inputting parameters indicative of changes in morphology or position of the cells with unknown differentiation state over time into the trained inference model; and The method includes a step of having the trained inference model infer the differentiation state of a cell whose differentiation state is unknown. Effect of the Invention

[0018] According to the present invention, it is possible to distinguish between cellular states such as polarization, in which a cell's phenotype changes, differentiation, in which a cell type changes, epithelial-mesenchymal transition, in which an epithelial cell acquires mesenchymal traits and becomes cancerous, and cell death, in which a cell dies. [Brief description of the drawings]

[0019] [Figure 1] 1 is a schematic diagram showing a configuration of an inference device 1 according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a schematic diagram showing a holographic microscope, which is an example of an acquisition device 20. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] An inference device and an inference method according to an embodiment of the present invention will now be described. 1 is a schematic diagram showing the configuration of an inference device 1 according to an embodiment of the present invention. The inference device 1 includes a computer 10, an acquisition device 20, an output device 30, and the like.

[0021] The computer 10 comprises a processor 11, a main memory device 12, an auxiliary memory device 13, an input interface 14, an output interface 15, and the like. The processor 11 reads out a program stored in the auxiliary storage device 13 into the main storage device 12 and executes the program, thereby causing the computer 10 to perform various processes.

[0022] For example, as described below, the computer 10 acquires parameters indicating the morphology of the cell and parameters indicating the change in the position of the cell over time from the cell image data acquired by the acquisition device 20. That is, the computer 10 functions as an acquisition unit that acquires parameters indicating the change in the morphology or position of a cell with an unknown state over time in cooperation with the acquisition device 20.

[0023] Moreover, a software library for use in machine learning is installed in the auxiliary storage device 13 of the computer 10, and the processor 11 reads the software from the auxiliary storage device 13 into the main storage device 12 and executes it, thereby functioning as an inference model. A trained inference model (inference unit) can be obtained by inputting a data set, which will be described later, into the computer 10 and performing supervised learning. In other words, the computer 10 functions as an inference unit that infers the state of a cell about which the acquisition unit has acquired information, based on the results of learning the association between a parameter indicating a time-dependent change in the morphology or position of a cell whose state is known and the state of the cell.

[0024] Here, the "state" of a cell includes, for example, the phenotype, type, trait, and life or death of a cell. Changes in the state of a cell include polarization, which changes the phenotype of a cell, differentiation, which changes the type of a cell, transformation, and cell death, which causes a cell to die.

[0025] "Phenotype" refers to the characteristics of a cell that are manifested as traits that can be observed from the surface, such as shape, color, size, and function. A cell with an "unknown phenotype" is a cell whose phenotype has not been identified at the time of observation, such as a cell whose phenotype changes in a living state. Examples of "cells whose phenotype changes in a living state" include macrophages, microglia, Kupffer cells, and alveolar macrophages (dust cells).

[0026] "Differentiation" of a cell refers to the change of an immature cell into a more mature type of cell, such as the change of a hematopoietic stem cell through a progenitor cell into various mature blood cells, including macrophages.

[0027] Cell "transformation" refers to the change in genetic properties of cells by introducing genes into them, or malignant transformation. "Malignant transformation" refers to the transformation of normal cells into cancer, such as epithelial-mesenchymal transition, in which epithelial cells acquire mesenchymal characteristics. "Epithelial-mesenchymal transition" refers to the process in which epithelial cells lose their cell polarity and cell adhesion function with surrounding cells, and gain migration and invasive capabilities, becoming mesenchymal-like cells.

[0028] "Cell death" is the phenomenon in which a living cell ceases to function. There are two types of cell death: apoptosis and necrosis.

[0029] FIG. 2 is a schematic diagram showing a holographic microscope. The holographic microscope is an example of an acquisition device 20 that acquires an image of a cell. The holographic microscope is a microscope that acquires a three-dimensional image of a sample using the interference of light. The holographic microscope is an example of a microscope that can observe the three-dimensional morphology of a cell. As shown in FIG. 2, the holographic microscope includes a diode laser 21, half mirrors 22 and 23, mirrors 24 and 25, an objective lens 26, a magnifying glass 27, an image sensor 28, and a sample stage 29.

[0030] Light emitted from a diode laser 21 is split by a half mirror 22. The light (object illumination light) transmitted through the half mirror 22 is reflected by a mirror 24 and then irradiated onto a sample on a sample stage 29. The sample placed on the sample stage 29 is, for example, a culture solution of cells whose state is unknown. The light (object light) transmitted through the sample is magnified by the objective lens 26, reflected by the half mirror 23, and enters the image sensor .

[0031] On the other hand, the light (reference light) reflected by the half mirror 22 is reflected by the mirror 25, magnified by the magnifying mirror 27, passes through the half mirror 23, and enters the image sensor 28. The objective lens 26 and the magnifying mirror 27 have the same magnification. An interference pattern (hologram) is formed on the light receiving surface of the image sensor 28 by interference between the object light and the reference light. The hologram data is input from the image sensor 28 to the computer 10 via the input interface 14 and stored in the auxiliary storage device 13. The computer 10 reconstructs a phase image using the hologram data and analyzes the reconstructed phase image. The computer 10 may output the phase image from the output interface 15 to an output device 30, which will be described later.

[0032] The computer 10 uses the hologram data to calculate parameters indicative of the morphology or parameters indicative of changes in position over time of cells whose state is unknown.

[0033] Parameters indicating cell morphology include "Area," "Irregularity," "Optical volume," and "Optical thickness," which are described below.

[0034] “Area” is a parameter that indicates the area of ​​a cell. The area (A) of a cell is expressed by the following formula (1).

number

[0035] Here, N is the number of pixels in the image of the cell captured by the image sensor 28, and s 2 xy is the area of ​​the region imaged by one pixel of the image sensor 28.

[0036] “Irregularity” is a parameter that indicates how much the outer shape of a cell in planar view deviates from a perfect circle. “Irregularity” is expressed by the following equation (2).

number

[0037] Here, A is the area (A) of the cell in formula (1), and “perimeter length” is the perimeter of the cell. When the perimeter of the cell is a perfect circle, the irregularity is 0.

[0038] “Optical volume” is a parameter that indicates the volume of a cell. The volume of a cell (V) is calculated by the following formula (3).

number

[0039] where λ is the wavelength and n c is the refractive index of the cell, and n m is the refractive index of the medium in which the cells are cultured. φ sum is the sum of the phase shifts of the cells. φ sum is expressed by the following equation (4).

number

[0040] Here, N is the number of pixels in the image of the cell captured by the image sensor 28, and φ i is the phase shift in radians of the area imaged at pixel i (i = 1 to N) of the cell divided by 2π.

[0041] “Optical thickness” is a parameter that indicates the average thickness of a cell. avg ) is calculated using the following formula (5):

number

[0042] Parameters indicating changes in cell position over time include "migration distance," "motility distance," "migration directness," and "motility speed," which are described below.

[0043] "Migration distance" is the shortest distance between the start and end points of a cell's migration path. "Motility distance" is the distance that a cell actually moves from the start point to the end point of its migration path. When calculating "Motility distance" from multiple images taken at regular time intervals by time-lapse photography, the total distance that the cell moves between each image is the "Motility distance."

[0044] "Motility speed" is the "motility distance" divided by the time it took for the cell to move from the starting point to the end point.

[0045] As a holographic microscope used by the acquisition unit to acquire such parameters, for example, Holomonitor (registered trademark) manufactured by PHI AB can be used. In addition, as the acquisition device 20, a microscope capable of observing the three-dimensional morphology of other cells may be used. For example, instead of the holographic microscope, the above parameters may be acquired from an image of a cell captured using a confocal laser microscope. Alternatively, as the acquisition device 20, any of the above parameters may be acquired from an image of a cell captured using a phase contrast microscope or a differential interference microscope. With the phase contrast microscope or the differential interference microscope, "Area" and "Irregularity" can be acquired as parameters indicating the morphology of the cell, and "Migration distance", "Motility distance", and "Motility speed" can be acquired as parameters indicating the change in the position of the cell over time.

[0046] The parameters indicating the time-dependent changes in the morphology or position of the cells described above can be obtained from the cells whose states are known, and the data set can be used for machine learning (supervised learning). The data set can be obtained by imaging the cells whose states are known using the above-mentioned holographic microscope and performing calculations using the computer 10.

[0047] Cells with a known state include, for example, cells with a known phenotype, cells with a known differentiation state (type), and cells with known traits or live or dead states.

[0048] For example, parameters indicating the above-mentioned changes in the morphology or position of cells over time may be obtained from cells with known phenotypes, and the data set may be used for machine learning (supervised learning). The data set may be obtained by imaging the cells with known phenotypes using the above-mentioned acquisition device 20 and performing calculations using the computer 10.

[0049] In the following, a case where a cell with a known phenotype is used as an example of a "cell with a known state" will be described.

[0050] A cell with a known phenotype can be obtained, for example, by inducing a specific phenotype in a cell whose genetic phenotype changes in living state, using a known method. For example, macrophages can be induced to become "cells with a known phenotype" by becoming the pro-inflammatory M1 type or the anti-inflammatory M2 type.

[0051] M1 type macrophages can be induced by classical activation (M1 type activation), for example, with lipopolysaccharide (LPS) and interferon-γ (IFN-γ). M2 type macrophages can be induced, for example, by selective activation (M2 type activation) with interleukin-4 (IL-4) or interleukin-13 (IL-13).

[0052] Similarly, polarization can be induced in microglia, Kupffer cells, alveolar macrophages, and the like by known methods.

[0053] Alternatively, parameters indicating the above-mentioned changes in cell morphology or position over time may be obtained from various types of cells whose differentiation state is known, and the data set may be used for machine learning (supervised learning). The data set can be obtained by imaging cells whose differentiation state is known using the above-mentioned acquisition device 20 and performing calculations using the computer 10.

[0054] Alternatively, parameters indicating the above-mentioned changes in cell morphology or position over time may be obtained from cells whose traits are known, and the data set may be used for machine learning (supervised learning). For example, the above-mentioned parameters may be obtained from normal epithelial cells and cancerous epithelial cells whose traits are known, and the data set may be used for machine learning (supervised learning). The data set may be obtained by imaging cells whose traits are known using the above-mentioned acquisition device 20 and performing calculations using the computer 10.

[0055] Alternatively, parameters indicating the above-mentioned changes in cell morphology or position over time may be obtained from cells whose live or dead status is known, and the data set may be used for machine learning (supervised learning). The data set can be obtained by imaging the cells whose live or dead status is known using the above-mentioned acquisition device 20 and performing calculations using the computer 10.

[0056] For the above-mentioned "cells whose state is known," parameters that indicate changes in cell morphology or position over time can be obtained to obtain a data set.

[0057] Parameters indicating cell morphology include the above-mentioned "Area", "Irregularity", "Optical volume", and "Optical thickness". Parameters indicating changes in cell position over time include the above-mentioned "migration distance," "motility distance," and "motility speed." The method for calculating the parameters indicating the change in cell morphology or position over time is as described above.

[0058] A trained inference model can be obtained by performing supervised learning on an inference model such as a neural network using the obtained data set. For example, the above data set can be input to a computer 10 having a software library for machine learning installed therein, and a trained inference model can be obtained by performing supervised learning. Software libraries used for machine learning (machine learning libraries) include, for example, scikit-learn, TensorFlow, PyTorch, Keras, and LightGBM.

[0059] Note that data augmentation may be performed in advance on the dataset to be input to the inference model. For example, the mean and variance of the parameters of each class may be calculated, and a random value within the range of (mean - variance) to (mean + variance) may be added to the dataset.

[0060] Additionally, outliers in the data set may be pre-deleted. For example, values ​​above a certain threshold may be replaced with the average value for that class. Alternatively, all values ​​may be used as is without replacing outliers.

[0061] Inference models can be evaluated by cross-validation, in which a dataset used in supervised learning is divided into training data and test data and evaluated. For example, a data set may be split into training data and test data, and hold-out validation may be performed to conduct learning and evaluation. Alternatively, k-fold cross-validation may be performed by dividing the dataset into k parts, using k-1 of the parts as training data and the remaining part as test data, and repeating the learning and evaluation process k times.

[0062] As evaluation indices, accuracy, precision, recall, F measure, ROC-AUC, and PR-AUC can be used.

[0063] The accuracy is expressed by the following formula (6). The precision is expressed by the following formula (7). The recall is expressed by the following formula (8).

number

[0064] In equations (6) to (8), TP is the number of true positive data, TN is the number of true negative data, FP is the number of false positive data, and FN is the number of false negative data.

[0065] The F measure is the harmonic mean of precision and recall, and is expressed by the following equation (9).

number

[0066] For example, when evaluating a trained inference model using test data of M1 type macrophages (hereinafter also referred to as "M1 macrophages") and M2 type macrophages (hereinafter also referred to as "M2 macrophages"), the number of predictions of M1 macrophages (Positive) from the test data of M1 macrophages (True data) is defined as TP; the number of predictions of M2 macrophages (Negative) from the True data is defined as TN; the number of predictions of Positive from the test data of M2 macrophages (False data) is defined as FP; and the number of predictions of Negative from the False data is defined as FN. Then, the accuracy rate, precision, and recall can be calculated, and an F-score can be calculated from the calculated precision and recall.

[0067] ROC-AUC is the area under the ROC curve (Receiver Operating Characteristic curve). The ROC curve is a curve in which the false positive rate (FPR) is plotted on the horizontal axis and the true positive rate (TPR) is plotted on the vertical axis while changing the threshold of the parameter for determining positive or negative. The more effective the inference model is for classification, the further the ROC curve moves to the upper left from the 45° line, and the closer the ROC-AUC is to 1. On the other hand, the less effective the inference model is for classification, the closer the ROC curve is to the 45° line, and the closer the ROC-AUC is to 0.5.

[0068] PR-AUC is the area under the Precision-Recall Curve (AUC). The PR curve is a curve that plots precision (Precision) on the vertical axis and recall (Recall) on the horizontal axis while varying the threshold of the parameter for determining positive or negative. The more effective the inference model is for classification, the farther the PR curve moves to the upper right from the -45° line, and the closer the PR-AUC is to 1. On the other hand, the less effective the inference model is for classification, the closer the ROC curve is to the -45° line, and the closer the PR-AUC is to 0.5.

[0069] For the above "cells whose states are known," machine learning is performed on an inference model using the parameters of the above dataset as training data, and then the inference model is evaluated using test data.

[0070] When using a trained inference model to predict the state of a cell whose state is unknown, the state can be predicted with a high probability by using parameters that indicate changes in the cell's morphology or position over time and that have high evaluation indexes based on the above test data.

[0071] The output device 30 outputs the state of the cell inferred by the inference unit. As the output device 30, for example, an output device such as a display or a printer connected to the computer 10 can be used.

[0072] For example, when inferring the phenotype of a cell, the output device 30 outputs the phenotype of the cell inferred by the inference section. Furthermore, when the differentiation state of a cell is inferred, the output device 30 outputs the differentiation state of the cell (cell type) inferred by the inference section. Alternatively, in the case of inferring the phenotype of a cell, the phenotype of the cell inferred by the inference unit is output from the output device 30. For example, the determination result of whether or not an epithelial cell has become cancerous, or whether or not an epithelial cell has acquired mesenchymal traits is output. Alternatively, in the case of inferring the life or death of a cell, the output device 30 outputs the result of the determination of the life or death state of the cell inferred by the inference section.

[0073] Hereinafter, a method for inferring the phenotype of a macrophage will be described with a specific example. EXAMPLES

[0074] (Unpolarized cells) A mouse monocyte-macrophage-like cell line (JCRB Cell Bank, Culture Resources Laboratory, National Institutes of Biomedical Innovation, Health and Nutrition, cell number: JCRB9108, cell name: J774A.1) was prepared as an unpolarized macrophage (M0 macrophage).

[0075] (M1 polarization-induced cell population) Lipopolysaccharide (LPS) (Fujifilm Wako Pure Chemical Industries, Ltd., product number: 121-05161) and interferon-γ (IFN-γ) (Fujifilm Wako Pure Chemical Industries, Ltd., product number: 094-04701) were administered to the unpolarized cell group and incubated at 37°C, 5% CO2 for 24 hours to induce M1 type macrophages (M1 macrophages).

[0076] (M2 polarization induced cell group) Interleukin-4 (IL-4) (Fujifilm Wako Pure Chemical Industries, Ltd., product number: 094-03944) was administered to the unpolarized cell group and incubated at 37°C, 5% CO2 for 24 hours to induce M1 type macrophages (M1 macrophages).

[0077] (Polarization evaluation) We isolated some cultured cells from the above M0, M1, and M2 macrophage cell groups, and evaluated whether they were polarized to M1 or M2 types by RT-PCR and immunostaining. Interleukin-1β (IL-1β) and inducible nitric oxide synthase (iNOS) were used as M1 markers, and arginase 1 (ARG1) was used as an M2 marker.

[0078] [RT-PCR] By RT-PCR, it was evaluated whether or not mRNA encoding IL-1β, mRNA encoding iNOS, and mRNA encoding ARG1 were produced in the cells.

[0079] First, mRNA was extracted from each cultured cell. Next, cDNA was created using reverse transcriptase. After that, the mRNA was degraded, and using the following forward and reverse primers, a portion of the cDNA complementary to the mRNA encoding IL-1β, a portion of the cDNA complementary to the mRNA encoding iNOS, and a portion of the cDNA complementary to the mRNA encoding ARG1 were amplified. In addition, a portion of the cDNA complementary to the mRNA encoding glyceraldehyde-3-phosphate dehydrogenase (GADPH) was amplified as an internal positive control.

[0080] To amplify a portion of cDNA complementary to the mRNA encoding IL-1β, a forward primer having the sequence GCAACTGTTCCTGAACTCAACT (SEQ ID NO: 1) and a reverse primer having the sequence ATCTTTTGGGGTCCGTCAACT (SEQ ID NO: 2) were used.

[0081] To amplify a portion of cDNA complementary to the mRNA encoding iNOS, a forward primer having the sequence GTTCTCAGCCAACAATACAAGA (SEQ ID NO: 3) and a reverse primer having the sequence GTGGACGGGTCGATGTCAC (SEQ ID NO: 4) were used.

[0082] To amplify a portion of cDNA complementary to the mRNA encoding ARG1, a forward primer having the sequence CTCCAAGCCAAAGTCCTTAGAG (SEQ ID NO: 5) and a reverse primer having the sequence AGGAGCTGTCATTAGGGACATC (SEQ ID NO: 6) were used.

[0083] To amplify a portion of cDNA complementary to the mRNA encoding GADPH, a forward primer having the sequence AGGTCGGGTGTGAACGGATTTG (SEQ ID NO: 7) and a reverse primer having the sequence TGTAGACCATGTAGTTGAGGTCA (SEQ ID NO: 8) were used.

[0084] For RT-PCR, a real-time PCR system (QuantStudio (registered trademark) 3 manufactured by Thermo Fisher Scientific Inc.) was used. An intercalator (or a probe) that emits fluorescence when bound to double-stranded DNA was used, and a fluorescent signal was detected by irradiating excitation light.

[0085] [Immunostaining] Whether iNOS was produced in the cells was evaluated by immunostaining with an iNOS antibody. The cells were stained with a rabbit polyclonal antibody against iNOS (Anti-iNOS, antiab3523, Abcam) as the primary antibody, and Goat Anti-Rabbit IgG H&L (Alexa Fluor (registered trademark) 594) (ab150080, Abcam) as the secondary antibody and fluorescent dye. The stained cells were observed under a fluorescence microscope (BZ-X710, KEYENCE).

[0086] [result] Both RT-PCR and immunohistochemistry confirmed that IL-1β and iNOS were expressed in M1 macrophages, that ARG1 was expressed in M2 macrophages, and that none of IL-1β, iNOS, or ARG1 was expressed in M0 macrophages.

[0087] [Get parameters] Culture fluids of M0 macrophages, M1 macrophages, and M2 macrophages were placed on the sample stage of a holographic microscope (HoloMonitor M4 manufactured by PHI AB) and cultured for 36 hours in a CO2 incubator, during which time-lapse photography was performed at 30-minute intervals. The number of samples photographed was 7991 for M0 macrophages, 8129 for M1 macrophages, and 7103 for M2 macrophages.

[0088] From the obtained images, using software (HoloMonitor M4 App Suite, PHI AB), the above-mentioned "Area," "Irregularity," "Optical volume," and "Optical thickness" were obtained as parameters indicating cell morphology, along with the time elapsed since cell division (Time), and the above-mentioned "Migration distance," "Motility distance," and "Motility speed" were obtained as parameters indicating changes in cell position over time.

[0089] [Data Expansion] Data augmentation was performed by calculating the mean and variance for all obtained parameters and adding random values ​​within the range of (mean - variance) to (mean + variance).

[0090] [Replace outliers] For "migration distance," the average value was calculated, and values ​​exceeding 35 were replaced with that average value. For "Motility distance," the average value was calculated and any value exceeding 40 was replaced with that average value. For "Motility speed," the average value was calculated and any value exceeding 60 was replaced with that average value.

[0091] [Training and evaluating inference models] Of the dataset obtained through the above, 80% was used as training data and 20% as test data. The training data was input into a computer with scikit-learn installed, and an inference model was trained. Next, test data was input into the trained inference model, and the inference model was evaluated. The accuracy, precision, recall, and F-measure mentioned above were used as evaluation indicators for the inference model.

[0092] When calculating the accuracy rate, precision rate, and recall rate, the number of predictions of M1 macrophages from the M1 macrophage test data was defined as TP, the number of predictions of non-M1 macrophages from the M1 macrophage test data was defined as FN, the number of predictions of M1 macrophages from test data other than M1 macrophages was defined as FP, and the number of predictions of non-M1 macrophages from test data other than M1 macrophages was defined as TN.

[0093] The classification results were evaluated by changing the type of parameters used. The parameters used in the inference model are shown in Table 1. In Table 1, "〇" indicates that the parameter was used, and "-" indicates that the parameter was not used.

[0094] In Example 1, "Time" was not used, and instead "Area", "Irregularity", "Optical thickness", "Migration distance", "Motility distance", and "Motility speed" were used.

[0095] In Example 2, "Time", "Area", "Irregularity", and "Optical thickness" were used.

[0096] In Example 3, all parameters, “Time”, “Area”, “Irregularity”, “Optical thickness”, “Migration distance”, “Motility distance”, and “Motility speed”, were used.

[0097] In Example 4, "Optical thickness" was not used, and the other parameters "Time", "Area", "Irregularity", "Migration distance", "Motility distance", and "Motility speed" were used.

[0098] In Example 5, "Area", "Irregularity", and "Optical thickness" were not used, and the other parameters "Time", "Migration distance", "Motility distance", and "Motility speed" were used.

[0099] [Table 1]

[0100] The evaluation indices used were accuracy, precision, recall, F-score, ROC-AUC, and PR-AUC. The evaluation results are shown in Table 2.

[0101] [Table 2]

[0102] As shown in Table 2, by selecting parameters that indicate cell morphology and parameters that indicate the change in cell position over time, along with the time elapsed since cell division (Time), a highly rated inference model can be obtained.

[0103] By using the inference model obtained in this way, it is possible to predict cell phenotypes with high accuracy. Specifically, a culture solution of cells with unknown phenotypes is placed on the sample stage of a holographic microscope, and the parameters that were used in the inference model that received the highest evaluation are obtained from among the parameters that indicate the morphology of the cells or the parameters that indicate the change in position over time. The trained inference model is then used to infer the phenotype of the cells from the obtained parameters, and the inferred phenotype of the cells is displayed on a display or printed. This makes it possible to identify the phenotype of cells that change phenotype while the cells are still alive.

[0104] In the above embodiment, the method for inferring the phenotype of macrophages has been described as an example, but the present invention is not limited thereto. The cell phenotype inference device and cell phenotype inference method of the present invention can be applied to other cells whose phenotypes change. For example, the present invention can also be applied to devices and methods for inferring the phenotypes of microglia, Kupffer cells, alveolar macrophages (dust cells), etc.

[0105] Furthermore, the subject of inference in the inference device or inference method is not limited to the phenotype of a cell, and the state of a cell, such as the differentiation state of a cell (cell type), the trait of a cell, or whether the cell is alive or dead, may be inferred. In this case, too, after machine learning is performed on the inference model using training data, the inference model is evaluated using test data, and among the parameters indicating the change over time in the morphology or position of the cell, the parameters with high evaluation indexes based on the test data may be used to infer the state of the cell.

[0106] Example 6 [Evaluation of cell differentiation] (undifferentiated cell group) Human acute monocytic leukemia cells (THP-1) were purchased and cultured in Roswell Park Memorial Institute medium (RPMI medium) containing 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin (PS).

[0107] (differentiated cell group) Phorbol 12-myristate 13-acetate (PMA) was added to the undifferentiated cells, and they were differentiated into macrophages by culturing them at 37°C for 24 hours. When observed under a phase-contrast microscope, the cells differentiated by the addition of PMA attached to the bottom of the culture plate and showed the extension of pseudopods like macrophages.

[0108] The culture plate containing undifferentiated cells and the culture plate containing differentiated cells were placed on the sample stage of a holographic microscope (HoloMonitor M4, manufactured by PHI AB), and time-lapse photography was performed at 30-minute intervals while culturing for 36 hours in a CO2 incubator. The number of samples photographed was 5,356 for undifferentiated cells and 10,328 for differentiated cells.

[0109] From the images obtained, software (HoloMonitor M4 App Suite, PHI AB) was used to obtain the above-mentioned "Area," "Irregularity," and "Optical thickness" as parameters indicating cell morphology, and "Migration distance," "Motility distance," and "Motility speed" as parameters indicating changes in cell position over time.

[0110] As in Examples 1 to 5, data augmentation, outlier replacement, and inference model training and evaluation were performed.

[0111] As a result of the evaluation, the accuracy of the inference model using the time elapsed since cell division ("Time"), parameters indicating cell morphology ("Area"), "Irregularity", and "Optical thickness"), and parameters indicating changes in cell position over time ("Migration distance", "Motility distance", and "Motility speed") was 0.98, which was a high result. In addition, the ROC curve and PR curve showed that the quality of the inference model was high. This indicates that a model that identifies cell differentiation by machine learning using migration and three-dimensional morphological data as features was successfully developed.

[0112] To determine the important features for distinguishing undifferentiated cells from differentiated cells, we calculated the importance of the time elapsed since cell division (Time), parameters indicating cell morphology (Area), (Irregularity), and (Optical thickness), and parameters indicating the change in cell position over time (Migration distance), (Motility distance), and (Motility speed). The results showed that the most important parameter was Migration distance, followed by Motility distance. Human acute monocytic leukemia cells (THP-1) float and do not adhere to the bottom of the culture plate. When THP-1 differentiate into macrophages, they adhere to the plate surface and cell migration decreases. Although there are many adherent cell lines that migrate by extending pseudopodia, macrophages derived from TPH-1 have the important characteristic of less migration. Therefore, it was suggested that Migration distance and Motility distance are the most important parameters for distinguishing pre- and post-differentiation cells.

[0113] The method was found to be able to identify THP-1 differentiation non-invasively and over time. By adding holographically specific information, such as movement and 3D morphology data, to the general 2D morphology data, the accuracy was further improved, reaching 98% accuracy. [Explanation of symbols]

[0114] 1 Reasoning device 10. Computers 11 Processors 12 Main storage 13 Auxiliary storage device 14 Input Interface 15 Output Interface 20 Acquisition device 21 Diode Laser 22,23 Half mirror 24,25 Mirror 26 Objective Lens 27 Magnifying Glass 28 Image Sensor 29 Sample stage 30 Output Device

Claims

1. A cell state inference device that infers a state of a cell, the state of which is unknown, based on a parameter indicating a time-dependent change in a morphology or position of the cell, the device comprising: an acquisition unit for acquiring parameters indicative of a change over time in the morphology or position of a cell whose state is unknown; an inference unit that infers the state of the cell about which the acquisition unit acquired the information, based on a result of learning the association between a parameter indicating a time-dependent change in the morphology or position of the cell, the state of which is known, and the state of the cell; and An apparatus for inferring a cell state comprising an output unit that outputs the cell state inferred by the inference unit.

2. The acquisition unit includes a microscope capable of observing the three-dimensional morphology of cells. The cell state inference device according to claim 1 .

3. The device for inferring a state of a cell according to claim 1 or 2, wherein the cell has taxis.

4. A cell phenotype inference device that infers a phenotype of a cell, the phenotype of which is unknown, based on a parameter indicating a time-dependent change in a morphology or position of the cell, the device comprising: an acquisition unit for acquiring a parameter indicative of a change in morphology or position of a cell having an unknown phenotype over time; an inference unit that infers the phenotype of a cell about which the acquisition unit has acquired the information, based on a result of learning the association between a parameter indicating a time-dependent change in the morphology or position of a cell whose phenotype is known and the cell's phenotype; and An apparatus for inferring a cell phenotype comprising an output unit that outputs the cell phenotype inferred by the inference unit.

5. The apparatus for inferring a phenotype of a cell according to claim 4, wherein the cell is a macrophage, a microglia, a Kupffer cell, or an alveolar macrophage.

6. An apparatus for inferring a differentiation state of a cell, the differentiation state of which is unknown, based on a parameter indicating a time-dependent change in a morphology or position of the cell, the apparatus comprising: an acquisition unit for acquiring parameters indicative of changes over time in morphology or position of cells whose differentiation state is unknown; an inference unit that infers the differentiation state of the cell about which the acquisition unit has acquired the information, based on a result of learning the association between a parameter indicating a time-dependent change in the morphology or position of a cell whose differentiation state is known and the cell's differentiation state; and An apparatus for inferring the differentiation state of a cell, comprising an output unit that outputs the differentiation state of the cell inferred by the inference unit.

7. inputting parameters indicating changes over time in the morphology or position of a cell whose state is known and the state of the cell into a machine learning model as learning data, and causing the machine learning model to learn the association between the parameters indicating changes over time in the morphology or position of a cell whose state is known and the state of the cell; acquiring a parameter indicative of a change in morphology or position of a cell, the state of which is unknown, over time; inputting parameters indicative of changes in morphology or position of the unknown cell over time into the trained machine learning model; and A method for inferring a state of a cell, comprising a step of having the trained machine learning model infer the state of a cell whose state is unknown.

8. The method for inferring a state of a cell according to claim 7 , wherein the parameters are acquired using a microscope capable of observing the three-dimensional morphology of the cell.

9. The method for inferring a state of a cell according to claim 7 or 8, wherein the cell has taxis.

10. inputting parameters indicating a change over time in the morphology or position of a cell whose phenotype is known and the phenotype of the cell into an inference model as learning data, and causing the inference model to learn the association between the parameters indicating a change over time in the morphology or position of a cell whose phenotype is known and the phenotype of the cell; obtaining a parameter indicative of a change in morphology or location of a cell of unknown phenotype over time; inputting parameters representing changes in morphology or location of the cells with unknown phenotype over time into the trained inference model; and A method for inferring a cell phenotype, comprising a step of having the trained inference model infer the phenotype of a cell whose phenotype is unknown.

11. The method of claim 10, wherein the cell is a macrophage, microglia, Kupffer cell, or alveolar macrophage.

12. inputting parameters indicating changes over time in the morphology or position of a cell whose differentiation state is known and the differentiation state of the cell into an inference model as learning data, and causing the inference model to learn the association between the parameters indicating changes over time in the morphology or position of a cell whose differentiation state is known and the differentiation state of the cell; acquiring a parameter indicative of the change over time in morphology or position of a cell whose differentiation state is unknown; inputting parameters representing changes in morphology or position of the cells with unknown differentiation state over time into the trained inference model; and A method for inferring the differentiation state of a cell, comprising a step of having the trained inference model infer the differentiation state of a cell whose differentiation state is unknown.

Images