Detection method, detection device, and program

Abstract

A detection method according to the present disclosure causes a computer to execute: a step for acquiring the physical properties of microorganisms to be detected; a step for inputting the physical properties of the microorganisms to be detected into a prediction model that has been subjected to a learning process using the physical properties of low-production microorganisms having a substance-producing ability less than a prescribed substance-producing ability, and acquiring an output from the prediction model; and a step for detecting the low-production microorganisms among the microorganisms to be detected on the basis of the output of the prediction model. The detection method of the present disclosure is also applicable to cases where a plant is the detection target.

Description

Detection Method, Detection Device, and Program

[0001] The present disclosure relates to a detection method, a detection device, and a program for detecting the production ability of substances possessed by a detection target.

[0002] As one of the technologies contributing to sustainable manufacturing, bio-manufacturing technology that uses materials derived from organisms such as plants for manufacturing, or further creates useful compounds by utilizing the abilities of organisms such as microorganisms, has attracted attention. Bio-manufacturing technology is a technology that uses various genetic manipulation technologies to produce useful substances in cells such as microorganisms, animals, and plants. Examples of bio-manufacturing technology methods include the random mutagenesis introduction method and the gene recombination technology. The random mutagenesis introduction method is a method of creating a library of mutant strains with different activities by randomly introducing mutations into the genes possessed by microorganisms. Gene recombination technology is a technology aimed at incorporating genes taken out from other organisms into the genes of the target organism and imparting some of the properties of other organisms to the target organism.

[0003] Mutant strains produced by bio-manufacturing technology may have different characteristics. In order to effectively utilize mutant strains, it is necessary to efficiently evaluate the characteristics possessed by the mutant strains.

[0004] As technologies for evaluating the characteristics possessed by microorganisms, there are, for example, technologies disclosed in Patent Documents 1, 2, 3, etc. Patent Documents 1, 2, 3 disclose technologies for creating a prediction model for predicting the characteristics of microorganisms using the reflection spectrum of microorganisms obtained by hyperspectral imaging as learning data, and identifying microorganisms using the prediction model.

[0005] Japanese Patent No. 6872566, Japanese Patent Publication No. 2021-506288, Japanese Patent Publication No. 2021-506286

[0006] As an example of the characteristics possessed by a microorganism, there is a substance production ability indicating whether the microorganism has the ability to produce a useful substance. Mutant strains produced by bio-manufacturing technology may include those having a substance production ability above a predetermined substance production ability.

[0007] When evaluating the substance production capacity of mutant strains using the techniques disclosed in Patent Documents 1, 2, and 3, the following problems may arise. In order to construct a predictive model for identifying microorganisms with superior substance production capacity using the techniques disclosed in Patent Documents 1, 2, and 3, it is necessary to prepare microorganisms with superior substance production capacity in advance. However, preparing microorganisms with superior substance production capacity in advance is extremely difficult. Furthermore, because the applicability of the predictive model is limited to the interpolation domain, the prediction accuracy for microorganisms with even superior substance production capacity decreases.

[0008] Furthermore, regarding plants that can produce useful substances, there is a need to accurately and efficiently identify plants with superior substance-producing capabilities.

[0009] In light of these circumstances, this disclosure aims to provide a detection method, a detection device, and a program that can detect the material production capacity of target microorganisms with high accuracy and reproducibility. Furthermore, this disclosure aims to provide a detection method, a detection device, and a program that can detect the material production capacity of target microorganisms or plants with high accuracy and reproducibility.

[0010] A detection method according to one aspect of the present disclosure involves a computer performing the following steps: acquiring the physical properties of a microorganism to be detected; inputting the physical properties of the microorganism to be detected into a prediction model that has been trained using the physical properties of low-productivity microorganisms having a material production capacity below a predetermined capacity, and obtaining the output from the prediction model; and detecting the low-productivity microorganisms among the microorganisms to be detected based on the output of the prediction model.

[0011] A detection device according to one aspect of the present disclosure includes: an acquisition unit that acquires the physical properties of a microorganism to be detected; an execution unit that inputs the acquired physical properties into a prediction model that has been trained using the physical properties of low-productivity microorganisms having a material production capacity below a predetermined capacity, and acquires an output from the prediction model; and a detection unit that detects the low-productivity microorganisms among the microorganisms to be detected based on the output.

[0012] A program according to one aspect of this disclosure causes a computer to execute the following steps: a step to acquire the physical properties of a microorganism to be detected; a step to input the acquired physical properties into a prediction model that has been trained using the physical properties of low-productivity microorganisms having a material production capacity below a predetermined capacity, and to obtain the output from the prediction model; and a step to perform detection to detect the low-productivity microorganisms among the microorganisms to be detected based on the output of the prediction model.

[0013] A detection method according to one aspect of the present disclosure involves a computer performing the following steps: acquiring the physical properties of a plant to be detected; inputting the physical properties of the plant to be detected into a prediction model that has been trained using the physical properties of low-productivity plants having a material production capacity below a predetermined capacity, and obtaining the output from the prediction model; and detecting the low-productivity plants among the plants to be detected based on the output of the prediction model.

[0014] A detection device according to one aspect of the present disclosure includes: an acquisition unit that acquires the physical properties of a plant to be detected; an execution unit that inputs the acquired physical properties to a prediction model that has been trained using the physical properties of low-productivity plants having a material production capacity below a predetermined capacity, and acquires an output from the prediction model; and a detection unit that detects the low-productivity plants among the plants to be detected based on the output.

[0015] A program according to one aspect of this disclosure causes a computer to perform the following steps: a step to acquire the physical properties of a plant to be detected; a step to input the acquired physical properties into a predictive model that has been trained using the physical properties of low-productivity plants having a material production capacity below a predetermined capacity, and to obtain the output from the predictive model; and a step to perform detection to detect the low-productivity plants among the plants to be detected based on the output of the predictive model.

[0016] A detection method according to one aspect of the present disclosure is a method in which a computer performs the following steps: acquiring the physical properties of a target to be detected, including microorganisms or plants; inputting the physical properties of the target to be detected into a predictive model that has been trained using the physical properties of a low-productivity target, which is a target to be detected having a material production capacity below a predetermined capacity, and obtaining the output from the predictive model; and detecting the low-productivity target among the target to be detected based on the output of the predictive model.

[0017] A detection device according to one aspect of the present disclosure includes: an acquisition unit that acquires the physical properties of a target to be detected, including microorganisms or plants; an execution unit that inputs the acquired physical properties to a prediction model that has been trained using the physical properties of a low-productivity target, which is a target to be detected having a material production capacity below a predetermined capacity, and acquires an output from the prediction model; and a detection unit that detects the low-productivity target among the targets to be detected based on the output.

[0018] A program according to one aspect of this disclosure causes a computer to perform the following steps: a step to acquire physical properties of a target to be detected, including microorganisms or plants; a step to input the acquired physical properties into a predictive model that has been trained using the physical properties of a low-productivity target, which is a target having a material production capacity below a predetermined capacity, and to obtain the output from the predictive model; and a step to perform detection to detect the low-productivity target among the targets to be detected, based on the output of the predictive model.

[0019] According to this disclosure, the ability to detect substance production capacity from target microorganisms can be performed with high accuracy and reproducibility. Furthermore, this disclosure enables the detection of substance production capacity from target microorganisms or plants with high accuracy and reproducibility.

[0020] Figure 1 is a diagram illustrating a detection system including a detection device according to the first embodiment of this disclosure. Figure 2A is a diagram illustrating an example of the configuration of a measuring device. Figure 2B is a diagram illustrating an example of the spectrum of light emitted by an LED as a light source. Figure 3 is a diagram illustrating an example of the functional configuration of a learning device. Figure 4A is a diagram illustrating an overview of k-NN as an example of an algorithm used in the learning process of a prediction model. Figure 4B is a diagram illustrating an overview of ABOD as an example of an algorithm used in the learning process of a prediction model. Figure 5 is a block diagram illustrating an example of the functional configuration of a detection device. Figure 6 is a flowchart illustrating an example of the operation of the detection device in the first embodiment. Figure 7 is a flowchart illustrating an example of the operation of the entire detection system in the first embodiment. Figure 8 is a diagram illustrating a detection system including a detection device in the second embodiment of this disclosure. Figure 9 is a flowchart illustrating an example of the operation of the detection device in the second embodiment. Figure 10 is a flowchart illustrating an example of the operation of the entire detection system in the second embodiment. Figure 11 is a diagram including a table showing the setting conditions for each embodiment. Figure 12 is a diagram including a table showing the evaluation results for each embodiment. Figure 13 shows an example of a computer hardware configuration as a learning device and detection device.

[0021] The embodiments of this disclosure will be described in detail below with reference to the drawings. However, unnecessary detailed explanations, such as detailed explanations of already well-known matters or redundant explanations of substantially identical configurations, may be omitted. However, the scope of this disclosure is not limited to those described in the following embodiments and drawings.

[0022] [First Embodiment] (Detection System) Figure 1 is a diagram illustrating a detection system including a detection device according to the first embodiment of the present disclosure. As shown in Figure 1, the detection system 100 comprises a measuring device 1, a learning device 2, and a detection device 3.

[0023] <Usage of the Detection System 100 in the First Embodiment> First, the usage of the detection system 100 in the first embodiment will be described. The user of the detection system 100 sets the type of microorganism to be detected by the detection system 100 and the substance produced by that microorganism. The user then applies bio-manufacturing technologies such as random mutagenesis or genetic recombination technology to the set microorganism to produce a microorganism with a substance production capacity below a predetermined capacity. In this specification, a microorganism with a substance production capacity below a predetermined capacity will be referred to as a low-productivity microorganism. An example of a low-productivity microorganism is a microorganism that has no substance production capacity at all.

[0024] Low-productivity microorganisms are cultured on a culture medium to form colonies, and their physical properties are measured using measuring device 1. Examples of physical properties include optical properties, more specifically, the reflectance spectrum obtained by irradiating the colonies with light. Using learning device 2, the reflectance spectra of low-productivity microorganisms are used as training data to construct a prediction algorithm for detecting low-productivity microorganisms. In the following description, microorganisms that are clearly produced by bio-manufacturing technology and have a material production capacity below a predetermined capacity are referred to as training microorganisms. The prediction algorithm takes the reflectance spectrum of a microorganism as input and outputs an anomaly score indicating the degree to which the microorganism is not a low-productivity microorganism. The prediction algorithm constructed using learning device 2 is introduced into detection device 3.

[0025] Next, the user uses bio-manufacturing technology to alter the base sequence of the genes of a designated microorganism (including deletions, insertions, mutations, etc.) to create multiple microorganisms of the same species, each potentially possessing different substance-producing capabilities. These multiple microorganisms are the target microorganisms for detection processing by detection device 3. In the following description, the multiple microorganisms of the same species, each potentially possessing different substance-producing capabilities, created using bio-manufacturing technology, will be referred to as the microorganisms to be detected.

[0026] Next, the reflectance spectrum of the target microorganism is measured using measuring device 1. Using detection device 3, a detection process is performed to detect low-productivity microorganisms and non-low-productivity microorganisms from among the target microorganisms based on their reflectance spectrum. This makes it possible to determine whether each of the target microorganisms is a low-productivity microorganism or a non-low-productivity microorganism.

[0027] Microorganisms that are not low-productivity microorganisms may include microorganisms with a material production capacity exceeding a predetermined capacity. By evaluating the material production capacity of microorganisms that are not low-productivity microorganisms detected by the detection device 3, useful microorganisms (those with high material production capacity) can be extracted from among multiple microorganisms produced by bio-manufacturing technology.

[0028] The effects obtained by using the detection system 100 in this manner are as follows: As described above, the detection process of the detection device 3 makes it easy to distinguish between low-productivity microorganisms and non-low-productivity microorganisms from among the microorganisms to be detected. Therefore, even if a large number of microorganisms are produced by bio-manufacturing technology, low-productivity microorganisms can be easily excluded from the evaluation target in advance by the detection device 3. This significantly reduces the time and effort required for evaluation.

[0029] Furthermore, as described above, the learning device 2 uses the reflectance spectra of microorganisms that are known to be low-productivity microorganisms as training data to train the prediction algorithm. Therefore, it becomes unnecessary to prepare microorganisms with a material production capacity above a certain level for training data, which is difficult to prepare in advance. Since low-productivity microorganisms are obtained when the host is set, it is also easy to prepare training data for low-productivity microorganisms.

[0030] In this disclosure, any microorganism capable of forming colonies on a culture medium can be suitably used as the learning microorganism and the microorganism to be detected. Specific examples include bacteria such as Escherichia coli, lactic acid bacteria, or actinomycetes, fungi such as yeast, or microalgae.

[0031] Furthermore, the useful substances produced by microorganisms in this disclosure can include any metabolites produced by microorganisms during cultivation. Specific examples include fatty acids, fluorescent proteins such as green fluorescent protein (GFP), dyes such as β-carotene, and polyesters such as polyhydroxybutyrate (PHB).

[0032] <Description of each component> The following describes each component of the detection system 100 according to the first embodiment.

[0033] <Measurement Device 1> Figure 2 is a diagram illustrating the measurement device 1. Figure 2A shows an example of the configuration of the measurement device 1. In the example shown in Figure 2A, the measurement device 1 includes a base 11, two light sources 12 and 13, and a camera 14.

[0034] The base 11 is a base on which Petri dish S is placed. Petri dish S contains a culture medium on which learning microorganisms or microorganisms to be detected have been cultured and colonies have formed.

[0035] Light sources 12 and 13 irradiate the petri dish S with light of a predetermined wavelength from different directions. Light sources 12 and 13 are, for example, LEDs (Light Emitting Diodes). Figure 2B shows an example of the spectrum of light emitted by LEDs as light sources 12 and 13. In the example shown in Figure 2, light sources 12 and 13 irradiate light with a wavelength of 400 nm to 1000 nm. Note that the wavelength of light emitted by the light source of the measuring device in this disclosure is not limited to this example and may be changed depending on, for example, the type of microorganism cultured in the petri dish S or the required accuracy of the test. Alternatively, a polarizing filter or the like may be provided in the measuring device 1 to generate S-polarized or P-polarized light and acquire optical properties in a crossed nicol state. In this case, specular reflected light from the colony can be removed and only scattered light from inside the colony can be acquired, which is preferable.

[0036] Furthermore, Figure 2A shows an example in which the measuring device 1 has two light sources 12 and 13, but the disclosure is not limited thereto. In this disclosure, the measuring device may have one or three or more light sources.

[0037] Camera 14 is a camera that spectrally separates light into wavelengths and takes images. Camera 14 is, for example, a hyperspectral camera or a multispectral camera. Light sources 12 and 13 are fixed in positions where light is suitably irradiated onto colonies formed on the culture medium in the petri dish S placed on the base 11. Camera 14 is also fixed in a position where it can suitably photograph the culture medium provided in the petri dish S and the colonies formed on the culture medium.

[0038] With this configuration, the measuring device 1 can suitably photograph the culture medium provided in the petri dish S and the colonies formed on the culture medium, and suitably measure the reflectance spectrum of microbial colonies formed on the culture medium. Furthermore, the change in the reflectance spectrum can be enhanced by adding dyes that change the photophysical properties in response to the physical properties of metabolites to the culture medium in advance. Examples include Nile red, Nile blue, methyl orange, bromothymol blue, 4-(4'-hydroxystyryl)-N-methylpyridinium iodide, and 2,6-diphenyl-4-(2,4,6-triphenyl-1-pyridinio)phenolate. In the following description, the reflectance spectrum of microbial colonies may be referred to as the microbial reflectance spectrum for simplicity.

[0039] <Learning Device 2> Learning device 2 is a computer, such as a PC. Learning device 2 realizes the various functional configurations described below by executing a program for training a predictive model.

[0040] Figure 3 shows an example of the functional configuration of the learning device 2. As shown in Figure 3, the learning device 2 comprises an acquisition unit 21, a learning target model storage unit 22, and a learning unit 23.

[0041] The acquisition unit 21 acquires the position information of the colonies formed on the agar medium of the learning microorganisms and the reflection spectra of the colonies from the measuring device 1. The position information is pixel position information in the image captured by the camera 14.

[0042] The learning target model storage unit 22 stores a prediction model (machine learning model) that is the target of the learning process in the learning device 2.

[0043] The learning unit 23 performs a learning process on the prediction model stored in the learning target model storage unit 22 using the reflection spectrum of the learning microorganisms acquired from the measuring device 1 as learning data. Since each colony in the captured image is composed of a plurality of pixels, the learning unit 23 performs the learning process for each colony using the average value of the pixels of the reflection spectrum of the learning microorganisms for each wavelength and the standard deviation of the pixels for each wavelength. Thereby, a prediction model that outputs the degree of abnormality indicating the degree to which the microorganism is not a low-producing microorganism when the position information and the reflection spectrum of the microbial colony are input is constructed. In the calculation of the average value and the standard deviation of the reflection spectrum of the learning microorganisms for each wavelength, there is no restriction on the number of pixels constituting the colony, but if it is less than 10 pixels, the influence of the variation between pixels and noise becomes large, so it is preferably 10 pixels or more.

[0044] As the algorithm used by the learning unit 23 for the learning process of the prediction model, an algorithm generally treated as an anomaly detection algorithm may be appropriately adopted. Examples of the anomaly detection algorithm include k-NN (k-Nearest Neighbors) or ABOD (Angle Based Outlier Detection).

[0045] FIG. 4 is a diagram for explaining the algorithm used for the learning process of the prediction model.

[0046] Figure 4A shows an overview of k-NN. In k-NN, a circle is drawn to include k data points in the neighborhood of a certain data point, and the radius (distance) of the drawn circle is regarded as the degree of abnormality of that data point. k-NN sets the degree of abnormality based on the idea that the radius of the circle drawn around a normal data point is smaller than the radius of the circle drawn around an abnormal data point. When k-NN is adopted, the reflection spectra of the learning microorganisms, which are the learning data, are data points with close distances to each other. By learning this as the correct answer, data points with a far distance from others can be detected as points with a high degree of abnormality.

[0047] Figure 4B shows an overview of ABOD. In ABOD, the cosine similarity between two data points is regarded as the degree of abnormality. ABOD utilizes the fact that the angle formed by two straight lines connecting a certain data point and two other data points is smaller in the case of an abnormal data point than in the case of a normal data point.

[0048] In the present disclosure, the algorithms used when the learning device performs learning of the prediction model are not limited to these, and other algorithms may be used as appropriate.

[0049] <Detection device 3> (Configuration) The detection device 3 is a computer such as a PC (Personal Computer). The detection device 3 realizes various functional configurations described below by executing a program according to the detection method of the present disclosure.

[0050] Figure 5 is a block diagram showing an example of the functional configuration of the detection device 3. As shown in Figure 5, the detection device 3 includes an acquisition unit 31, an execution unit 32, a detection unit 33, and a result output unit 34. With these configurations, the detection process of the detection device 3 is executed.

[0051] The acquisition unit 31 acquires the position information (pixel information) of the colonies formed on the microbial agar medium to be detected and the reflection spectra of the colonies from the measuring device 1. When the microorganism to be detected includes a plurality of microorganisms, the acquisition unit 31 acquires the position information and reflection spectra for each colony.

[0052] The execution unit 32 inputs the location information and reflection spectrum for each colony acquired by the acquisition unit 31 into the prediction model and obtains the output from the prediction model. When the execution unit 32 uses the prediction model, the prediction model may be stored in a memory unit (not shown) of the detection device 3, or it may be used by communicating with an external memory device, for example, and using the prediction model stored in said memory device. In the latter case, the prediction model may be stored in a cloud server, for example.

[0053] As described above, the prediction model is trained by the learning device 2 to use the location information and reflectance spectrum (physical properties) of low-productivity microbial colonies as training data (ground truth data), so that when the location information and reflectance spectrum of a microbial colony are input, it outputs an abnormality score indicating the degree to which the microorganism is not a low-productivity microorganism. Therefore, the execution unit 32 inputs the location information and reflectance spectrum of the microbial colony to be detected into the prediction model and obtains the abnormality score of the microorganism to be detected as an output.

[0054] The detection unit 33 detects low-productivity microorganisms and non-low-productivity microorganisms among the target microorganisms based on the anomaly score, which is the output obtained from the prediction model by the execution unit 32. Specifically, the detection unit 33 detects microorganisms whose anomaly score is below a predetermined threshold as low-productivity microorganisms. The detection unit 33 also detects microorganisms whose anomaly score is above a predetermined threshold as non-low-productivity microorganisms.

[0055] The predetermined threshold used to determine the degree of anomaly is a value calculated based on the training data. More specifically, the predetermined threshold is a value used to determine whether a microorganism is low-productivity or not, and is calculated and set for each training data set. The method for calculating the threshold may be set appropriately by the algorithm used in the training process. As an example of a method for calculating the threshold, one method is to assume that the data set follows a normal distribution and set the threshold to the mean + 3 × standard deviation, i.e., the 99.7% confidence interval in a normal distribution. Alternatively, one method is to set an arbitrary contamination coefficient and set the threshold by determining the proportion of data from the training data that should be considered outliers. In this disclosure, the method for calculating the threshold is not limited to the examples described above, and other methods may be adopted as appropriate.

[0056] The result output unit 34 outputs the detection results for microorganisms that are not low-productivity microorganisms among the microorganisms to be detected. The result output unit 34 can output and display the detection results to, for example, a display unit on the detection device 3 or a display device provided outside the detection device 3. Alternatively, the detection results may be output as an image or the like using the colony location information acquired by the acquisition unit 31. In this case, it is preferable because the microbial colonies that are not low-productivity microorganisms can be visually identified, and thus non-low-productivity microorganisms can be acquired more efficiently.

[0057] Thus, the detection device 3 can output as a detection result microbial colonies that are not low-productivity microorganisms among the multiple microbial colonies to be detected. As mentioned above, it is expected that the microbial colonies that are not low-productivity microorganisms among the multiple microbial colonies to be detected will include microorganisms with a material production capacity of a predetermined level or higher. As a result, when a variety of microorganisms are produced by bio-manufacturing technology and it becomes necessary to evaluate them, the target of evaluation can be easily narrowed down, and the time and effort required for evaluation can be reduced.

[0058] (Detection Process) An example of the operation of the detection device 3 in the first embodiment will be described. Figure 6 is a flowchart illustrating an example of the operation of the detection device 3 in the first embodiment.

[0059] In step S1, the detection device 3 obtains the reflectance spectrum of the target microbial colony from the measuring device 1.

[0060] In step S2, the detection device 3 inputs the reflectance spectrum of the microbial colony to be detected into the prediction model and obtains the output from the prediction model. As described above, the prediction model is trained to output an anomaly score indicating the degree to which the microorganism is not a low-productivity microorganism when the reflectance spectrum of the microbial colony is input. Therefore, in step S2, the detection device 3 obtains the anomaly score of the microbial colony to be detected as the output from the prediction model.

[0061] In step S3, the detection device 3 detects low-productivity microorganisms and non-low-productivity microorganisms from among the target microbial colonies based on the anomaly score, which is the output from the prediction model. More specifically, in step S3, the detection device 3 detects microbial colonies with an anomaly score below a predetermined threshold as low-productivity microorganisms, and detects microbial colonies with an anomaly score equal to or greater than the threshold as non-low-productivity microorganisms.

[0062] In step S4, the detection device 3 outputs as detection results microbial colonies that are not low-productivity microorganisms among the microbial colonies to be detected.

[0063] The detection process, including steps S1 to S4 described above, allows for the output of microbial colonies that are not low-productivity microorganisms among the target microbial colonies as the detection result.

[0064] <Example of Detection System Operation> An example of the operation of the entire detection system 100 will be described. Figure 7 is a flowchart illustrating an example of the operation of the entire detection system 100 in the first embodiment.

[0065] In step S11, the measuring device 1 uses a petri dish in which colonies of learning microorganisms (low-productivity microorganisms) for training a predictive model have formed, and measures the reflectance spectrum of the culture medium in the petri dish and the colonies formed on the culture medium.

[0066] In step S12, the learning device 2 acquires the positional information and the reflectance spectrum of the learning microorganism colonies measured in step S11.

[0067] In step S13, the learning device 2 uses the reflection spectra of the learning microbial colonies acquired in step S12 as learning data to perform a learning process on the predictive model so that it can accurately detect low-productivity microbial colonies and microbial colonies that are not low-productivity.

[0068] In step S14, the detection device 3 stores the prediction model that was trained in step S13.

[0069] In step S15, the measuring device 1 measures the positional information of the culture medium in the petri dish in which the microorganism to be detected is cultured, as well as the reflectance spectrum of the microbial colonies formed on the culture medium.

[0070] In step S16, the detection device 3 acquires the positional information and the reflectance spectrum of the target microbial colony, which were measured in step S15.

[0071] In step S17, the detection device 3 uses the reflectance spectrum of the target microbial colony acquired in step S16 to detect low-productivity microorganisms and non-low-productivity microbial colonies from among the target microbial colonies, and outputs the non-low-productivity microbial colonies as the detection result.

[0072] Through this operation, the detection system 100 can accurately detect microbial colonies that are not low-productivity microorganisms from among the target microbial colonies.

[0073] [Second Embodiment] A second embodiment of the present disclosure will be described below. The detection system 100A according to the second embodiment can detect plants. Figure 8 is a diagram illustrating a detection system including a detection device according to the second embodiment of the present disclosure. The detection system 100A according to the second embodiment includes a measuring device 1A, a learning device 2A, and a detection device 3A, similar to the first embodiment described above. Below, the differences between the detection system 100A according to the second embodiment and the detection system 100 according to the first embodiment will be mainly described.

[0074] In this disclosure, "plant" refers to a photosynthetic eukaryote belonging to the kingdom Plantae, whether unicellular or multicellular. In this disclosure, "plant" includes the whole plant, plant-derived cells, or tissue cultures. More specifically, in this disclosure, "plant" includes the whole plant, plant components, or organs (such as leaves, stems, and roots), plant tissues, seeds, plant cells, protoplasts, and their offspring. Offspring may include any generation. Plant cells include biological cells isolated from a plant, or cells obtained by culturing cells isolated from a plant. Furthermore, in this disclosure, "plant" includes monocots and dicots.

[0075] The plants in this disclosure may include food crops (cereals and pasture grasses, legumes and fodder beans, tuber crops (such as potatoes), fruit trees and nuts, and vegetables), as well as industrial crops (oilseed crops, fiber crops, and sugar crops and starch crops). Other plantation crops, ornamental plants, turfgrasses, flavor crops, flowers, and forest trees may also be included. The following are some representative plant species, but are not limited to these: Cereals and pasture grasses: alfalfa (Medicago sativa), rice (Oryza sativa), maize (Zea mays), wheat (Triticum genus), barley (Hordeum vulgare), oats (Avena sativa), sorghum (Sorghum bicolor), pearl millet (Cenchrus americanus = Pennisetum glaucum), finger millet (Eleusine coracana), cold-climate pasture grasses (Festuca pratensis, Lolium perenne, etc.), bahia grass (Paspalum notatum), etc. - Legumes and feed beans: Kidney beans (Phaseolus vulgaris), cowpeas (Vigna unguiculata), peas (Pisum sativum), broad beans (Vicia faba), lentils (Lens culinaris), teparii beans (Phaseolus acutifolius), pigeon peas (Vigna aconitifolia), chickpeas (Cicer arietinum), lupines (Lupinus genus), alfalfa (Medicago sativa), clover (Trifolium genus), etc. - Tuberous crops (potatoes, etc.): Beets (Beta vulgaris), parsnips (Pastinaca sativa), potatoes (Solanum tuberosum), turnips (Brassica rapa subsp. rapa), sweet potatoes (Ipomoea batatas), etc.Fruit trees and nuts: Apples (Malus domestica), pears (Pyrus pyrifolia or Pyrus communis), peaches (Prunus persica), plums (Prunus salicina), berries (Rubus fruticosus, Fragaria × ananassa, Vaccinium corymbosum, etc.), cherries (Prunus avium), grapes (Vitis vinifera), olives (Olea europaea), almonds (Prunus dulcis), walnuts (Juglans regia), etc. Citrus fruits (Citrus species such as lime, orange, and grapefruit), bananas (Musa acuminata, Musa × paradisiaca), plantains (Musa × paradisiaca), pineapples (Ananas comosus), papayas (Carica papaya), mangoes (Mangifera indica), avocados (Persea americana), kiwifruit (Actinidia deliciosa), passion fruit (Passiflora edulis), and persimmons (Diospyros kaki), etc. Vegetable crops: Solanaceae plants (tomato (Solanum lycopersicum), eggplant (Solanum melongena), chili pepper (Capsicum annuum), etc.), Brassicaceae vegetables (Brassica genus: Chinese cabbage, cabbage, komatsuna, etc.), radish (Raphanus sativus), carrot (Daucus carota subsp. sativus), Cucurbitaceae plants (Cucumis sativus, Cucurbita pepo, Citrullus lanatus, etc.), Allium genus plants (Allium cepa, Allium fistulosum, etc.), asparagus (Asparagus officinalis), leafy vegetables (Lactuca sativa, Spinacia oleracea, etc.).- Oilseed crops: Soybeans (Glycine max), Brassicaceae oilseed crops (Brassica napus, Brassica rapa, Brassica juncea, Brassica carinata, etc.), sunflower (Helianthus annuus), peanut (Arachis hypogaea), flax (Linum usitatissimum), sesame (Sesamum indicum), safflower (Carthamus tinctorius), etc. - Fiber crops, sugar crops and starch crops: Sugarcane (Saccharum officinarum), sugar beet (Beta vulgaris subsp. vulgaris var. altissima), stevia (Stevia rebaudiana), cassava (Manihot esculenta), cotton (Gossypium hirsutum), etc. Others: Tobacco (Nicotiana tabacum), coffee (Coffea arabica), cacao (Theobroma cacao), tea (Camellia sinensis), rubber tree (Hevea brasiliensis), medicinal plants (Digitalis purpurea, Panax ginseng, etc.), turfgrass (Zoysia japonica, Poa pratensis, etc.). Forest tree species: Japanese cedar (Cryptomeria japonica), Japanese cypress (Chamaecyparis obtusa), Japanese pine (Pinus densiflora), etc.

[0076] <Usage of Detection System 100A in the Second Embodiment> The usage of the detection system 100A in the second embodiment will be described. The user of the detection system 100A sets the type of plant to be detected by the detection system 100A and the substance that the plant produces. The user then applies bio-manufacturing technologies such as random mutation introduction or genetic modification technology to the set plant to produce a plant with a substance production capacity below a predetermined capacity. In this specification, a plant with a substance production capacity below a predetermined capacity is referred to as a low-productivity plant. An example of a low-productivity plant is a plant that has no substance production capacity at all.

[0077] The entire plant or a part of it (leaves, seeds, cells, etc.) is placed on a solid material such as an agar medium, and the physical properties of the low-yielding plant are measured using the measuring device 1A. Examples of physical properties include photophysical properties, more specifically, the reflectance spectrum obtained by irradiating the entire plant or a part of it with light. In the second embodiment, the physical properties may be measured after processing the entire plant or a part of it. Examples of processing include processing the dried plant or a part of it into a powder. In the following description, the reflectance spectrum obtained by irradiating the entire plant, a part of it, or a processed version thereof with light may be referred to as the plant's reflectance spectrum.

[0078] Next, the learning device 2A is used to train a predictive algorithm for detecting low-productivity plants by using the reflectance spectra of low-productivity plants as training data. In the following description, plants that are clearly produced using bio-manufacturing technology and have a material production capacity below a predetermined capacity are referred to as learning plants. The predictive algorithm in the second embodiment is an algorithm that takes the reflectance spectrum of a plant as input and outputs an anomaly score indicating the degree to which the plant is not a low-productivity plant. The predictive algorithm constructed using the learning device 2A is introduced into the detection device 3A.

[0079] Next, the user uses bio-manufacturing technology to alter the base sequence of a set plant's gene (including deletions, insertions, mutations, etc.) to create multiple plants of the same species, each potentially possessing different substance-producing capabilities. These multiple plants are the plants targeted for detection by detection device 3A. In the following description, the multiple plants of the same species, each potentially possessing different substance-producing capabilities, created using bio-manufacturing technology, will be referred to as the plants to be detected.

[0080] Next, the reflectance spectrum of the target plant is measured using measuring device 1A. Using detection device 3A, a detection process is performed to detect low-productivity plants and non-low-productivity plants from among the target plants based on the reflectance spectrum of the target plant. This makes it possible to accurately determine whether each target plant is a low-productivity plant or a non-low-productivity plant.

[0081] Plants that are not low-productivity plants may include plants with a material production capacity exceeding a predetermined limit. By evaluating the material production capacity of plants that are not low-productivity plants detected by the detection device 3A, useful plants (those with high material production capacity) can be efficiently extracted from among multiple plants produced by bio-manufacturing technology.

[0082] The effects obtained by using the detection system 100A in this manner are as follows: As described above, the detection process of the detection device 3A makes it easy to distinguish between low-yielding plants and non-low-yielding plants from among the plants to be detected. Therefore, even if a large number of plants are produced using bio-manufacturing technology, the detection device 3A can exclude low-yielding plants from the evaluation beforehand, and only evaluation of non-low-yielding plants needs to be performed. This significantly reduces the time and effort required for evaluation.

[0083] Furthermore, as described above, the learning device 2A uses the reflectance spectra of plants that are known to be low-productivity plants as training data to train the prediction algorithm. Therefore, it becomes unnecessary to prepare plants with a material production capacity above a certain level for use as training data, which is difficult to do in advance.

[0084] <Explanation of Each Configuration> The differences between each configuration of the detection system 100A according to the second embodiment and the first embodiment will be explained.

[0085] <Measuring device 1A> The measuring device 1A according to the second embodiment has the same configuration as the measuring device 1 according to the first embodiment (see Figure 2).

[0086] However, the petri dish S used in the measuring device 1A contains a culture medium in which a learning plant, or the whole, a part, or a processed version thereof of the plant to be detected, is placed.

[0087] <Learning Device 2A> The learning device 2A according to the second embodiment has the same configuration as the learning device 2 according to the first embodiment (see Figure 3). However, the learning device 2A differs from the learning device 2 according to the first embodiment in that it performs learning processing on plants to be detected instead of microorganisms to be detected.

[0088] <Detection device 3A> The detection device 3A according to the second embodiment has the same configuration as the detection device 3 according to the first embodiment (see Figure 5). However, the detection device 3A differs from the detection device 3 according to the first embodiment in that it performs detection processing on plants instead of microorganisms.

[0089] (Detection Process) An example of the operation of the detection device 3A in the second embodiment will be described. Figure 9 is a flowchart illustrating an example of the operation of the detection device 3A in the second embodiment.

[0090] In step S21, the detection device 3A acquires the reflection spectrum of the object to be detected from the measuring device 1A.

[0091] In step S22, the detection device 3A inputs the reflectance spectrum of the plant to be detected into the prediction model and obtains the output from the prediction model. As described above, the prediction model in the second embodiment is trained to output an abnormality score indicating the degree to which the plant is not a low-yielding plant when the reflectance spectrum of the plant to be detected is input. Therefore, in step S22, the detection device 3A obtains the abnormality score of the plant to be detected as the output from the prediction model.

[0092] In step S23, the detection device 3A detects low-productivity plants and non-low-productivity plants from among the plants to be detected, based on the abnormality score which is the output from the prediction model. More specifically, in step S23, the detection device 3A detects plants whose abnormality score is below a predetermined threshold as low-productivity plants, and plants whose abnormality score is equal to or greater than the threshold as non-low-productivity plants.

[0093] In step S24, the detection device 3A outputs the plants that are not low-yielding plants among the plants to be detected as detection results.

[0094] The detection process, including steps S21 to S24 described above, allows for the output of plants that are not low-yielding plants among the target plants as detection results.

[0095] <Example of Detection System Operation> An example of the overall operation of the detection system 100A according to the second embodiment will be described. Figure 10 is a flowchart for explaining the overall operation example of the detection system 100A according to the second embodiment.

[0096] In step S31, the measuring device 1A uses a petri dish in which a learning plant (low-yielding plant) for training a predictive model is placed on the culture medium, and measures the reflectance spectra of the culture medium in the petri dish and the plant placed on the culture medium.

[0097] In step S32, the learning device 2A acquires the positional information and reflectance spectrum of the learning plant measured in step S31.

[0098] In step S33, the learning device 2A uses the reflection spectra of the training plants acquired in step S32 as training data to perform training processing on the prediction model so that it can accurately detect low-yielding plants and plants that are not low-yielding plants.

[0099] In step S34, the detection device 3A stores the prediction model that was trained in step S33.

[0100] In step S35, the measuring device 1A measures the positional information and reflectance spectrum of the culture medium in the petri dish and the plant to be detected placed on the culture medium.

[0101] In step S36, the detection device 3A acquires the positional information and reflectance spectrum of the plant to be detected, which were measured in step S35.

[0102] In step S37, the detection device 3A uses the reflectance spectrum of the target plant acquired in step S36 to detect low-yielding plants and non-low-yielding plants from among the target plants, and outputs the non-low-yielding plants as the detection result.

[0103] Through this operation, the detection system 100A can accurately detect plants that are not low-yielding plants from among the plants being detected.

[0104] <Examples> Below, we will describe the results of evaluating the detection results when various microorganisms or plants were targeted, as examples of the detection systems 100 and 100A according to the first and second embodiments of this disclosure. Figure 11 is a diagram including a table showing the setting conditions for each example. Figure 12 is a diagram including a table showing the evaluation results for each example. Note that the black cells in the table in Figure 11 indicate setting conditions that are different from those of Example 1.

[0105] (Example 1) In Example 1, Escherichia coli with or without the ability to produce fatty acids were used as the target microorganisms for detection and the training microorganisms, and a training process for a prediction model and a detection process using the trained prediction model were performed.

[0106] In Example 1, E. coli cells introduced with an empty vector were inoculated onto an agar plate as training microorganisms, and approximately 100 colonies were formed by culturing at 37°C for 17 hours. Subsequently, the colonies were irradiated from two directions using light in the wavelength range of 400 nm to 1000 nm in the measurement device 1, and the reflectance spectra of the colonies were measured using a hyperspectral camera. In the learning device 2, the mean and standard deviation of the measured reflectance spectra for each wavelength were used as training data to train a prediction model. ABOD was adopted as the learning algorithm.

[0107] To evaluate the performance of the detection device 3, which performs detection processing using the predictive model that has undergone the learning process described above, the following was performed: E. coli strains known to produce a predetermined amount or more of fatty acids and E. coli strains known to produce less than a predetermined amount of fatty acids were inoculated into different regions on an agar plate. Subsequently, multiple colonies were formed by culturing at 37 degrees Celsius for 17 hours. In the measurement device 1, the colonies were irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the colonies was measured using a hyperspectral camera. In the detection device 3, the detection process was performed by inputting the obtained reflectance spectrum into the predictive model described above.

[0108] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in detection device 3. In the evaluation test, it was assessed whether E. coli, whose fatty acid production was known to be above a predetermined amount, was detected by detection device 3 as a non-low-productive microorganism. In addition, the evaluation test was assessed whether E. coli, whose fatty acid production was known to be below a predetermined amount, was detected by detection device 3 as a low-productive microorganism.

[0109] The Nile Red test was used as the evaluation method. Specifically, E. coli detected by detection device 3 as low-productivity microorganisms and E. coli detected as non-low-productivity microorganisms were cultured in liquid. 10 μL of each pre-culture solution was dropped onto different areas of Nile Red-containing agar medium and cultured at 37°C for 20 hours. Subsequently, the fluorescence emission of the colonies formed by the culture was observed under excitation light of 470 nm. Whether or not a colony was a low-productivity microorganism was evaluated by whether the intensity of the fluorescence emission from the colony was stronger than the fluorescence emission from E. coli colonies whose fatty acid production was known to be below a predetermined amount.

[0110] The results of the evaluation test in Example 1 showed an accuracy rate of 70% or more and less than 90%, a recall rate of 90% or more, and a precision rate of 70% or more and less than 90%. The accuracy rate is the percentage of all detection results from the detection device 3 that include a correct detection result. The recall rate is the percentage of E. coli that are known to produce a predetermined amount or more of fatty acids that are correctly detected by the detection device 3. The precision rate is the percentage of microorganisms detected by the detection device 3 as not being low-productivity microorganisms that are known to produce a predetermined amount or more of fatty acids that are E. coli.

[0111] This confirmed that, in Example 1, the detection device 3 had a detection accuracy of a certain degree or higher.

[0112] (Example 2) In Example 2, the learning process for the prediction model was performed in the learning device 2 using the same procedure as in Example 1, except that the algorithm was changed from ABOD to k-NN. Then, in Example 2, the detection process was performed in the detection device 3 using the same procedure as in Example 1, and further evaluation tests of the detection results were performed using the same procedure as in Example 1.

[0113] The results of the evaluation test in Example 2 showed an accuracy rate of 70% or more and less than 90%, a recall rate of 70% or more and less than 90%, and a precision rate of 90% or more.

[0114] This confirmed that even when the learning algorithm is set to k-NN, the detection device 3 has the same detection accuracy as in Example 1.

[0115] (Example 3) In Example 3, the learning process for the prediction model was performed in the learning device 2 using the same procedure as in Example 1, except that the camera 14 (see Figure 2) in the measuring device 1 was changed from a hyperspectral camera to a multispectral camera. Then, in Example 3, the detection process in the detection device 3 was performed using the prediction model using the same procedure as in Example 1, and further evaluation tests of the detection results were performed using the same procedure as in Example 1.

[0116] The results of the evaluation test in Example 3 showed an accuracy rate of 60% or more but less than 70%, a recall rate of 70% or more but less than 90%, and a precision rate of 60% or more but less than 70%.

[0117] This confirmed that, in order to improve detection accuracy, a hyperspectral camera is preferable to a multispectral camera in the measurement device 1.

[0118] (Example 4) In Example 4, the learning process for the prediction model was performed in the learning device 2 using the same procedure as in Example 1, except that the two light sources 12 and 13 (see Figure 2) in the measuring device 1 were changed to one light source. Then, in Example 4, the detection process in the detection device 3 was performed using the prediction model using the same procedure as in Example 1, and further evaluation tests of the detection results were performed using the same procedure as in Example 1.

[0119] The results of the evaluation test in Example 4 showed an accuracy rate of 70% or more and less than 90%, a recall rate of 60% or more and less than 70%, and a precision rate of 70% or more and less than 90%.

[0120] This confirmed that, in order to improve detection accuracy, it is preferable to have two lights rather than one when measuring with measuring device 1.

[0121] (Example 5) In Example 5, the learning process for the prediction model was performed in the learning device 2 using the same procedure as in Example 1, except that the standard deviation for each wavelength of the reflection spectrum was not used as learning data during the learning process in the learning device 2. Then, in Example 5, the detection process in the detection device 3 was performed using the prediction model using the same procedure as in Example 1, and an evaluation test of the detection results was performed using the same procedure as in Example 1.

[0122] The results of the evaluation test in Example 5 showed that the accuracy rate was 70% or more and less than 90%, the recall rate was 70% or more and less than 90%, and the precision rate was 60% or more and less than 70%.

[0123] This confirmed that, in order to improve detection accuracy, it is preferable to use not only the mean but also the standard deviation in the training data for the learning process in the learning device 2.

[0124] (Example 6) In Example 6, the learning process for the predictive model was performed in the learning device 2 in the same procedure as in Example 1, except that the number of colonies of learning microorganisms used for learning the predictive model in the learning device 2 was changed from 100 to 200. Then, in Example 6, the detection process in the detection device 3 was performed using the predictive model in the same procedure as in Example 1, and further evaluation tests of the detection results were performed in the same procedure as in Example 1.

[0125] The results of the evaluation test in Example 6 showed an accuracy rate of 90% or higher, a recall rate of 90% or higher, and a precision rate of 70% or higher but less than 90%.

[0126] This confirmed that, in order to improve detection accuracy, it is preferable to have 200 colonies of learning microorganisms used in the learning process in the learning device 2 rather than 100.

[0127] (Example 7) In Example 7, the learning process for the prediction model was performed in the learning device 2 using the same procedure as in Example 1, except that the learning data from which outliers had been removed was used in the learning process in the learning device 2. Then, in Example 7, the detection process in the detection device 3 was performed using the prediction model using the same procedure as in Example 1, and an evaluation test of the detection results was performed using the same procedure as in Example 1. Specifically, outliers were first calculated using k-NN with the learning data, and the reflectance spectra of colonies whose outliers were larger than the third quartile of the outlier + 1.5 × IQR (interquartile range) were removed as outliers, and the remaining reflectance spectra were used again as the learning data set.

[0128] The results of the evaluation test in Example 7 showed that the accuracy, recall, and precision were all 90% or higher.

[0129] This confirmed that, in order to improve detection accuracy, it is preferable to use training data from which outliers have been removed during the training process in the learning device 2.

[0130] (Example 8) In Example 8, Escherichia coli with or without the ability to produce GFP (green fluorescent protein) was used as the target microorganism for detection and the learning microorganism for training. The training process for the prediction model and the detection process using the prediction model trained through the training process were performed.

[0131] The learning process in Example 8 is the same as in Example 1.

[0132] In Example 8, to evaluate the prediction model, E. coli strains known to produce GFP above a predetermined amount and E. coli strains known to produce GFP below a predetermined amount were inoculated into different regions on an agar plate. Multiple colonies were then formed by culturing at 37°C for 17 hours. In the measuring device 1, the colonies were irradiated from two directions using light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectra of the colonies were measured using a hyperspectral camera. In the detection device 3, the obtained reflectance spectra were input into the prediction model described above for detection processing.

[0133] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in detection device 3. In the evaluation test, it was assessed whether E. coli, whose GFP production was known to be above a predetermined amount, was detected by detection device 3 as a non-low-productive microorganism. In addition, the evaluation test was assessed whether E. coli, whose GFP production was known to be below a predetermined amount, was detected by detection device 3 as a low-productive microorganism.

[0134] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in detection device 3. The evaluation test was performed by fluorescence observation of GFP in each colony.

[0135] The results of the evaluation test in Example 8 showed that the accuracy, recall, and precision were all 90% or higher.

[0136] This confirmed that the detection device 3 can output highly accurate detection results even when E. coli with or without the ability to produce GFP are used as the target microorganism and the learning microorganism.

[0137] (Example 9) In Example 9, yeast having or not having the ability to produce β-carotene was used as the target microorganism for detection and the learning microorganism for learning. The process involved training a prediction model and performing a detection process using the prediction model that had undergone the training process.

[0138] In Example 9, yeast introduced with an empty vector was inoculated onto an agar plate as a training microorganism, and approximately 100 colonies were formed by culturing at 30°C for 72 hours. Subsequently, the colonies were irradiated from two directions using light in the wavelength range of 400 nm to 1000 nm in the measurement device 1, and the reflectance spectrum of the colonies was measured using a hyperspectral camera. In the learning device 2, the mean and standard deviation of the measured reflectance spectrum were used as training data to train a prediction model. ABOD was adopted as the learning algorithm.

[0139] In Example 9, to evaluate the prediction model, yeast known to produce β-carotene above a predetermined amount and yeast known to produce β-carotene below a predetermined amount were inoculated into different regions on an agar plate. Multiple colonies were then formed by culturing at 30°C for 72 hours. In the measuring device 1, the colonies were irradiated from two directions using light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the colonies was measured using a hyperspectral camera. In the detection device 3, the obtained reflectance spectrum was input into the prediction model described above for detection processing.

[0140] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in detection device 3. In the evaluation test, it was assessed whether yeast, which was known in advance to produce a predetermined amount of β-carotene or more, was detected by detection device 3 as a microorganism that was not a low-productivity microorganism. In addition, the evaluation test was assessed whether yeast, which was known in advance to produce less than a predetermined amount of β-carotene, was detected by detection device 3 as a low-productivity microorganism.

[0141] The evaluation test was conducted using the following method: Yeast detected as a low-productivity microorganism by detection device 3 and yeast detected as a non-low-productivity microorganism were cultured in liquid at 30°C for 72 hours, after which only the yeast was recovered by centrifugation. Acetone and zirconia beads were added to the yeast pellet obtained by centrifugation and stirred for 20 minutes. The solution and solid were separated by filtration, and the obtained solution was subjected to high-performance liquid chromatography (HPLC) analysis.

[0142] The results of the evaluation test in Example 9 showed that the accuracy, recall, and precision were all 90% or higher.

[0143] This confirmed that the detection device 3 can output highly accurate detection results even when yeast with or without the ability to produce β-carotene is used as the target microorganism and the learning microorganism.

[0144] (Example 10) In Example 10, yeast having or not having the ability to produce PHB (polyhydroxybutyrate) was used as the target microorganism for detection and the learning microorganism, and a training process for a prediction model and a detection process using the prediction model that had undergone the training process were performed.

[0145] The learning process in Example 10 is the same as in Example 9.

[0146] In Example 10, to evaluate the prediction model, yeast known to produce more than a predetermined amount of PHB and yeast known to produce less than a predetermined amount of PHB were inoculated into different regions on an agar plate. Multiple colonies were then formed by culturing at 30°C for 72 hours. In the measuring device 1, the colonies were irradiated from two directions using light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the colonies was measured using a hyperspectral camera. In the detection device 3, the obtained reflectance spectrum was input into the prediction model described above for detection processing.

[0147] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in detection device 3. In the evaluation test, it was assessed whether yeast, whose PHB production was known to be above a predetermined amount, was detected by detection device 3 as a non-low-productive microorganism. In addition, the evaluation test was assessed whether yeast, whose PHB production was known to be below a predetermined amount, was detected by detection device 3 as a low-productive microorganism. The evaluation test was conducted using the Nile Red test.

[0148] The results of the evaluation test in Example 10 showed an accuracy rate of 70% or more and less than 90%, a recall rate of 60% or more and less than 70%, and a precision rate of 90% or more.

[0149] This confirmed that the detection device 3 has a certain level of detection accuracy, even when yeast with or without the ability to produce PHB is used as the target microorganism and the learning microorganism.

[0150] (Example 11) In Example 11, the learning process for the prediction model was performed in the learning device 2 in the same procedure as in Example 10, except that a culture medium to which Nile Red had been added in advance was used when culturing the learning microorganisms and the microorganisms to be detected. The culture medium to which Nile Red had been added is an example of a culture medium to which a dye has been added as described herein. Then, in Example 11, the detection process in the detection device 3 was performed using the prediction model in the same procedure as in Example 10, and further evaluation tests of the detection results were performed in the same procedure as in Example 10.

[0151] The results of the evaluation test in Example 11 showed an accuracy rate of 90% or higher, a recall rate of 70% or higher but less than 90%, and a precision rate of 90% or higher.

[0152] This confirmed that, in order to improve detection accuracy, it is preferable to use a culture medium to which dyes have been added when culturing the learning microorganisms and the microorganisms to be detected.

[0153] (Example 12) In Example 12, the learning device 2 performed the same procedure as in Example 10, except that the measuring device 1 used a light source with a polarization generation function to generate S-polarized or P-polarized light and acquired the positional information of the learning microbial colony and the reflection spectrum of the colony in a crossed nicol state. Then, in Example 12, the detection process in the detection device 3 was performed using the prediction model in the same procedure as in Example 10, and further evaluation tests were performed on the detection results in the same procedure as in Example 10.

[0154] The results of the evaluation test in Example 12 showed that the accuracy, recall, and precision were all between 70% and 90%.

[0155] This confirmed that, in order to improve detection accuracy, it is preferable to use a light source with a polarization generation function in the measuring device 1.

[0156] (Example 13) In Example 13, the training process for a prediction model and the detection process using the prediction model that has undergone the training process were performed, with Benthamia tobacco having the ability to produce GFP or not being used as the target plant and the training plant.

[0157] In Example 13, tobacco plants that have or do not have the ability to produce GFP were prepared according to the following procedure.

[0158] (1-1) Construction of a transformed vector In Example 13, a binary vector into which the target gene was introduced was constructed by a recombination reaction using Gateway® LR Clonase® II (Thermo Fisher Scientific). Specifically, an entry clone was introduced into pDONR207 (Thermo Fisher Scientific) by a recombination reaction using Gateway LR Clonase II, containing DNA with a synthetic green fluorescent protein (sGFP) sequence, or DNA with a sequence encoding p19 derived from Tomato bushy stunt virus (TBSV), which is known to suppress cosuppression of the expression of the introduced gene. This entry clone was then introduced into the binary vector pDEST_35S_HSP_GWB5 by a recombination reaction using Gateway technology. The binary vector pDEST_35S_HSP_GWB5 was described in Fujiwara, S. et al., VP16 fusion induces the multiple-knockout phenotype of redundant transcriptional repressors partly by Med25-independent mechanisms in Arabidopsis, FEBS Letters Volume 588, Issue 20, 16 October 2014, Pages 3665-3672.

[0159] (1-2) Preparation of Benthamiana tobacco leaves expressing GFP Using the binary vectors obtained in (1-1), the Agrobacterium tumefaciens GV3101 strain was transformed by electroporation. Then, the Agroinfiltration method was performed in accordance with the method described in Kim, WY. et al., ZEITLUPE is a circadian photoreceptor stabilized by GIGANTEA in blue light, Nature 449, 356-360 (2007). Specifically, seeds were sown directly onto Nippi Horticultural Soil No. 1 (Nippon Fertilizer) and incubated for about one month at a set temperature of 26 degrees Celsius, with 16 hours of light and 8 hours of darkness, and a photosynthetic photon flux density of approximately 100 μmol / m³. ２ To tobacco leaves of Citrus benthamiana cultivated under conditions of / s, either an Agrobacterium suspension containing a p19 expression vector (control) or a 1:1 mixture of Agrobacterium suspensions containing sGFP expression vector and p19 expression vector, respectively (GFP), was injected using a 1 ml syringe without a needle (ss-01T, Terumo). After injection, the leaves were cultivated for a further 3 days, and then leaf fragments were punched out using a 5 mm diameter cork borer and used.

[0160] In Example 13, leaf discs of tobacco plants (Citricola benthamiana) into which vectors had been introduced by the agroinfiltration method were placed on an agar medium as learning plants. Subsequently, the leaf discs were irradiated from two directions using light in the wavelength range of 400 nm to 1000 nm in the measurement device 1A, and the reflectance spectrum of the leaf discs was measured using a hyperspectral camera. In the learning device 2A, the mean and standard deviation of the measured reflectance spectra for each wavelength were used as training data to train a prediction model. A k-NN was adopted as the learning algorithm.

[0161] To evaluate the performance of the detection device 3A, which performs detection processing using the predictive model that has undergone the learning process described above, the following was performed. Leaf discs of tobacco benthamiana, whose GFP production was known to be above a predetermined amount, and leaf discs of tobacco benthamiana, whose GFP production was known to be below a predetermined amount, were placed in different regions on agar. Then, in the measuring device 1A, the leaf discs were irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the leaf discs was measured using a hyperspectral camera. In the detection device 3A, the detection process was performed by inputting the obtained reflectance spectrum into the predictive model described above.

[0162] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in the detection device 3A. In the evaluation test, it was evaluated whether leaf discs of tobacco benthamiana, whose GFP production was known to be above a predetermined amount, were detected by the detection device 3A as plants that were not low-yielding plants. In addition, the evaluation test was evaluated whether leaf discs of tobacco benthamiana, whose GFP production was known to be below a predetermined amount, were detected by the detection device 3A as low-yielding plants.

[0163] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in the detection device 3A. The evaluation test was performed by fluorescence observation of GFP in each leaf disk.

[0164] The results of the evaluation test in Example 13 showed an accuracy rate of 70% or more and less than 90%, a recall rate of 60% or more and less than 70%, and a precision rate of 90% or more.

[0165] This confirmed that in Example 13, the detection device 3A can output highly accurate detection results when the leaf discs of tobacco benthamiana, which have or do not have the ability to produce GFP, are used as the target plant and the learning plant. It was also confirmed that the detection device 3A can be applied not only to microorganisms but also to plants.

[0166] (Example 14) In Example 14, Arabidopsis thaliana, which has or does not have the ability to produce curcumin, was used as the target plant for detection and the learning plant, and the training process of the prediction model and the detection process using the prediction model that had undergone the training process were performed.

[0167] In Example 14, Arabidopsis thaliana plants capable of producing curcumin, or not capable of producing curcumin, were prepared according to the following procedure.

[0168] (1-1) Construction of the transformation vector ProPkC4H:LOC4342896 The ProPkC4H:LOC4342896 vector was constructed by recombining plasmid DNA containing an amino acid encoding DNA sequence in the pDEST_PkC4H_HSP_GWB4 vector using an LR reaction. The pDEST_PkC4H_HSP_GWB4 vector is described in Sakamoto, S. et al, Identification of enzymatic genes with the potential to reduce biomass recalcitrance through lignin manipulation in Arabidopsis, Biotechnology for Biofuels, volume 13, Article number: 97 (2020), or Koncz, C. et al., The promoter of TL-DNA gene 5 controls the tissue-specific expression of chimaeric genes carried by a novel type of Agrobacterium binary vector, Molecular and General Genetics, Volume 204, pages 383-396, (1986).

[0169] (1-2) Transformation of Arabidopsis thaliana with ProPkC4H: LOC4342896 using the floral dip method The plasmid obtained in (1-1) was introduced into a soil bacterium strain (Agrobacterium tumefaciens strain GV3101 (C58C1Rifr) pMP90 (Gmr) (koncz and Schell 1986)) by electroporation. The introduced bacteria were cultured in 50 ml of LB medium for 1 day.

[0170] Next, the bacterial cells were collected from the culture medium and suspended in 250 ml of immersion medium. Wild Arabidopsis thaliana strain (Col-0) was then immersed in this medium for 2 minutes to infect the cells, and the cultures were grown as usual to harvest T1 seeds. Subsequently, the T1 seeds were sterilized with 50% bleach and 0.02% Triton X-100 solution for 7 minutes, rinsed three times with sterile water, and seeded on MS selective medium containing 50 mg / l kanamycin.

[0171] Transgenic plants (ProPkC4H: LOC4342896, T1) grown on the kanamycin plate described above were selected, transplanted into soil, and grown for approximately 8 weeks to obtain recombinant seeds (T2 seeds). The obtained seeds were selected using MS selective medium containing 50 mg / l kanamycin, and then the selected individuals (T2 individuals) were transplanted into soil and cultivated for 8 weeks.

[0172] (1-3) Preparation of flower stalk powder from ProPkC4H:LOC4342896 Arabidopsis thaliana plants In the ProPkC4H:LOC4342896 Arabidopsis thaliana T2 plants obtained in (1-2), irrigation was stopped approximately 80 days after sowing to kill the plants, and all above-ground flower stalks, excluding the leaves, were collected. The dried flower stalks were crushed using a ShakeMaster Auto at 1100 rpm for 10 minutes, and the powder that passed through a 500 μm mesh was collected and used as the analysis sample.

[0173] The wild-type Arabidopsis thaliana stem powder obtained in this way was placed on agar in a circular shape. Then, using measuring device 1A, the Arabidopsis thaliana stem powder was irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the stem powder was measured using a hyperspectral camera. In learning device 2A, the mean and standard deviation of the measured reflectance spectra for each wavelength were used as training data to train a prediction model. k-NN was adopted as the learning algorithm.

[0174] To evaluate the performance of the detection device 3A, which performs detection processing using the predictive model that has undergone the learning process described above, the following was performed. Stem powder of Arabidopsis thaliana, in which the amount of curcumin production was known to be above a predetermined amount, and stem powder of Arabidopsis thaliana, in which the amount of curcumin production was known to be below a predetermined amount, were placed in different regions on agar. Then, in the measuring device 1, the stem powder was irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the stem powder was measured with a hyperspectral camera. In the detection device 3A, the detection process was performed by inputting the obtained reflectance spectrum into the predictive model described above.

[0175] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in the detection device 3A. The evaluation test was performed by fluorescence observation of curcumin in each stem powder. In the evaluation test, it was assessed whether the stem powder of Arabidopsis thaliana, which was known to produce a predetermined amount of curcumin or more, was detected by the detection device 3A as a plant that was not a low-productivity plant. In addition, the evaluation test was assessed whether the stem powder of Arabidopsis thaliana, which was known to produce less than a predetermined amount of curcumin, was detected by the detection device 3A as a low-productivity plant.

[0176] The results of the evaluation test in Example 14 showed an accuracy rate of 70% or more and less than 90%, a recall rate of 90% or more, and a precision rate of 70% or more and less than 90%.

[0177] This confirmed that, in Example 14, the detection device 3A can output highly accurate detection results even when the target plant and learning plant are Arabidopsis thaliana stem powder, which may or may not have the ability to produce curcumin. It was also confirmed that the detection device 3A can be applied not only to plant leaf discs but also to stem powder.

[0178] (Example 15) In Example 15, as a model assuming plants with differences in chlorophyll production capacity, analysis was performed using Arabidopsis thaliana plants with different chlorophyll content. Leaves of Arabidopsis thaliana with high or low chlorophyll content were used as the target plant for detection and the training plant, and the training process of the prediction model and the detection process using the prediction model that had undergone the training process were performed.

[0179] In Example 15, Arabidopsis thaliana plants with high or low chlorophyll content were prepared according to the following procedure.

[0180] (1-1) Dark-induced senescence of Arabidopsis thaliana (Columbia-0) under dark conditions was performed as follows (Nakano, Y. et al., NSR1 / MYR2 is a negative regulator of ASN1 expression and its possible involvement in regulation of nitrogen reutilization in Arabidopsis, Plant Science, Volume 263, October 2017, Pages 219-225).

[0181] First, wild-type seeds, whose dormancy was broken by being left to stand in water at 4 degrees Celsius for 3-4 days, were sown in artificial culture medium (a mixture of black peat "Super Mix A" (Sakata Seed Corporation) and vermiculite (Asahi Kogyo Co., Ltd.) in a 1:1 volume ratio). After sowing, the temperature was 22 degrees Celsius, with a light period of 16 hours and a dark period of 8 hours, and a photosynthetic photon flux density of 50-70 μmol / m². －２ s －１The plants were grown in a plant cultivation room for 21 days. The grown above-ground parts were placed in a sterile petri dish FX (φ90 x 20 mm, 6-8663-02, Sansei Medical Equipment Co., Ltd.), placed on a water-containing qualitative filter paper No. 2 (φ70, Advantec), and then the petri dish was sealed with Micropore® surgical tape (1530-0, width 12.5 mm, 3M Japan Ltd.). After that, the petri dish was placed in a bright or dark place (22 degrees Celsius) under the above cultivation conditions for 4 days.

[0182] (1-2) Quantitative measurement of photosynthetic capacity by measuring the maximum quantum yield of photosystem II Using the pulse-modulated chlorophyll fluorescence analyzer MINI-PAM-II (WALZ), the maximum quantum yield of photosystem II (Fv / Fm) was measured in the first and fourth leaves of Arabidopsis thaliana treated under the above conditions. The measurement results showed that the Fv / Fm values ​​were significantly higher in plants placed in light compared to plants placed in darkness, confirming that the function of photosystem II was maintained.

[0183] Subsequently, the aerial parts of Arabidopsis thaliana were placed on agar. Then, using measurement device 1A, the leaf disk was irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the leaves of the aerial parts of Arabidopsis thaliana was measured using a hyperspectral camera. In learning device 2A, the mean and standard deviation of the measured reflectance spectra for each wavelength were used as training data to train a prediction model. k-NN was adopted as the learning algorithm.

[0184] To evaluate the performance of the detection device 3A, which performs detection processing using the predictive model that has undergone the learning process described above, the following was performed. Leaves of the above-ground part of Arabidopsis thaliana, whose chlorophyll content is known to be above a predetermined amount, and leaves of the above-ground part of Arabidopsis thaliana, whose chlorophyll content is known to be below a predetermined amount, were placed in different regions on agar. Then, in the measuring device 1A, the above-ground leaves were irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the above-ground leaves was measured with a hyperspectral camera. In the detection device 3A, the detection process was performed by inputting the obtained reflectance spectrum into the predictive model described above.

[0185] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in the detection device 3A. In the evaluation test, it was assessed whether the leaves of the above-ground parts of Arabidopsis thaliana, which were known to have a chlorophyll content of at least a predetermined amount, were detected by the detection device 3A as plants that were not low-yielding plants. In addition, the evaluation test was assessed whether the leaves of the above-ground parts of Arabidopsis thaliana, which were known to have a chlorophyll content below a predetermined amount, were detected by the detection device 3A as low-yielding plants.

[0186] Chloroplast extracts were performed from Arabidopsis thaliana (Nakano, Y. et al., NSR1 / MYR2 is a negative regulator of ASN1 expression and its possible involvement in regulation of nitrogen reutilization in Arabidopsis, Plant Science, Volume 263, October 2017, Pages 219-225), and the chlorophyll a+b content was analyzed by absorbance measurement. Specifically, four plant specimens (entire above-ground parts) placed in both light and dark conditions were freeze-dried and pulverized with liquid nitrogen. After adding ice-cold 80% acetone and holding at 4°C for approximately 18 hours, the supernatant obtained by centrifugation was used to measure the absorbance at 664 nm, 647 nm, and 750 nm using a spectrophotometer. Based on the measured values, the chlorophyll content (μg / gFW) was calculated. The chlorophyll content was calculated according to the method described in Porra, RJ et al., "Determination of accurate extinction coefficients and simultaneous equations for assaying chlorophylls a and b extracted with four different solvents: verification of the concentration of chlorophyll standards by atomic absorption spectroscopy," Biochimica et Biophysica Acta (BBA) - Bioenergetics, Volume 975, Issue 3, August 1989, Pages 384-394.

[0187] As a result, it was confirmed that plants placed in bright light had a significantly higher chlorophyll content compared to plants placed in dark light.

[0188] The results of the evaluation test in Example 15 showed an accuracy rate of 90% or higher, a recall rate of 70% or higher but less than 90%, and a precision rate of 90% or higher.

[0189] This confirmed that in Example 15, the detection device 3A can output highly accurate detection results even when the above-ground leaves of Arabidopsis thaliana, which have different chlorophyll contents, are used as the target plant and the learning plant. It was also confirmed that the detection device 3A can be applied not only to plant leaf discs and stem powder but also to unprocessed leaves.

[0190] (Example 16) In Example 16, seeds of Arabidopsis thaliana that have or do not have the ability to produce RFP (red fluorescent protein) were used as the target plant for detection and the learning plant, and the training process of the prediction model and the detection process using the prediction model that had undergone the training process were performed.

[0191] In Example 16, Arabidopsis thaliana seeds with or without the ability to produce TagRFP (hereinafter referred to as RFP for simplicity), a type of RFP, were prepared according to the following procedure.

[0192] The plasmid pKI1.1R (Tsutsui, H. et al., pKAMA-ITACHI vectors for highly efficient CRISPR / Cas9-mediated gene knockout in Arabidopsis thaliana., Plant Cell Physiol. 2016 Nov 17. pii:pcw191), which causes transformed seeds to exhibit RFP fluorescence, was introduced into a soil bacterium strain (Agrobacterium tumefaciens strain GV3101 (Koncz and Schell 1986)) by electroporation. The introduced bacteria were cultured at 28 degrees Celsius in 50 ml of LB medium for 2 days. The bacterial cells were collected from the culture medium and suspended in 50 ml of infection solution (10% sucrose, 0.05% Silwet-77). Flower buds of wild Arabidopsis thaliana (Col-0 strain) were immersed in this solution for 2 minutes to infect them. The cells were then cultured for about 1 month at 22 degrees Celsius, with a light intensity of approximately 60 PPFD, in a 16-hour light / 8-hour dark cycle, and T1 seeds were harvested. Subsequently, the fluorescence of the T1 seeds was evaluated using a fluorescence stereomicroscope, and seeds showing RFP fluorescence were sown and cultivated to obtain T2 seeds in about 3 months. Of the obtained T2 seeds, at least 20 seeds were selected visually using a fluorescence stereomicroscope to distinguish between seeds showing RFP fluorescence and those that did not. Each type of seed was used as an analytical sample.

[0193] Seeds that did not emit RFP fluorescence, selected in this manner, were placed on agar using toothpicks to serve as learning plants. Subsequently, in measuring device 1A, the seeds were irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the Arabidopsis thaliana seeds was measured using a hyperspectral camera. In learning device 2A, the mean and standard deviation of the measured reflectance spectra for each wavelength were used as training data to train a prediction model. ABOD was adopted as the learning algorithm.

[0194] To evaluate the performance of the detection device 3A, which performs detection processing using the predictive model that has undergone the learning process described above, the following was performed. Seeds of Arabidopsis thaliana, whose RFP production volume was known to be above a predetermined amount, and seeds of Arabidopsis thaliana, whose RFP production volume was known to be below a predetermined amount, were placed in different regions on agar. Then, in the measuring device 1A, the seeds were irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the seeds was measured using a hyperspectral camera. In the detection device 3, the detection process was performed by inputting the obtained reflectance spectrum into the predictive model described above.

[0195] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in detection device 3A. The evaluation test was performed by fluorescence observation of RFP in various seeds. In the evaluation test, it was assessed whether Arabidopsis thaliana seeds, whose RFP production was known to be above a predetermined amount, were detected by detection device 3A as plants that were not low-yielding plants. In addition, the evaluation test was assessed whether Arabidopsis thaliana seeds, whose RFP production was known to be below a predetermined amount, were detected by detection device 3 as low-yielding plants.

[0196] The results of the evaluation test in Example 16 showed that the accuracy rate was 60% or more and less than 70%, the recall rate was 60% or more and less than 70%, and the precision rate was 60% or more and less than 70%.

[0197] This confirmed that in Example 16, the detection device 3A can output detection results with a certain level of accuracy, even when Arabidopsis thaliana seeds, which may or may not be capable of producing RFP, are used as the target plant and learning plant. It was also confirmed that the detection device 3A can be applied not only to plant leaf discs, stem powder, and unprocessed leaves, but also to extremely small and unprocessed seeds.

[0198] (Example 17) In Example 17, seeds of Arabidopsis thaliana that have or do not have the ability to produce anthocyanins were used as the target plant for detection and the learning plant, and the training process of the predictive model and the detection process using the predictive model that had undergone the training process were performed.

[0199] In Example 17, Arabidopsis thaliana seeds with or without the ability to produce anthocyanins were prepared according to the following procedure.

[0200] To obtain training data for seeds with low anthocyanin productivity, we obtained SALK_005260 (tt2-5 strain) seeds from the Arabidopsis Biological Resource Center (ABRC). The obtained seeds were sown and cultured for approximately three months at 22 degrees Celsius, with a light intensity of approximately 60 PPFD and a 16-hour light / 8-hour dark cycle, before harvesting the next generation of seeds. The tt2-5 strain is known to have low anthocyanin accumulation in the seed coat (Chen, M. et al., The effect of transparent TESTA2 on seed fatty acid biosynthesis and tolerance to environmental stresses during young seedling establishment in Arabidopsis, Plant Physiol. 2012 Oct; 160(2): 1023-36.). The tt2-5 strain seeds were used as "pigment-free seeds," and the wild-type Col-0 seeds were used as "pigment-containing seeds" for training and analysis.

[0201] The seeds of the tt2-5 strain obtained in this way were placed on agar using toothpicks to serve as learning plants. Subsequently, in measurement device 1A, the seeds were irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the Arabidopsis thaliana seeds was measured using a hyperspectral camera. In learning device 2A, the mean and standard deviation of the measured reflectance spectra for each wavelength were used as training data to train a prediction model. A k-NN was adopted as the learning algorithm.

[0202] To evaluate the performance of the detection device 3A, which performs detection processing using the predictive model that has undergone the learning process described above, the following was performed. Seeds of Arabidopsis thaliana, whose anthocyanin production was known to be above a predetermined amount, and seeds of Arabidopsis thaliana, whose anthocyanin production was known to be below a predetermined amount, were placed in different regions on agar. Then, in the measuring device 1A, the seeds were irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the seeds was measured using a hyperspectral camera. In the detection device 3A, the detection process was performed by inputting the obtained reflectance spectrum into the predictive model described above.

[0203] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in detection device 3. The evaluation test was performed by visually observing the seeds of each species. In the evaluation test, it was assessed whether Arabidopsis thaliana seeds, which were known to produce anthocyanins above a predetermined amount, were detected by detection device 3 as plants that were not low-productivity plants. In addition, the evaluation test was assessed whether Arabidopsis thaliana seeds, which were known to produce anthocyanins below a predetermined amount, were detected by detection device 3 as low-productivity plants.

[0204] The results of the evaluation test in Example 17 showed an accuracy rate of 70% or more and less than 90%, a recall rate of 70% or more and less than 90%, and a precision rate of 90% or more.

[0205] This confirms that in Example 17, the detection device 3A can output highly accurate detection results even when the target plant and learning plant are Arabidopsis thaliana seeds that have or do not have the ability to produce anthocyanins.

[0206] (Example 18) In Example 18, colonies of Euglena gracilis (NIES-48) with high or low chlorophyll production capacity were used as the target microorganism for detection and the learning microorganism for training. The training process for the prediction model and the detection process using the prediction model trained through the training process were performed.

[0207] In Example 18, Euglena colonies with high or low chlorophyll production capacity were prepared according to the following procedure. A culture medium with an optical density of 1 OD was diluted with phosphate buffer at a predetermined ratio and then inoculated onto agar medium (KH medium) by the Misler method. Euglena with high chlorophyll production was then obtained by culturing at 25°C for 10 days without light shielding. On the other hand, Euglena with low chlorophyll production was obtained by culturing at 25°C for 10 days while shielded from light with aluminum foil. Subsequently, the colonies were irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm using measuring device 1, and the reflectance spectrum of the Euglena colonies was measured using a hyperspectral camera. In learning device 2, the mean and standard deviation of the measured reflectance spectra for each wavelength were used as training data to train a predictive model. ABOD was adopted as the learning algorithm.

[0208] To evaluate the performance of the detection device 3, which performs detection processing using the predictive model that has undergone the learning process described above, the following was performed. Petri dishes were prepared in which Euglena colonies that had been light-shielded and in which Euglena colonies that had not been light-shielded were formed. Then, in the measuring device 1, the colonies were irradiated from two directions with light in the wavelength range of 400 nm to 1000 nm, and the reflectance spectrum of the colonies was measured with a hyperspectral camera. In the detection device 3, the detection process was performed by inputting the obtained reflectance spectrum into the predictive model described above.

[0209] Subsequently, an evaluation test was conducted to assess the detection accuracy of the detection process in detection device 3. The evaluation test involved evaluating chlorophyll production by fluorescence observation of each colony. In the evaluation test, it was assessed whether Euglena colonies that had not been subjected to light-shielding treatment were detected by detection device 3 as non-low-productive microorganisms. In addition, the evaluation test assessed whether Euglena colonies whose chlorophyll production was known to be below a predetermined amount were detected by detection device 3 as low-productive microorganisms.

[0210] In the evaluation tests, almost no chlorophyll-derived red fluorescence was observed in Euglena colonies that had been treated with light shielding. On the other hand, strong chlorophyll-derived red fluorescence was observed in Euglena colonies that had not been treated with light shielding.

[0211] The results of the evaluation test in Example 18 showed an accuracy rate of 90% or higher, a recall rate of 90% or higher, and a precision rate of 90% or higher.

[0212] This confirmed that in Example 18, the detection device 3 can output highly accurate detection results even when Euglena colonies with high or low chlorophyll production capacity are used as the target microorganisms and learning microorganisms. It also became clear that the detection device 3 can be applied to microalgae as well.

[0213] (Comparative Example 1) In Comparative Example 1, the learning process for the prediction model was performed in the learning device 2 using the same procedure as in Example 1, except that Escherichia coli whose fatty acid production was known to be above a predetermined amount was used as the learning microorganism. Then, in Comparative Example 1, the detection process in the detection device 3 was performed using the prediction model using the same procedure as in Example 1, and further evaluation tests of the detection results were performed using the same procedure as in Example 1.

[0214] In Comparative Example 1, the evaluation test results showed that the accuracy, recall, and precision were all less than 60%.

[0215] This confirmed that, in Example 1, the detection accuracy of the detection device 3 was significantly improved by using E. coli with an empty vector introduced, which clearly does not possess the ability to produce substances as a learning microorganism.

[0216] (Comparative Example 2) In Comparative Example 2, the learning process for the prediction model was performed in the learning device 2 using the same procedure as in Example 1, except that Escherichia coli whose fatty acid production was known to be above a predetermined amount was used as the learning microorganism, and k-NN was used instead of ABOD as the learning processing algorithm in the learning device 2. Then, in Comparative Example 2, the detection process in the detection device 3 was performed using the prediction model in the same procedure as in Example 1, and further evaluation tests of the detection results were performed in the same procedure as in Example 1.

[0217] In Comparative Example 2, the evaluation test results showed that the accuracy, recall, and precision were all less than 60%.

[0218] This confirmed that, regardless of the algorithm used in the learning process, if E. coli bacteria whose fatty acid production is known to be above a predetermined amount are used as the learning microorganism, the same detection accuracy as in Example 1 cannot be ensured.

[0219] As explained above, the detection device 3 of this disclosure can perform detection processing with significantly higher accuracy than the comparative example which uses microorganisms with high material production capacity as learning microorganisms, by using low-productivity microorganisms as learning microorganisms.

[0220] <Example of Computer Hardware Configuration> As described above, the learning devices 2, 2A and detection devices 3, 3A of the detection systems 100, 100A are computers. Below, an example of the hardware configuration of the computer as the learning devices 2, 2A and detection devices 3, 3A will be described. Figure 13 is a diagram showing an example of the hardware configuration of computer 1000 as the learning devices 2, 2A and detection devices 3, 3A.

[0221] The computer 1000 includes a drive device 101, a storage device 102, a memory device 103, a processor 104, a user interface (UI) device 105, and a communication device 106, all interconnected via bus B.

[0222] The programs or instructions that implement the various functions and processes described above in the computer 1000 may be stored in a removable storage medium such as a CD-ROM (Compact Disk - Read Only Memory) or flash memory. When the storage medium is set in the drive device 101, the programs or instructions are installed from the storage medium to the storage device 102 or memory device 103 via the drive device 101. However, the programs or instructions do not necessarily have to be installed from the storage medium; they may also be downloaded from an external device via a network or the like.

[0223] The storage device 102 is implemented by a hard disk drive or the like, and stores installed programs or instructions along with files, data, etc., used to execute the programs or instructions.

[0224] The memory device 103 is implemented using random access memory, static memory, etc., and when a program or instruction is activated, it reads the program or instruction, data, etc. from the storage device 102 and stores it. The storage device 102, the memory device 103, and the removable storage medium may be collectively referred to as a non-transitory tangible storage medium.

[0225] The processor 104 may be implemented by one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), processing circuits, etc., which may consist of one or more processor cores, and executes various functions and processes of the computer 1000, as described later, according to data such as programs, instructions, and parameters necessary to execute the programs or instructions stored in the memory device 103.

[0226] The user interface (UI) device 105 may consist of input devices such as a keyboard, mouse, camera, and microphone, output devices such as a display, speaker, headset, and printer, and input / output devices such as a touch panel, and realizes an interface between the user and the computer 1000. For example, the user operates the computer 1000 by operating a GUI (Graphical User Interface) displayed on the display or touch panel using a keyboard, mouse, etc.

[0227] The communication device 106 is realized by various communication circuits that perform wired and / or wireless communication processing with external devices, the Internet, a LAN (Local Area Network), a cellular network, and other communication networks.

[0228] The hardware configuration described above is merely an example, and the computer 1000 according to this disclosure may be implemented by any other suitable hardware configuration.

[0229] <Modification> In the first embodiment described above, a detection system 100 that targets microorganisms was described. In the second embodiment described above, a detection system 100A that targets plants was described.

[0230] This disclosure is not limited to these embodiments, and detection systems for detecting microorganisms or plants are also included in this disclosure. Furthermore, the detection systems of this disclosure may also detect cells of breedable animals (such as cattle, pigs, sheep, horses, dogs, and cats) in addition to microorganisms or plants. In this case, animal cells can be used as the detection target and the same processing as in the embodiments described above can be performed.

[0231] All disclosures in the specification, drawings, and abstract contained in the Japanese application No. 2024-193914, filed on November 5, 2024, are incorporated herein by reference.

[0232] This disclosure is useful for detection methods that perform detection of material production capacity.

[0233] 100, 100A Detection system 1, 1A Measurement device 11 Base 12, 13 Light source 14 Camera 2, 2A Learning device 21 Acquisition unit 22 Storage unit for learning target model 23 Learning unit 3, 3A Detection device 31 Acquisition unit 32 Execution unit 33 Detection unit 34 Result output unit

Claims

1. A detection method in which a computer performs the following steps:

1. Obtain the physical properties of a microorganism to be detected; 2. Input the physical properties of the microorganism to be detected into a predictive model that has been trained using the physical properties of low-productivity microorganisms having a material production capacity below a predetermined capacity, and obtain the output from the predictive model; and 3. Detect the low-productivity microorganisms among the microorganisms to be detected based on the output of the predictive model.

2. The detection method according to claim 1, wherein the prediction model is trained to output an abnormality score indicating the degree to which the microorganism to be detected is not a low-productivity microorganism when the physical properties of the microorganism to be detected are input.

3. The detection method according to claim 2, wherein the detection step involves detecting microorganisms among the target microorganisms whose degree of abnormality is below a threshold as low-productivity microorganisms.

4. The detection method according to any one of claims 1 to 3, wherein the computer further performs a step of outputting as a detection result the microorganisms obtained by removing the low-productivity microorganisms detected in the detection step from the microorganisms to be detected.

5. The detection method according to claim 1, wherein the physical property is the photophysical property exhibited by the microorganism.

6. The detection method according to claim 5, wherein the optical properties are optical properties obtained in a crossed nicol state using S-polarized or P-polarized light.

7. The detection method according to claim 5, wherein the optical properties are obtained by irradiating the colony of the microorganism with light from multiple directions.

8. The detection method according to claim 7, wherein the optical property is the reflectance spectrum of the colony obtained by irradiation with light having a wavelength of 400 nm or more and 1000 nm or less.

9. The detection method according to claim 8, wherein the photophysical properties are obtained by irradiating the colonies cultured in a culture medium to which the dye has been added with light.

10. The detection method according to any one of claims 1 to 3, wherein the prediction model is trained using the physical properties with outliers removed.

11. A detection device comprising: an acquisition unit that acquires the physical properties of microorganisms to be detected; an execution unit that inputs the acquired physical properties into a prediction model that has been trained using the physical properties of low-productivity microorganisms having a material production capacity below a predetermined capacity, and acquires the output from the prediction model; and a detection unit that detects the low-productivity microorganisms among the microorganisms to be detected based on the output.

12. A program that causes a computer to perform the following steps: a procedure for obtaining the physical properties of microorganisms to be detected; a procedure for inputting the obtained physical properties into a predictive model that has been trained using the physical properties of low-productivity microorganisms having a material production capacity below a predetermined capacity, and obtaining the output from the predictive model; and a procedure for detecting the low-productivity microorganisms among the microorganisms to be detected based on the output of the predictive model.

13. A detection method in which a computer performs the following steps: acquiring the physical properties of a plant to be detected; inputting the physical properties of the plant to be detected into a predictive model that has been trained using the physical properties of low-productivity plants having a material production capacity below a predetermined capacity, and obtaining the output from the predictive model; and detecting the low-productivity plants among the plants to be detected based on the output of the predictive model.

14. The detection method according to claim 13, wherein the prediction model is trained to output an abnormality score indicating the degree to which the plant to be detected is not a low-productivity plant when the physical properties of the plant to be detected are input.

15. The detection method according to claim 14, wherein the detection step involves detecting plants among the plants to be detected whose degree of abnormality is below a threshold as low-productivity plants.

16. The detection method according to any one of claims 13 to 15, wherein the computer further performs a step of outputting as a detection result the plants obtained by removing the low-yielding plants detected in the detection step from the plants to be detected.

17. The detection method according to claim 13, wherein the physical property is the photophysical property exhibited by the plant.

18. The detection method according to claim 17, wherein the optical properties are optical properties obtained in a crossed nicol state using S-polarized or P-polarized light.

19. The detection method according to claim 17, wherein the optical properties are obtained by irradiating the plant with light from multiple directions.

20. The detection method according to claim 19, wherein the optical property is the reflectance spectrum of the plant obtained by irradiation with light having a wavelength of 400 nm or more and 1000 nm or less.

21. The detection method according to any one of claims 13 to 15, wherein the prediction model is trained using the physical properties with outliers removed.

22. A detection device comprising: an acquisition unit that acquires the physical properties of a plant to be detected; an execution unit that inputs the acquired physical properties into a prediction model that has been trained using the physical properties of low-productivity plants having a material production capacity below a predetermined capacity, and acquires the output from the prediction model; and a detection unit that detects the low-productivity plants among the plants to be detected based on the output.

23. A program that causes a computer to perform the following steps: a procedure for obtaining the physical properties of a plant to be detected; a procedure for inputting the obtained physical properties into a predictive model that has been trained using the physical properties of low-productivity plants having a material production capacity below a predetermined capacity, and obtaining the output from the predictive model; and a procedure for detecting the low-productivity plants among the plants to be detected based on the output of the predictive model.

24. A detection method in which a computer performs the following steps: acquiring the physical properties of a target to be detected, including microorganisms or plants; inputting the physical properties of the target to be detected into a predictive model that has been trained using the physical properties of a low-productivity target, which is a target to be detected having a material production capacity below a predetermined capacity, and obtaining the output from the predictive model; and detecting the low-productivity target among the target to be detected based on the output of the predictive model.

25. A detection device comprising: an acquisition unit that acquires the physical properties of a target to be detected, including microorganisms or plants; an execution unit that inputs the physical properties of the target to be detected into a prediction model that has been trained using the physical properties of a low-productivity target, which is a target to be detected having a material production capacity below a predetermined capacity, and acquires the output from the prediction model; and a detection unit that detects the low-productivity target among the target to be detected based on the output of the prediction model.

26. A program that causes a computer to perform the following steps: a procedure for obtaining the physical properties of a target to be detected, including microorganisms or plants; a procedure for inputting the physical properties of the target to be detected into a predictive model that has been trained using the physical properties of a low-productivity target, which is a target to be detected having a material production capacity below a predetermined capacity, and obtaining the output from the predictive model; and a procedure for detecting the low-productivity target among the target to be detected based on the output of the predictive model.

Images