Particle analysis system, information processing device, and fractionation device

Abstract

This particle analysis system is capable of obtaining analysis results with high reproducibility from multi-dimensional data, the particle analysis system including: a particle analysis device comprising a light receiving unit that receives light emitted from bio-derived particles; and an information processing device comprising a dimensionality reduction unit that generates dimensionality reduction data by dimensionality reducing the multi-dimensional data related to the light outputted from the light receiving unit, and a learning unit that constructs a learning model by machine learning in which correlation between the dimensionality reduction data and the multi-dimensional data is the teacher.

Description

[Technical Field]

[0001] The present disclosure relates to a particle analysis system, an information processing device, and a fractionation device. [Background technology]

[0002] In recent years, in the fields of medicine and biochemistry, flow cytometers have been used to rapidly analyze the characteristics of large amounts of particles, and in the fields of medicine and biochemistry, cells or tissues stained with multiple fluorescent dyes have been measured using a fluorescence microscope to analyze the internal structure and dynamics of cells or tissues.

[0003] In analytical instruments such as flow cytometers and fluorescence microscopes, for example, fluorescence from multiple fluorescent dyes is dispersed using a prism or the like, and the dispersed fluorescence is detected using a light-receiving element array in which multiple light-receiving elements with different detection wavelength ranges are arranged. Therefore, the measurement data obtained by analytical instruments such as flow cytometers and fluorescence microscopes is multidimensional data that includes detection values ​​from multiple light-receiving elements.

[0004] For example, Patent Document 1 listed below discloses a technique for analyzing multidimensional data measured by a flow cytometer at higher speed in order to quickly determine whether or not a cell is a target for sorting. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] International Publication No. 2020 / 241722 Summary of the Invention [Problem to be solved by the invention]

[0006] However, because the technology disclosed in Patent Document 1 focuses on determining whether a cell is a target for sorting, it is difficult to obtain highly reproducible analysis results for the entire measured sample. Therefore, with the technology disclosed in Patent Document 1, it is difficult to compare, for example, the presence or absence of cell populations or differences in their distribution between samples.

[0007] Therefore, the present disclosure proposes a new and improved particle analysis system, information processing device, and fractionation device that are capable of obtaining highly reproducible analysis results from multidimensional data. [Means for solving the problem]

[0008] According to the present disclosure, there is provided a particle analysis system including: a particle analysis device having a light receiving unit that receives light emitted from biogenic particles; a dimensional compression unit that generates dimensionally compressed data by dimensionally compressing multidimensional data related to the light output from the light receiving unit; and a learning unit that constructs a learning model by machine learning using the correspondence between the dimensionally compressed data and the multidimensional data as a teacher.

[0009] Furthermore, according to the present disclosure, there is provided an information processing device including: a dimensional compression unit that generates dimensionally compressed data by dimensionally compressing multidimensional data related to light emitted from a biological particle; and a learning unit that constructs a learning model by machine learning using the correspondence between the dimensionally compressed data and the multidimensional data as a teacher.

[0010] Furthermore, according to the present disclosure, there is provided a sorting device including: a light-receiving unit that receives light emitted from the biological particles; a determining unit that determines whether the biological particles are to be sorted based on dimensionally reduced data corresponding to the multidimensional data, the dimensionally reduced data being estimated by a learning model that has been machine-learned to determine a correspondence between multidimensional data related to the light emitted from the biological particles and dimensionally reduced data obtained by reducing the dimensions of the multidimensional data; and a sorting unit that sorts the biological particles determined to be to be sorted. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a schematic diagram illustrating the configuration of a flow cytometer. [Figure 2] 1 is a block diagram showing a functional configuration of an information processing device according to a first embodiment of the present disclosure. [Figure 3] FIG. 2 is an explanatory diagram illustrating an example of multidimensional data and dimensionally compressed data. [Figure 4] FIG. 2 is an explanatory diagram illustrating an example of a learning model constructed by a learning unit. [Figure 5] FIG. 10 is a flowchart illustrating an example of the flow of operations of the information processing device according to the embodiment. [Figure 6] FIG. 10 is a graph showing the results of plotting dimensionality compressed data by UMAP on a two-dimensional coordinate space in an example of the embodiment. [Figure 7] FIG. 10 is a graph showing the results of plotting, on a two-dimensional coordinate space, dimensionally compressed data obtained by inputting multidimensional data used in machine learning into a learning model in an example of the same embodiment. [Figure 8] FIG. 10 is a graph showing the results of plotting, in a two-dimensional coordinate space, dimensionally reduced data obtained by inputting other measurement data of the same sample into a learning model in an example of the same embodiment. [Figure 9] FIG. 10 is a block diagram showing a functional configuration of an information processing device according to a second embodiment of the present disclosure. [Figure 10] FIG. 10 is a flowchart illustrating an example of the flow of operations of the information processing device according to the embodiment. [Figure 11] FIG. 10 is a graph showing the results of plotting dimensionally compressed data obtained by inputting multidimensional data into a learning model in a two-dimensional coordinate space in an example of the same embodiment. [Figure 12] FIG. 10 is an explanatory diagram showing the results of comparing dimensionality-reduced data estimated using a learning model between samples in an example of the same embodiment. [Figure 13] FIG. 10 is an explanatory diagram showing the results of comparing dimensionality compressed data by UMAP between samples in an example of the embodiment. [Figure 14] FIG. 10 is a block diagram showing a functional configuration of a particle analysis system according to a third embodiment of the present disclosure. [Figure 15] FIG. 10 is a flowchart showing an example of the operation flow of the particle analysis system according to the embodiment. [Figure 16] FIG. 1 is a schematic diagram showing a general configuration of a fluorescence imaging device. [Figure 17] 1 is a block diagram showing an example of a hardware configuration of an information processing device according to first to third embodiments of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] Preferred embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0013] The explanation will be given in the following order. 1. First embodiment 1.1. Overview of Flow Cytometers 1.2. Example of information processing device configuration 1.3. Example of operation of information processing device 1.4. Example of dimensionality reduction data 2. Second embodiment 2.1. Example of information processing device configuration 2.2. Example of operation of information processing device 2.3. Example of dimensionality reduction data 3. Third embodiment 3.1. Example of particle analysis system configuration 3.2. Example of particle analysis system operation 4. Variations 5. Hardware configuration example

[0014] <1. First embodiment> (1.1. Overview of Flow Cytometers) First, an overview of a flow cytometer to which the technology according to the present disclosure is applied will be described with reference to Fig. 1. Fig. 1 is a schematic diagram showing the general configuration of a flow cytometer 10.

[0015] As shown in FIG. 1, a flow cytometer 10 includes a laser light source 11, a flow cell 12, a detection optical unit 13, and a photodetector 14.

[0016] The flow cytometer 10 is an analytical device that irradiates a measurement object S flowing at high speed through a flow cell 12 with laser light from a laser light source 11, and detects fluorescence and the like emitted from each measurement object S with a photodetector 14.

[0017] The measurement target S is, for example, a cell, tissue, microorganism, or biological particle derived from a living body, such as a cell stained with multiple fluorescent dyes, tissue, microorganism, or biological particle. The cell is, for example, an animal cell (blood cell) or a plant cell. The tissue is, for example, a tissue collected from the human body or a part of the tissue (including tissue cells). The microorganism is, for example, a bacterium such as Escherichia coli, a virus such as tobacco mosaic virus, or a fungus such as yeast. The biological particle is, for example, various organelles (cell organelles) constituting a cell, such as chromosomes, liposomes, or mitochondria, or a biological macromolecule such as a nucleic acid, protein, lipid, sugar chain, or a complex thereof. These biological particles may be spherical or non-spherical, and there are no particular limitations on their size or mass.

[0018] The measurement target S may also be artificial particles such as latex particles, gel particles, or industrial particles. The industrial particles may be, for example, organic resin particles such as polystyrene or polymethyl methacrylate, inorganic material particles such as glass, silica, or magnetic materials, or metal particles such as gold colloid or aluminum. Similarly, these artificial particles may be spherical or non-spherical, and there are no particular limitations on their size or mass.

[0019] The object to be measured S is stained (labeled) in advance with multiple fluorescent dyes. The labeling of the object to be measured S with fluorescent dyes may be performed by a known method. Specifically, when the object to be measured S is a cell, the cell can be fluorescently labeled by mixing the cell of the object to be measured S with a fluorescently labeled antibody that selectively binds to an antigen present on the cell surface, and allowing the fluorescently labeled antibody to bind to the antigen on the cell surface through an antigen-antibody reaction. Alternatively, the cell can be fluorescently labeled by mixing the cell of the object to be measured S with a fluorescent dye that is selectively taken up by specific cells, and allowing the fluorescent dye to be taken up by the cells.

[0020] A fluorescently labeled antibody is an antibody to which a fluorescent dye is bound as a label. For example, the fluorescently labeled antibody may be an antibody to which a fluorescent dye is directly bound, or may be an antibody to which a fluorescent dye is bound via avidin using the avidin-biotin reaction. The antibody may be either a polyclonal antibody or a monoclonal antibody. Any known fluorescent dye used for cell staining or the like can be used.

[0021] The laser light source 11 emits, for example, laser light of a wavelength capable of exciting a fluorescent dye that has stained (labeled) the measurement object S. When multiple fluorescent dyes are used to stain the measurement object S, multiple laser light sources 11 may be provided so as to be able to emit laser light corresponding to the excitation wavelengths of the multiple fluorescent dyes. For example, the laser light source 11 may be a semiconductor laser light source. The laser light emitted from the laser light source 11 may be pulsed light or continuous light.

[0022] The flow cell 12 is a flow path through which measurement objects S, such as cells, are aligned in one direction and allowed to flow. Specifically, the flow cell 12 allows the measurement objects S to flow in one direction and aligned by causing a sheath liquid that envelops a sample liquid containing the measurement objects S to flow as a laminar flow at high speed. The measurement objects S flowing through the flow cell 12 are irradiated with laser light emitted from a laser light source 11. Fluorescence emitted from the measurement objects S irradiated with the laser light passes through a detection optical unit 13 and is then detected by a photodetector 14.

[0023] The detection optical unit 13 is an optical element that allows light in a predetermined detection wavelength range, among the light emitted from the measurement target S irradiated with laser light, to reach the photodetector 14. The detection optical unit 13 may be, for example, a spectroscopic element such as a prism or grating that can obtain a spectrum by dispersing incident light.

[0024] Alternatively, the detection optical unit 13 may be an optical element that separates the fluorescence emitted from the measurement target S irradiated with laser light into light components in a predetermined detection wavelength range. In such a case, the detection optical unit 13 is configured to include, for example, at least one dichroic mirror or optical filter, and can separate the fluorescence from the measurement target S into light components in a predetermined detection wavelength range using optical members such as a dichroic mirror and an optical filter. In this way, the detection optical unit 13 can allow the separated light components in the predetermined detection wavelength range to reach the corresponding photodetectors 14.

[0025] The photodetector 14 includes a group of light-receiving elements that detect fluorescence emitted from the measurement target S irradiated with laser light. The group of light-receiving elements may be, for example, a light-receiving element array in which a plurality of light-receiving elements, such as photomultiplier tubes (PMTs) or photodiodes, each having a different detectable wavelength range, are arranged one-dimensionally along the direction in which light is separated by the detection optical unit 13.

[0026] Alternatively, the photodetector 14 may be configured to include, for example, a plurality of light-receiving elements that receive light corresponding to the wavelength range of the fluorescent dye separated by the detection optical unit 13. Furthermore, the photodetector 14 may be configured to include, for example, an imaging element such as a CCD image sensor or a CMOS image sensor. In such a case, the photodetector 14 can acquire an image of the measurement target S (for example, a bright-field image, a dark-field image, or a fluorescent image) using the imaging element.

[0027] In the flow cytometer 10 configured as described above, first, laser light from the laser light source 11 is irradiated onto the measurement object S flowing at high speed through the flow cell 12, causing fluorescence to be emitted from the measurement object S. As a result, the fluorescence emitted from the measurement object S is separated by the detection optical unit 13 and then detected by the photodetector 14. In the photodetector 14, the fluorescence emitted from the measurement object S is detected by each of a plurality of light-receiving elements with different detectable wavelength ranges of light. This allows the flow cytometer 10 to acquire multidimensional data from the measurement object S, including the fluorescence intensity of the fluorescent substance that labels the measurement object S.

[0028] (1.2. Configuration example of information processing device) Next, an information processing device according to the first embodiment of the present disclosure will be described with reference to Fig. 2. Fig. 2 is a block diagram showing the functional configuration of the information processing device 100 according to this embodiment.

[0029] The information processing device 100 according to this embodiment can generate lower-dimensional dimension-reduced data using multidimensional data obtained by the flow cytometer 10, etc. This allows the information processing device 100 to convert the multidimensional data obtained by the flow cytometer 10, etc. into dimension-reduced data that is easier to view or analyze, and present it to the user.

[0030] As shown in FIG. 2, the information processing device 100 includes, for example, an input unit 110, a dimensionality reduction unit 120, a learning unit 130, a learning model storage unit 140, an analysis unit 150, and an output unit 160.

[0031] The input unit 110 is an input port for inputting multidimensional data to the information processing device 100. Specifically, the input unit 110 is a connection port capable of receiving various data from an external device such as the flow cytometer 10. For example, the input unit 110 is a USB (Universal Serial Bus) port, IEEE1394 port, or SCSI (Small It may be a Computer System Interface (CSI) port or the like.

[0032] The multidimensional data input to the input unit 110 may include, for example, data regarding the light intensity received by each light-receiving element included in the photodetector 14 of the flow cytometer 10. For example, the multidimensional data input to the input unit 110 may be light intensity data that has been normalized after correcting for overlapping of the fluorescence wavelength ranges of each fluorescent dye from the light intensity received by each light-receiving element (i.e., fluorescence correction). Alternatively, the multidimensional data input to the input unit 110 may be raw data of the light intensity received by each light-receiving element.

[0033] Alternatively, the multidimensional data input to the input unit 110 may include data on the fluorescence intensity from each fluorescent dye calculated by analyzing the fluorescence spectrum measured by the light-receiving element array. For example, the multidimensional data input to the input unit 110 may be fluorescence intensity data of each fluorescent dye that has been separated from the fluorescence spectrum measured by the light-receiving element array in consideration of the fluorescence wavelength range of the fluorescent dye and then normalized. Furthermore, the multidimensional data input to the input unit 110 may be raw data of the fluorescence spectrum measured by the light-receiving element array.

[0034] The dimensionality reduction unit 120 performs dimensional compression on the multidimensional data input to the input unit 110 to generate dimensionally reduced data having a reduced number of dimensions compared to the multidimensional data. Specifically, the dimensionality reduction unit 120 performs dimensional compression on the multidimensional data input to the input unit 110 using t-SNE (t-distributed Stochastic Neighbor Embedding) or UMAP (Uniform Manifold The dimension reduction unit 120 may generate dimension-reduced data by reducing the dimension of the multidimensional data using a nonlinear technique such as linear approximation and projection. The dimension reduction unit 120 may reduce the dimension of the multidimensional data to two or three dimensions, for example.

[0035] Here, the dimensional compression performed by the dimensional compression unit 120 will be described in more detail with reference to Fig. 3. Fig. 3 is an explanatory diagram showing an example of multidimensional data and dimensionally compressed data.

[0036] 3, the multidimensional data MD is, for example, data relating to the light intensity emitted from each of M cells and received by each of N light receiving elements (channels: CH). That is, the multidimensional data MD is M pieces of N-dimensional (e.g., 20-dimensional) data. The multidimensional data MD may be raw data output from each of the N light receiving elements, or may be data that has been normalized after fluorescence correction.

[0037] The dimensionality reduction unit 120 generates M pieces of two-dimensional dimensionally reduced data DC by reducing the dimensionality of the multidimensional data MD using a nonlinear method such as t-SNE or UMAP. The dimensionality reduced data DC can be used as coordinate data for plotting M cells in a two-dimensional coordinate space so that similar cells are placed close to each other and dissimilar cells are placed far from each other. This allows the dimensionality reduced data DC plotted in the two-dimensional coordinate space to be a collection of subsets according to the type or characteristics of cells, making it possible to clearly visualize what cell populations are included in the M cells.

[0038] The learning unit 130 constructs a learning model by machine learning using the correspondence between the multidimensional data and the dimensionally compressed data generated by the dimensionality compression unit 120 as a teacher.

[0039] For example, dimension reduction methods such as t-SNE or UMAP perform nonlinear transformations during the dimension reduction process, resulting in changes in the dimension reduction results each time dimension reduction is performed, resulting in low reproducibility. Specifically, with dimension reduction methods such as t-SNE or UMAP, even when the same multidimensional data is input, the distribution of subsets in the dimension reduction results may change each time dimension reduction is performed. Therefore, when multidimensional data is dimension reduced using t-SNE or UMAP, it becomes difficult to compare dimension-reduced data between different samples, making it difficult to discover unknown cell populations, for example.

[0040] The learning unit 130 performs machine learning using one of the dimensionality reduction processes from multidimensional data to dimensionality-compressed data performed by the dimensionality reduction unit 120 as a teacher. Specifically, the learning unit 130 performs machine learning using the multidimensional data and the dimensionality-compressed data generated from the multidimensional data by the dimensionality reduction process as a teacher. In this way, by inputting multidimensional data, the learning unit 130 can construct, by machine learning, a learning model that outputs dimensionality-compressed data corresponding to the input multidimensional data. In other words, the learning unit 130 can construct a learning model that reproduces the dimensionality reduction process that generates dimensionality-compressed data used in machine learning from multidimensional data.

[0041] Therefore, the learning model constructed by the learning unit 130 can generate dimension-reduced data from multidimensional data with high reproducibility. In such cases, the dimension-reduced data generated via the learning model has high reproducibility of the distribution of subsets in the dimension-reduced results, making it possible to more easily identify each subset. Furthermore, the dimension-reduced data generated via the learning model facilitates comparison between samples due to the above-mentioned characteristics, making it possible to more easily discover unknown cell populations.

[0042] Here, the learning model constructed by the learning unit 130 will be described in more detail with reference to Fig. 4. Fig. 4 is an explanatory diagram showing an example of the learning model constructed by the learning unit 130.

[0043] As shown in FIG. 4, the learning unit 130 may construct a learning model using a neural network DN including an input layer IL, two or more hidden layers ML, and an output layer OL.

[0044] The number of nodes in the input layer IL is the same as the number of dimensions of the multidimensional data (e.g., 20 dimensions), and the number of nodes in the output layer OL is the same as the number of dimensions of the dimensionally reduced data (e.g., 2 dimensions). The number of nodes in each layer of the hidden layer ML is set to a number between the number of nodes in the input layer IL and the number of nodes in the output layer OL (e.g., 15, 10, 5, etc.) so that the number of nodes gradually decreases from the input layer IL to the output layer OL.

[0045] In the above-described neural network DN, data for each dimension of the multidimensional data is applied to each node of the input layer IL, and data for each dimension of the dimension-reduced data is applied to each node of the output layer OL. For example, in the multidimensional data and dimension-reduced data shown in FIG. 3, data related to the light intensity received by each of N light-receiving elements (channels: CH) is applied to each node of the input layer IL, and coordinate data in a two-dimensional coordinate space is applied to each node of the output layer OL. The learning unit 130 can construct a learning model that reproduces the dimension reduction process by optimizing the network structure and weighting in the neural network DN to which the above-described data is applied.

[0046] However, the learning unit 130 may construct a learning model using a machine learning method other than the neural network DN. For example, the learning unit 130 may reproducibility It is possible to construct a learning model using any method as long as it is supervised machine learning with

[0047] The above-described dimensionality reduction unit 120 and learning unit 130 may be provided on a server or cloud external to the information processing device 100. For example, the information processing device 100 may transmit multidimensional data to the server or cloud via a network, thereby causing the dimensionality reduction unit 120 on the server or cloud to generate dimensionality-reduced data. Furthermore, the information processing device 100 may receive, via a network, a learning model constructed by machine learning the correspondence between the multidimensional data and the dimensionality-reduced data in the learning unit 130 on the server or cloud.

[0048] The learning model storage unit 140 stores the learning model constructed by the learning unit 130. Specifically, the learning model storage unit 140 may store, as the learning model, a neural network DN whose network structure and weighting have been optimized by machine learning by the learning unit 130. The learning model storage unit 140 may be configured, for example, with a magnetic storage device such as an HDD (Hard Disk Drive), a semiconductor storage device, an optical storage device, or a magneto-optical storage device.

[0049] The analysis unit 150 uses the learning model constructed by the learning unit 130 to estimate dimensionality-reduced data corresponding to the multidimensional data input to the input unit 110. For example, when the learning model is constructed using the neural network DN shown in FIG. 4, the analysis unit 150 may estimate, as dimensionality-reduced data corresponding to the multidimensional data, data output from the output layer OL when multidimensional data is input to the input layer IL of the neural network DN. By using the learning model constructed by the learning unit 130, the analysis unit 150 can reproduce the same dimensionality reduction process performed by the dimensionality reduction unit 120, and therefore can estimate dimensionality-reduced data corresponding to the multidimensional data with high reproducibility.

[0050] The output unit 160 is a device capable of presenting the dimensionality reduced data estimated by the analysis unit 150 to a user. The output unit 160 is, for example, an LCD (Liquid Crystal The output unit 160 may be a display device such as a PDP (Plasma Display Panel), an OLED (Organic Light Emitting Diode) display, a hologram, or a projector, or may be a printing device such as a printer. The output unit 160 may output the estimated dimensionality reduced data as a scatter diagram plotted in a two-dimensional or three-dimensional coordinate space.

[0051] Alternatively, the output unit 160 may be an external output port that outputs the dimensionally compressed data to an external device that can present the dimensionally compressed data to a user. For example, the output unit 160 may be a connection port such as a USB port, an IEEE 1394 port, or a SCSI port that can transmit the dimensionally compressed data to an external device. In this case, the output unit 160 can output the dimensionally compressed data to an external display device or printer, and cause the display device or printer to present the dimensionally compressed data to a user.

[0052] The information processing device 100 having the above configuration can construct a learning model by machine learning the correspondence between multidimensional data obtained by the flow cytometer 10 or the like and dimension-reduced data obtained by dimensionally reducing the multidimensional data. By using the constructed learning model, the information processing device 100 can estimate dimension-reduced data corresponding to the multidimensional data without using a nonlinear dimension reduction method. Furthermore, since the constructed learning model can reproduce the same dimension reduction process, it has high reproducibility and can return the same output for the same input. Therefore, the information processing device 100 can obtain dimension-reduced data corresponding to the multidimensional data with high reproducibility, making it easier to perform analysis between samples.

[0053] (1.3. Example of operation of information processing device) Next, an example of the operation of the information processing device 100 according to this embodiment will be described with reference to Fig. 5. Fig. 5 is a flowchart showing an example of the flow of the operation of the information processing device 100 according to this embodiment.

[0054] 5, first, a portion of a sample is measured by an analytical device such as a flow cytometer 10 to acquire multidimensional data (S101). The acquired multidimensional data is data that includes information related to the fluorescence intensity measured by the photodetector 14 of the flow cytometer 10. The multidimensional data acquired by the analytical device is input to the information processing device 100 via the input unit 110.

[0055] Next, the dimension reduction unit 120 uses a dimension reduction method such as t-SNE or UMAP to reduce the dimension of the multidimensional data, thereby generating dimension-reduced data (S102).

[0056] Next, using the correspondence between the multidimensional data acquired in step S101 and the dimension-reduced data obtained in step S102 as a teacher, the learning unit 130 performs machine learning to construct a learning model (S103). For example, the learning unit 130 may construct the learning model by optimizing the network structure and weighting of a neural network DN that includes two or more hidden layers ML, in which multidimensional data is applied to an input layer IL and dimension-reduced data is applied to an output layer OL. The constructed learning model is stored in the learning model storage unit 140, for example.

[0057] Thereafter, the remaining portion of the sample is measured by an analytical device to obtain multidimensional data (S104). Subsequently, using the constructed learning model, the analysis unit 150 estimates dimension-reduced data corresponding to the multidimensional data obtained in step S104 (S105). For example, if the learning model is constructed using a neural network DN, the analysis unit 150 may estimate the output of the output layer OL when the multidimensional data obtained in step S104 is input to the input layer IL as dimension-reduced data corresponding to the multidimensional data.

[0058] Next, the estimated dimensionality-reduced data is presented to the user via the output unit 160 (S106). This allows the information processing device 100 to estimate two-dimensional or three-dimensional dimensionality-reduced data from the multidimensional data measured by the analysis device using the constructed learning model. The two-dimensional or three-dimensional dimensionality-reduced data can be easily visually understood by plotting it in coordinate space, allowing the user to intuitively understand the distribution of a subset of biological particles (e.g., cells) measured by the analysis device.

[0059] (1.4. Example of dimensionally compressed data) Next, an example of dimensionally compressed data in the information processing device according to this embodiment will be described with reference to FIGS.

[0060] First, a simulator was used to generate 20-dimensional multidimensional data that simulated the measurement results of a flow cytometer. Specifically, measurement data (multidimensional data) of a total of 10,000 cell populations, each containing 500 cells singly stained with 20 different fluorescent dyes, was generated. In other words, the generated measurement data was a total of 10,000 pieces of 20-dimensional multidimensional data.

[0061] Next, of the 10,000 pieces of multidimensional data, a total of 8,000 pieces of multidimensional data, each containing 400 pieces of single-stained cells of 20 different types, were subjected to dimensionality compression using UMAP to generate 8,000 pieces of dimensionally compressed data. Figure 6 shows the results of plotting the dimensionality compressed data compressed using UMAP on a two-dimensional coordinate space.

[0062] We also constructed a learning model by performing machine learning using the correspondence between the above 8,000 multidimensional data items and the dimensionality-reduced data obtained by UMAP as training data. Specifically, we used a neural network with 20 nodes in the input layer, 2 nodes in the output layer, and three intermediate layers with nodes numbering 15, 10, and 5, respectively, from the input layer. We applied the multidimensional data to the input layer and the dimensionality-reduced data to the output layer to perform machine learning and construct a learning model. Next, we input the multidimensional data used in the machine learning into the learning model constructed above, and output dimensionality-reduced data from the output layer. Figure 7 shows the results of plotting the output dimensionality-reduced data in two-dimensional coordinate space.

[0063] Furthermore, the remaining 2,000 pieces of multidimensional data, consisting of 100 pieces of each of 20 types of single-stained cells (a total of 2,000 pieces of multidimensional data) after subtracting the 8,000 pieces of multidimensional data from the 10,000 pieces of multidimensional data, were input into the learning model constructed above, and dimension-reduced data was output from the output layer. Figure 8 shows the result of plotting the output dimension-reduced data in two-dimensional coordinate space.

[0064] 6 and 7, the information processing device according to this embodiment can reproduce the dimensionality reduction process learned by machine learning using UMAP with high accuracy. Furthermore, as shown in FIGS. 7 and 8, the information processing device according to this embodiment can obtain similar dimensionality reduced data with high accuracy even when multidimensional data of the same sample that was not used as training data is input.

[0065] Therefore, the information processing device according to this embodiment can reproduce the same dimensionality reduction process with high accuracy by constructing a learning model that uses machine learning to learn a dimensionality reduction process that includes a nonlinear transformation such as UMAP. Furthermore, the information processing device according to this embodiment can obtain dimensionality reduced data with high reproducibility of the distribution of subsets from multiple samples that have similar trends, making it easier to compare dimensionality reduced data between multiple samples.

[0066] 2. Second embodiment (2.1. Configuration example of information processing device) Next, an information processing device according to a second embodiment of the present disclosure will be described with reference to Fig. 9. Fig. 9 is a block diagram showing the functional configuration of an information processing device 101 according to this embodiment. The information processing device according to this embodiment is capable of estimating dimensionally reduced data at high speed even for new multidimensional data by using the learning model constructed by the information processing device 100 according to the first embodiment.

[0067] As shown in FIG. 9, the information processing device 101 includes, for example, an input unit 110, a learning model storage unit 140, an analysis unit 150, and an output unit 160.

[0068] The input unit 110, the analysis unit 150, and the output unit 160 have substantially the same configuration as those described in the information processing device 100 according to the first embodiment, and therefore description thereof will be omitted here.

[0069] The learning model storage unit 140 stores a learning model for estimating dimension-reduced data from multidimensional data. Specifically, similar to the first embodiment, the learning model storage unit 140 may store a learning model in which a dimension reduction process such as UMAP is machine-learned using a neural network including an input layer IL, two or more hidden layers ML, and an output layer OL. The learning model stored in the learning model storage unit 140 is machine-learned, for example, by the information processing device 100 according to the first embodiment, and the network structure and weighting are optimized. Therefore, the subsequent analysis unit 150 can estimate dimension-reduced data corresponding to the multidimensional data acquired by the input unit 110, similar to the information processing device 100 according to the first embodiment.

[0070] The learning model storage unit 140 may be provided on a server or cloud external to the information processing device 101. For example, the information processing device 101 may estimate dimensionally compressed data corresponding to multidimensional data by referring to a learning model stored in the learning model storage unit 140 on a server or cloud via a network.

[0071] The information processing device 101 having the above configuration can analyze the measurement results of the flow cytometer 10 and the like in a shorter time by using a learning model in which the network structure and weighting have already been optimized by machine learning. Furthermore, the information processing device 101 can reduce the time required for calculations to construct a learning model, thereby suppressing deterioration of the measured sample over time.

[0072] (2.2. Example of operation of information processing device) Next, an example of the operation of the information processing device 101 according to this embodiment will be described with reference to Fig. 10. Fig. 10 is a flowchart showing an example of the flow of the operation of the information processing device 101 according to this embodiment.

[0073] 10, first, a sample is measured by an analytical device such as a flow cytometer 10 to acquire multidimensional data (S201). The acquired multidimensional data is data including information related to the fluorescence intensity measured by the photodetector 14 of the flow cytometer 10. The multidimensional data acquired by the analytical device is input to the information processing device 101 via the input unit 110.

[0074] Next, using the learning model stored in the learning model storage unit 140, the analysis unit 150 estimates dimension-reduced data corresponding to the multidimensional data (S202). The learning model is, for example, a neural network that includes an input layer, two or more intermediate layers, and an output layer, and has undergone machine learning of a dimension reduction process such as t-SNE or UMAP. For example, the analysis unit 150 may estimate the output of the output layer OL when the multidimensional data acquired in step S201 is input to the input layer IL of the learning model as dimension-reduced data corresponding to the multidimensional data.

[0075] The estimated dimensionally reduced data is presented to the user via the output unit 160 (S203). This allows the user to visualize the multidimensional data measured by the analysis device in a two-dimensional or three-dimensional format that is easy to understand visually. Coordinate Space Therefore, the information processing device 101 allows the user to more intuitively understand the distribution of various groups contained in the sample.

[0076] (2.3. Example of dimensionally compressed data) Next, an example of dimensionally compressed data in the information processing device according to this embodiment will be described with reference to FIGS.

[0077] First, using a simulator, measurement data (multidimensional data) of a total of 10,000 cell populations, each containing 500 cells singly stained with 20 different fluorescent dyes, was generated, as in (1.4. Example of dimensionally reduced data). However, the generated measurement data was different from the measurement data generated in (1.4. Example of dimensionally reduced data), having similar trends, by changing the noise added to it.

[0078] Next, the generated 10,000 pieces of multidimensional data were input into the learning model constructed in (1.4. Example of dimensionality-reduced data), and dimensionality-reduced data was output from the output layer. The output dimensionality-reduced data was plotted in two-dimensional coordinate space, as shown in Figure 11.

[0079] 11, it can be seen that the information processing device according to this embodiment can obtain dimensionality-reduced data that is almost identical to the dimensionality-reduced data used as training data (the dimensionality-reduced data shown in FIG. 7) even when multidimensional data having similar tendencies but that are not identical is input to the learning model. Therefore, it can be seen that the information processing device according to this embodiment can estimate dimensionality-reduced data from multidimensional data using an already constructed learning model, without constructing a new learning model, for samples having similar tendencies.

[0080] Next, a simulator was used to generate measurement data (multidimensional data) for a total of 9,000 cell populations, each containing 500 cells singly stained with 18 different fluorescent dyes. The generated 9,000 multidimensional data were then subjected to dimensionality reduction using UMAP to generate dimensionality-reduced data, and a learning model was constructed by performing machine learning using the correspondence between the 9,000 multidimensional data and the dimensionality-reduced data as training data. The learning model used a neural network with 20 nodes in the input layer, 2 nodes in the output layer, and 15, 10, and 5 nodes in the three hidden layers, starting from the input layer.

[0081] The multidimensional data used in machine learning was input into the constructed learning model, and dimensionality-reduced data was output from the output layer. The output dimensionality-reduced data was plotted in a two-dimensional coordinate space, as shown on the left in Figure 12. The same 9,000 pieces of multidimensional data were also dimensionally reduced using UMAP, and the resulting dimensionality-reduced data was plotted in a two-dimensional coordinate space, as shown on the left in Figure 13.

[0082] Furthermore, we used a simulator to generate measurement data (multidimensional data) for a total of 10,000 cell populations, each containing 500 cells singly stained with 20 fluorescent dyes, including the 18 fluorescent dyes mentioned above. The generated 10,000 multidimensional data were then input into the learning model constructed above (the learning model that generated the dimensionality-reduced data shown on the left in Figure 12), and dimensionality-reduced data was output from the output layer. The output dimensionality-reduced data was plotted in two-dimensional coordinate space, as shown on the right in Figure 12. Furthermore, the same 10,000 multidimensional data were dimensionally reduced using UMAP, and the resulting dimensionality-reduced data was plotted in two-dimensional coordinate space, as shown on the right in Figure 13.

[0083] The plot results of the dimensionality-reduced data shown on the right of Figures 12 and 13 include new sets UC1 and UC2 derived from two cell populations not included in the teacher multidimensional data or the dimensionality-reduced data. However, the dimensionality-reduced data generated by dimensionality reduction using UMAP shown on the right of Figure 13 has low reproducibility and the distribution of each cell population has changed, making it difficult to determine the new sets UC1 and UC2 from the distribution shape and arrangement of the cell populations. On the other hand, the dimensionality-reduced data generated by estimation using the learning model shown on the right of Figure 12 has high reproducibility and the distribution of each cell population has remained almost unchanged, making it easy to determine the new sets UC1 and UC2 from the distribution shape and arrangement of the cell populations.

[0084] Therefore, the information processing device according to this embodiment can estimate dimensionality-reduced data corresponding to multidimensional data by using a learning model that has undergone machine learning to learn the dimensionality reduction process from multidimensional data to dimensionality-reduced data. This increases the reproducibility of dimensionality-reduced data, making it easier to compare dimensionality-reduced data between different samples and to easily identify unknown cell populations.

[0085] Comparison of measurement data using such dimension-reduced data can be used, for example, to compare a sample (e.g., blood) collected from a patient with a sample collected from a healthy individual. This makes it possible to easily identify cell populations that are specifically expressed in the patient. Comparison of measurement data using dimension-reduced data can also be used to compare samples collected on different dates from the same patient, or to compare measurement data of samples actually collected from the patient with model data. Furthermore, comparison of measurement data using dimension-reduced data can be used to compare cell samples cultured under different conditions. This makes it possible to easily detect changes in cell samples due to the presence or absence of different drugs, thereby making it possible to easily determine the efficacy of drugs on cell samples.

[0086] 3. Third Embodiment (3.1. Example of particle analysis system configuration) Next, a particle analysis system according to a third embodiment of the present disclosure will be described with reference to Fig. 14. Fig. 14 is a block diagram showing the functional configuration of a particle analysis system 20 according to this embodiment. The particle analysis system 20 according to this embodiment can quickly determine a group to be separated from the measurement target S by using the machine-learned learning model described in the information processing device 100 according to the first embodiment. For example, the particle analysis system 20 according to this embodiment is a so-called cell sorter that can selectively separate a group to be separated from the measurement target S.

[0087] As shown in FIG. 14, the particle analysis system 20 includes an information processing device 102 and a particle analysis device 200.

[0088] The information processing device 102 includes an input unit 110, a dimensionality reduction unit 120, a learning unit 130, a learning model storage unit 140, an analysis unit 150, an output unit 160, and a fractionation designation unit 170.

[0089] The input unit 110, dimensional compression unit 120, learning unit 130, learning model storage unit 140, analysis unit 150, and output unit 160 are substantially the same as the configurations described in the information processing device 100 according to the first embodiment, and therefore will not be described here.

[0090] The dimensionality-reduced data output from the output unit 160 is visually recognized by the user as, for example, a scatter diagram plotted in a two-dimensional or three-dimensional coordinate space. From the scatter diagram on which the dimensionality-reduced data is plotted, the user can confirm the number of groups included in the measurement target S, the variance within the groups, the distance between the groups, and the like.

[0091] Based on input from the user, the fractionation designation unit 170 designates the range of the measurement target S to be fractionated by the particle analysis device 200. For example, based on input from the user, the fractionation designation unit 170 may designate the range of the measurement target S to be fractionated by designating the coordinate range of dimensionally compressed data plotted in a two-dimensional or three-dimensional coordinate space.

[0092] Particle analysis device 200 includes input unit 210, analysis unit 220, determination unit 230, and fractionation unit 240. Particle analysis device 200 estimates dimension-reduced data corresponding to the multidimensional data of measurement target S in real time using a machine-learned learning model, thereby determining whether measurement target S is a fractionation target or not, and can fractionate measurement target S that is a fractionation target.

[0093] The input unit 210 acquires measurement data of the measurement target S from the photodetector 14 or the like. Specifically, the input unit 210 acquires measurement data of each particle of the measurement target S as multidimensional data. For example, the input unit 210 may acquire measurement data relating to the fluorescence intensity of the particles of the measurement target S for each wavelength range as multidimensional data.

[0094] The analysis unit 220 uses the learning model stored in the learning model storage unit 140 to estimate dimension-reduced data corresponding to the multidimensional data input to the input unit 210. In the particle analysis system 20 according to this embodiment, the analysis unit 220 estimates dimension-reduced data using the same learning model as the analysis unit 150 of the information processing device 102, so the particle analysis device 200 can obtain dimension-reduced data similar to that of the information processing device 102. The learning model stored in the learning model storage unit 140 may be a learning model constructed using a portion of the measurement target S including the fractionation target, or may be a learning model constructed in advance using other samples, etc.

[0095] The dimensionality reduced data estimated by the analysis unit 220 may be presented to the user in real time via the output unit 160. This allows the output unit 160 to plot the dimensionality reduced data estimated by the analysis unit 220 together with the range specified as the collection target in a two-dimensional or three-dimensional coordinate space and present it to the user. This allows the user to observe the dimensionality reduced data plotted in the two-dimensional or three-dimensional coordinate space in real time, and therefore allows the user to perform collection while checking that the measurement target S in the range specified as the collection target has been collected.

[0096] Furthermore, the learning model used in the analysis unit 220 may be implemented on rewritable hardware such as a field-programmable gate array (FPGA). For example, the network structure and weighting of the neural network DN, which is the learning model, may be implemented on hardware such as an FPGA. The neural network DN constructed using the technology disclosed herein has a relatively simple and small-scale structure, and therefore can be implemented on hardware such as an FPGA.

[0097] In such a case, the analysis unit 220 can estimate dimensionally reduced data corresponding to the multidimensional data faster than when software is executed by a calculation device such as a CPU. This allows the particle analysis device 200 to shorten the time from acquiring multidimensional data from the measurement object S to determining whether the measurement object S is a sample to be dispensed. Therefore, the particle analysis device 200 can more reliably dispense a measurement object S that has been determined to be a sample to be dispensed.

[0098] The determination unit 230 determines whether the measurement object S is a target for fractionation based on the dimensionality reduced data estimated by the analysis unit 220. Specifically, the determination unit 230 may determine that the measurement object S is a target for fractionation when the coordinates indicated by the dimensionality reduced data estimated by the analysis unit 220 are included in the range specified by the fractionation designation unit 170.

[0099] The sorting unit 240 separates the measurement target S determined by the determination unit 230 to be the measurement target S from the other measurement targets S, and collects them in a sorting well or tube. Specifically, the sorting unit 240 charges droplets containing the measurement target S determined to be the measurement target S, and passes the charged droplets between a pair of deflection plates to which a voltage is applied. This allows the sorting unit 240 to deflect and separate the droplets containing the measurement target S determined to be the measurement target S from the droplets containing the other measurement targets S by electrostatic attraction. The separated droplets containing the measurement target S are collected, for example, in a sorting well or tube.

[0100] In the particle analysis system 20 according to this embodiment, dimension-reduced data is estimated from multidimensional data using a learning model that has been machine-learned to learn a specific dimension reduction process. In this case, the particle analysis system 20 uses the same learning model in the analysis unit 150 of the information processing device 102 and the analysis unit 220 of the particle analysis device 200, and therefore, similar dimension-reduced data can be obtained. Therefore, the determination unit 230 can determine whether the measurement target S is a sample collection target based on whether the dimension-reduced data estimated by the analysis unit 220 is included in the range of sample collection targets set for the dimension-reduced data estimated by the analysis unit 150.

[0101] Furthermore, in the particle analysis system 20 according to this embodiment, the analysis units 150 and 220 use a learning model previously constructed by machine learning, so that dimensionally reduced data can be obtained quickly in a short time. Therefore, the particle analysis system 20 according to this embodiment can quickly determine whether the measurement target S is a sample to be collected, even within the time constraints from measurement to collection.

[0102] The particle analysis system 20 having the above configuration can more quickly separate a population of known measurement targets S due to the real-time nature of the calculations for estimating the dimension-reduced data. Furthermore, the particle analysis system 20 can quickly separate a population of unknown measurement targets S due to the real-time nature of the calculations for estimating the dimension-reduced data and the high reproducibility of the dimension-reduced data.

[0103] (3.2. Example of particle analysis system operation) Next, an example of the operation of the particle analysis system 20 according to this embodiment will be described with reference to Fig. 15. Fig. 15 is a flow chart showing an example of the flow of the operation of the particle analysis system 20 according to this embodiment.

[0104] As shown in Fig. 15, first, a sample is measured by an analytical device such as a flow cytometer 10 to acquire multidimensional data (S301). The acquired multidimensional data is data containing information about the fluorescence intensity measured by the photodetector 14 of the flow cytometer 10. The multidimensional data acquired by the analytical device is input to an information processing device via the input unit 110. 102 is entered into

[0105] Next, using the learning model stored in the learning model storage unit 140, the analysis unit 150 estimates dimension-reduced data corresponding to the multidimensional data (S302). The learning model is, for example, a neural network that includes an input layer, two or more intermediate layers, and an output layer, and has undergone machine learning of a dimension reduction process such as t-SNE or UMAP. For example, the analysis unit 150 may estimate the output of the output layer OL when the multidimensional data acquired in step S301 is input to the input layer IL of the learning model as dimension-reduced data corresponding to the multidimensional data.

[0106] The dimensionality reduced data estimated by the analysis unit 150 is presented to the user via the output unit 160. This allows the user to specify the coordinate range of the measurement target S to be fractionated based on the output dimensionality reduced data (S303). For example, the coordinate range of the measurement target S to be fractionated is input to the fractionation specifying unit 170, for example.

[0107] Thereafter, the measurement object S is measured again, and multidimensional data of the measurement object S is acquired via the input unit 210 (S304). Next, using the learning model stored in the learning model storage unit 140, the analysis unit 220 estimates dimensionally reduced data corresponding to the multidimensional data (S305).

[0108] Next, the determination unit 230 determines whether the dimensionally compressed data estimated by the analysis unit 220 is included in the coordinate range of the measurement object S to be fractionated (S306). If the dimensionally compressed data estimated by the analysis unit 220 is included in the coordinate range of the measurement object S to be fractionated (S306 / Yes), the determination unit 230 determines that the measurement object S is to be fractionated. As a result, the measurement object S determined to be to be fractionated is fractionated by the fractionation unit 240 (S307).

[0109] On the other hand, if the dimensionally compressed data estimated by the analysis unit 220 is not included in the coordinate range of the measurement target S to be collected (S306 / No), the determination unit 230 determines that the measurement target S is not a target for collection. As a result, the measurement target S determined not to be a target for collection is collected in a waste liquid tank or the like.

[0110] According to the above operation, the particle analysis system 20 according to this embodiment can precisely separate the measurement target S included in the range designated by the user as the separation target.

[0111] <4. Modifications> In the above-described embodiments, measurement data from a flow cytometer 10 is exemplified as multidimensional data, but the technology according to the present disclosure is not limited to this. The technology according to the present disclosure can also be applied to a fluorescence imaging device that measures multidimensional data such as fluorescence spectra using an image sensor (two-dimensional image sensor). In other words, the information processing device described in the above-described embodiments can also estimate dimension-reduced data corresponding to multidimensional data measured by a fluorescence imaging device.

[0112] An example of a schematic configuration of a fluorescence imaging device is shown in Fig. 16. Fig. 16 is a schematic diagram showing the schematic configuration of a fluorescence imaging device.

[0113] As shown in FIG. 16, a fluorescence imaging device 30 includes, for example, a laser light source 31, a movable stage 32 on which a fluorescently stained specimen 33 is placed, a spectroscopic unit 34, and an image sensor 35.

[0114] The laser light source 31 emits, for example, laser light of a wavelength capable of exciting the fluorescent dye used to stain the fluorescently stained specimen 33. When multiple fluorescent dyes are used to stain the fluorescently stained specimen 33, multiple laser light sources 31 may be provided so as to be able to emit laser light corresponding to the excitation wavelengths of the multiple fluorescent dyes. For example, the laser light source 31 may be a semiconductor laser light source. The laser light emitted from the laser light source 31 may be pulsed light or continuous light.

[0115] The movable stage 32 is a stage on which a measurement target such as a fluorescent stained specimen 33 is placed. The movable stage 32 can move horizontally so that the laser light emitted from the laser light source 31 scans the fluorescent stained specimen 33 two-dimensionally.

[0116] The fluorescent stained specimen 33 is a specimen prepared from a specimen or tissue sample collected from a human body and stained with multiple fluorescent dyes for purposes such as pathological diagnosis. The fluorescent stained specimen 33 includes a large number of measurement targets S, such as cells that make up the collected tissue. The fluorescent stained specimen 33 is moved horizontally on the movable stage 32, so that the laser light emitted from the laser light source 31 can be sequentially irradiated onto the large number of measurement targets S included in the fluorescent stained specimen 33.

[0117] The spectroscopic unit 34 is an optical element that separates the fluorescence emitted from the measurement target S irradiated with laser light into a continuous spectrum. The fluorescence separated by the spectroscopic unit 34 can be detected by the downstream image sensor 35. The spectroscopic unit 34 may be, for example, a prism or a grating.

[0118] Alternatively, the spectroscopic unit 34 may be an optical element that separates the fluorescence emitted from the measurement target S irradiated with laser light into light components in a predetermined detection wavelength range. In such a case, the spectroscopic unit 34 is configured to include, for example, at least one dichroic mirror or optical filter, and can separate the fluorescence from the measurement target S into light components in a predetermined detection wavelength range by optical members such as the dichroic mirror and the optical filter.

[0119] The imaging element 35 is a two-dimensional image sensor in which light receiving elements such as a CCD sensor or a CMOS sensor are arranged two-dimensionally.

[0120] The image sensor 35 outputs an image signal by receiving, with each of the two-dimensionally arranged light-receiving elements, the fluorescence emitted from the measurement object S included in the fluorescently stained specimen 33 and dispersed by the spectroscopic unit 34. Because the fluorescence emitted from the measurement object S irradiated with laser light is dispersed by the spectroscopic unit 34, the image sensor 35 can receive fluorescence in different wavelength ranges for each region and output an image signal according to the intensity of the received fluorescence.

[0121] In the fluorescence imaging device 30 having the above configuration, fluorescence emitted from the measurement target S contained in the fluorescently stained specimen 33 is dispersed by the spectroscopic section 34 and then detected by each of the light-receiving elements of the image sensor 35. Therefore, the image signal output by the image sensor 35 becomes multidimensional data. As a result, the information processing device described in each of the above embodiments can use a machine-learned learning model to estimate lower-dimensional dimension-reduced data corresponding to the multidimensional data measured by the fluorescence imaging device 30, similar to the measurement data of the above-mentioned flow cytometer 10.

[0122] The multidimensional data input to the information processing device according to each of the above-described embodiments may be image signals associated with position information acquired by the image sensor 35. Furthermore, when an image of the fluorescently stained specimen 33 is subjected to segmentation processing, the multidimensional data input to the information processing device may be image data associated with the region obtained by the segmentation processing.

[0123] The technology disclosed herein is not limited to fluorescence imaging devices that acquire fluorescence information, but can also be applied to general microscope devices that acquire images of biological specimens using image sensors. For example, when a biological specimen consisting of multiple sections is stained with hematoxylin eosin (HE) staining or immunohistochemical staining, and images of each stained section are acquired, image data acquired from the multiple images can be used as multidimensional data in association with positional information on the biological specimen. Furthermore, when multiple images of the biological specimen are segmented, image data associated with regions obtained by the segmentation process can also be used as multidimensional data.

[0124] <5. Hardware configuration example> Furthermore, the hardware configuration of the information processing devices 100, 101, and 102 according to this embodiment will be described with reference to Fig. 17. Fig. 17 is a block diagram showing an example of the hardware configuration of the information processing devices 100, 101, and 102 according to this embodiment.

[0125] The functions of the information processing devices 100, 101, and 102 according to the present embodiment are realized by cooperation between software and the hardware described below. For example, the functions of the dimensionality reduction unit 120, the learning unit 130, the analyzing unit 150, and the fraction designating unit 170 described above may be executed by the CPU 901.

[0126] As shown in FIG. 17, the information processing devices 100, 101, and 102 each include a CPU (Central Processing Unit) 901, a ROM (Read Only Memory) 902, and a RAM (Random Access Memory) 903.

[0127] The information processing devices 100, 101, and 102 may further include a host bus 904a, a bridge 904, an external bus 904b, an interface 905, an input device 906, an output device 907, a storage device 908, a drive 909, a connection port 910, and a communication device 911. Furthermore, the information processing devices 100, 101, and 102 may further include a DSP (Digital Signal Processor) or an ASIC (Application Specific Integrated Circuit) instead of or in addition to the CPU 901. The arithmetic processing circuit may include other processing circuits such as a arithmetic processing circuit.

[0128] The CPU 901 functions as an arithmetic processing device or control device, and controls the overall operation of the information processing devices 100, 101, and 102 in accordance with various programs recorded in the ROM 902, the RAM 903, the storage device 908, or a removable recording medium attached to the drive 909. The ROM 902 stores programs used by the CPU 901, calculation parameters, etc. The RAM 903 temporarily stores programs used in the execution of the CPU 901, and parameters used during the execution of the programs.

[0129] The CPU 901, ROM 902, and RAM 903 are interconnected by a host bus 904a, which is an internal bus such as a CPU bus. The host bus 904a is further connected to an external bus 904b, such as a PCI (Peripheral Component Interconnect / Interface) bus, via a bridge 904.

[0130] The input device 906 is a device that accepts input from a user, such as a mouse, keyboard, touch panel, button, switch, or lever. The input device 906 may be a microphone that detects the user's voice. The input device 906 may also be, for example, a remote control device that uses infrared rays or other radio waves, or may be an externally connected device that supports operation of the information processing devices 100, 101, and 102.

[0131] The input device 906 further includes an input control circuit that outputs an input signal generated based on information input by a user to the CPU 901. By operating the input device 906, a user can input various data or instruct the information processing devices 100, 101, and 102 to perform processing operations.

[0132] The output device 907 is a device capable of visually or audibly presenting information acquired or generated by the information processing devices 100, 101, and 102 to a user. The output device 907 may be, for example, a display device such as an LCD (Liquid Crystal Display), a PDP (Plasma Display Panel), an OLED (Organic Light Emitting Diode) display, a hologram, or a projector. The output device 907 may also be an audio output device such as a speaker or headphones, or a printing device such as a printer. The output device 907 may output information acquired by processing of the information processing devices 100, 101, and 102 as video such as text or images, or sound such as voice or audio. The output device 907 may function as the above-mentioned output unit 160, for example.

[0133] The storage device 908 is a data storage device configured as an example of a storage unit of the information processing devices 100, 101, and 102. The storage device 908 is, for example, a hard disk drive (HDD). The storage device 908 may be configured by a magnetic storage device such as a hard disk drive, a semiconductor storage device, an optical storage device, or a magneto-optical storage device. The storage device 908 can store programs executed by the CPU 901, various data, and various data acquired from the outside. The storage device 908 may function as the learning model storage unit 140 described above, for example.

[0134] The drive 909 is a device for reading or writing data from or to a removable recording medium such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, and is built into or externally attached to the information processing devices 100, 101, and 102. For example, the drive 909 can read information recorded on an attached removable recording medium and output the information to the RAM 903. The drive 909 can also write information to an attached removable recording medium.

[0135] The connection port 910 is a port for directly connecting an external device to the information processing devices 100, 101, 102. The connection port 910 may be, for example, a Universal Serial Bus (USB) port, an IEEE 1394 port, or a Small Computer System Interface (SCSI) port. The connection port 910 may also be an RS-232C port, an optical audio terminal, or a High-Definition Multimedia Interface (HDMI) (registered trademark) port. When connected to an external device, the connection port 910 enables transmission and reception of various types of data between the information processing devices 100, 101, 102 and the external device. The connection port 910 may function as, for example, the input unit 110 or the output unit 160 described above.

[0136] The communication device 911 is, for example, a communication interface configured with a communication device for connecting to the communication network 920. The communication device 911 may be, for example, a communication card for a wired or wireless LAN (Local Area Network), Bluetooth (registered trademark), or WUSB (Wireless USB). The communication device 911 may also be a router for optical communication, a router for ADSL (Asymmetric Digital Subscriber Line), or a modem for various types of communication. The communication device 911 may function as, for example, the input unit 110 or the output unit 160 described above.

[0137] The communication device 911 can transmit and receive signals, for example, via the Internet or other communication devices using a predetermined protocol such as TCP / IP. The communication network 920 connected to the communication device 911 is a wired or wireless network, and may be, for example, the Internet communication network, a home LAN, an infrared communication network, a radio wave communication network, or a satellite communication network.

[0138] It is also possible to create a program that causes hardware such as the CPU 901, ROM 902, and RAM 903 built into a computer to perform functions equivalent to those of the information processing devices 100, 101, and 102. It is also possible to provide a computer-readable recording medium on which the program is recorded.

[0139] Although the preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings, the technical scope of the present disclosure is not limited to such examples. It is clear that a person skilled in the art of the present disclosure can conceive of various modified or altered examples within the scope of the technical idea described in the claims, and it is understood that these also naturally fall within the technical scope of the present disclosure.

[0140] Furthermore, the effects described herein are merely descriptive or exemplary and are not limiting. In other words, the technology according to the present disclosure may achieve other effects that will be apparent to those skilled in the art from the description of this specification, in addition to or in place of the above-described effects.

[0141] The following configurations also fall within the technical scope of the present disclosure. (1) a particle analysis device including a light receiving unit that receives light emitted from the biological particles; an information processing device including: a dimensional compression unit that generates dimensionally compressed data by dimensionally compressing the multidimensional data related to the light output from the light receiving unit; and a learning unit that constructs a learning model by machine learning using a correspondence relationship between the dimensionally compressed data and the multidimensional data as a teacher; A particle analysis system, including: (2) The particle analysis system according to (1) above, wherein the dimensionality reduction unit reduces the dimension of the multidimensional data using a nonlinear method. (3) The particle analysis system according to (1) or (2), wherein the dimensionality reduction unit reduces the dimension of the multidimensional data to two or three dimensions. (4) The particle analysis system according to (3) above, wherein the dimensionally reduced data is coordinate data in a two-dimensional or three-dimensional coordinate space. (5) The particle analysis system according to any one of (1) to (4) above, wherein the learning unit constructs the learning model using a neural network including an input layer, two or more intermediate layers, and an output layer. (6) The particle analysis system according to (5) above, wherein the number of nodes in the input layer is the same as the number of dimensions of the multidimensional data, and the number of nodes in the output layer is the same as the number of dimensions of the dimensionally reduced data. (7) The particle analysis system according to (6) above, wherein the learning model is the neural network in which the multidimensional data is applied to the input layer and the dimensionality-reduced data is applied to the output layer to optimize the network structure and weighting. (8) The particle analysis system according to any one of (1) to (7), wherein the information processing device further includes an analysis unit that estimates the dimensionality-reduced data corresponding to the multidimensional data using the learning model constructed by the learning unit. (9) The particle analysis system according to any one of (1) to (8) above, wherein the multidimensional data includes data on the light intensity of the fluorescence emitted from the biogenic particles. (10) The particle analysis system according to any one of (1) to (9) above, wherein the particle analysis device further comprises a sorting unit that sorts the biological particles to be sorted. (11) The particle analysis system according to (10) above, wherein the particle analysis device further comprises a determination unit that determines whether the biogenic particles are the target of sorting based on the dimensionality-reduced data corresponding to the multidimensional data estimated by the learning model constructed by the learning unit. (12) The particle analysis system according to (11) above, wherein in the particle analysis device, the learning model constructed by the learning unit is implemented in hardware. (13) The information processing device includes: The particle analysis system according to any one of (1) to (12) above, further comprising an output unit that presents to a user the dimension-reduced data corresponding to the multidimensional data estimated by the learning model constructed by the learning unit. (14) a dimensional compression unit that generates dimensionally compressed data by dimensionally compressing multidimensional data related to light emitted from the biological particles; a learning unit that constructs a learning model by machine learning using the correspondence relationship between the dimensionality reduced data and the multidimensional data as a teacher; An information processing device comprising: (15) a light receiving unit that receives light emitted from the biological particles; a determination unit that determines whether the biogenic particles are to be sorted based on dimension-reduced data corresponding to the multidimensional data, the dimension-reduced data being estimated by a learning model that has been machine-learned to determine a correspondence relationship between multidimensional data related to light emitted from the biogenic particles and dimension-reduced data obtained by dimensionally reducing the multidimensional data; a sorting unit that sorts the biogenic particles determined to be sorted; A fraction collection device comprising: [Explanation of symbols]

[0142] 10 Flow cytometer 11 Laser light source 12 flow cells 13 Detection optics 14 Photodetector 20 Particle Analysis System 100, 101, 102 Information processing equipment 110 Input section 120 Dimensionality reduction section 130 Learning Department 140 Learning model memory unit 150 Analysis Department 160 Output section 170 Fractionation Designation Section 200 Particle analysis device 210 Input section 220 Analysis Department 230 Judgment section 240 Preparation section S Measurement target DN Neural Network IL input layer ML middle layer OL output layer

Claims

1. a particle analysis device including a light receiving unit that receives light emitted from the biological particles; an information processing device including: a dimensional compression unit that generates dimensionally compressed data by dimensionally compressing the multidimensional data related to the light output from the light receiving unit; a learning unit that constructs a learning model by machine learning using a correspondence relationship between the dimensionally compressed data and the multidimensional data as a teacher; and an analysis unit that uses the learning model constructed by the learning unit to estimate dimensionally compressed data corresponding to the multidimensional data newly output from the light receiving unit; A particle analysis system, including:

2. The particle analysis system according to claim 1 , wherein the dimensionality reduction unit reduces the dimensionality of the multidimensional data using a nonlinear method.

3. The particle analysis system according to claim 1 , wherein the dimensionality reduction unit reduces the dimension of the multidimensional data to two or three dimensions.

4. The particle analysis system according to claim 3 , wherein the dimensionally reduced data is coordinate data in a two-dimensional or three-dimensional coordinate space.

5. The particle analysis system according to claim 1 , wherein the learning unit constructs the learning model using a neural network including an input layer, two or more intermediate layers, and an output layer.

6. The particle analysis system according to claim 5 , wherein the number of nodes in the input layer is the same as the number of dimensions of the multidimensional data, and the number of nodes in the output layer is the same as the number of dimensions of the dimensionally reduced data.

7. The particle analysis system according to claim 6 , wherein the learning model is the neural network in which the multidimensional data is applied to the input layer and the dimensionality-reduced data is applied to the output layer to optimize the network structure and weighting.

8. The particle analysis system according to claim 1 , wherein the multidimensional data includes data relating to the intensity of fluorescence emitted from the biological particles.

9. The particle analysis system according to claim 1 , wherein the particle analysis device further comprises a sorting unit that sorts the biological particles to be sorted.

10. 10. The particle analysis system according to claim 9, wherein the particle analysis device further comprises a determination unit that determines whether the biogenic particles are the target of sorting based on the dimensionality-reduced data estimated by the learning model constructed by the learning unit.

11. The particle analysis system according to claim 10 , wherein in the particle analysis device, the learning model constructed by the learning unit is implemented in hardware.

12. The particle analysis system of claim 1, wherein the information processing device further includes an output unit that presents to a user the dimensional compression data estimated by the learning model constructed in the learning unit.

13. a dimensional compression unit that generates dimensionally compressed data by dimensionally compressing multidimensional data related to light emitted from the biological particles; a learning unit that constructs a learning model by machine learning using the correspondence relationship between the dimensionality reduced data and the multidimensional data as a teacher; an analysis unit that estimates dimension-reduced data corresponding to multidimensional data related to light emitted from the biogenic particles using the learning model constructed by the learning unit; An information processing device comprising:

14. a light receiving unit that receives light emitted from the biological particles; a determination unit that determines whether the biogenic particles are to be sorted based on dimensionality-reduced data estimated by a learning model in response to the multidimensional data related to the light output from the light-receiving unit; a sorting unit that sorts the biogenic particles determined to be sorted; Equipped with The learning model is a learning model that has been machine-learned in advance to determine the correspondence between multidimensional data regarding the light output from the light receiving unit and dimensionally compressed data that has been dimensionally compressed from the multidimensional data.

Images