Inspection method, inspection device, learning method, learning device, inspection program, and learning program

Abstract

Provided is a highly accurate method for testing a disease without giving an excess burden to a medical setting. A test device (100) executes steps (S20, S21) for acquiring maker data showing the result of measurement of a disease marker in a specimen and preparation data showing the preparation conditions of the specimen, step (S22) for inputting the marker data and preparation data to a learned model (121), said learned model having learned, by machine learning, correlation between the set of marker data showing the result of measurement of a disease marker in a specimen and preparation data of the specimen and the presence or absence of affection with the disease, and step (S23) for determining the presence or absence of affection with the disease.

Description

[Technical field]

[0001] The present invention relates to a testing method and testing device for testing a disease using a disease marker, a learning method and learning device for generating a trained model used in the testing, and a testing program and a learning program. [Background technology]

[0002] Analysis of proteins, DNA, and RNA contained in blood for the purpose of testing for diseases has been conducted since the 1960s. In particular, analysis of miRNA has become more common since around 2007. In recent years, efforts are being made around the world to determine the presence or absence of cancer and other diseases from the expression profile of miRNA in serum.

[0003] In tests using these biomolecules (disease markers) that are related to disease onset and serve as indicators of disease onset, it is necessary to suppress fluctuations in the abundance of disease markers associated with sample collection. For example, Patent Document 1 discloses that after storing a serum sample at 4°C for 72 or 168 hours, the abundance of some miRNAs in the sample fluctuates significantly. For this reason, it is common to standardize protocols, such as by standardizing test conditions, including sample collection. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication WO2017 / 146033 Summary of the Invention [Problem to be solved by the invention]

[0005] Since sampling of subject samples is performed at each medical institution, the conditions may differ from institution to institution, and therefore, in order to prevent fluctuations in disease markers, we are forced to ask medical institutions for their cooperation in standardizing the conditions.

[0006] Furthermore, in the course of the study, the present inventors have found that not only when the relatively long storage time (72 hours or 168 hours) shown in Patent Document 1 is performed, but also when the serum is adjusted for a short time (for example, within 6 hours), the expression profile of miRNA can unexpectedly vary greatly. Therefore, in order to perform a test with high accuracy, it is necessary to exclude miRNAs that are prone to variation from the analysis target, or to request medical institutions to comply with a unified protocol or to sample under more detailed conditions in order to suppress the variation as much as possible. However, discarding data by excluding some miRNAs from the analysis target is a great loss in performing a test with high accuracy. In addition, compliance with strict protocols and sampling under more detailed conditions will impose an additional burden on the medical field, which may hinder the implementation or dissemination of the test.

[0007] Therefore, the present invention has been made in consideration of the above-mentioned problems, and its object is to provide a highly accurate disease testing method that allows for a wide range of feasible sample collection conditions without placing an excessive burden on medical facilities. [Means for solving the problem]

[0008] In order to solve the above-mentioned problems, the testing method of the present invention is a testing method for testing for a disease using a disease marker, and includes a specimen data acquisition step of acquiring marker data indicating the results of measuring a disease marker in a body fluid sample collected from a subject, and preparation data indicating the preparation conditions of the body fluid sample, and a step of determining the presence or absence of a disease in the subject by inputting the marker data and the preparation data acquired in the specimen data acquisition step into a trained model that has been machine-learned to determine the correlation between a set of marker data indicating the results of measuring a disease marker in a body fluid sample and the preparation data of the body fluid sample, and the presence or absence of a disease in the subject from whom the body fluid sample was collected.

[0009] Furthermore, one aspect of the testing method according to the present invention is configured to further include a training data acquisition step of acquiring marker data indicating the results of measuring a disease marker in a body fluid sample, preparation data indicating preparation conditions for the body fluid sample, and clinical data indicating the presence or absence of a disease in a subject from whom the body fluid sample was collected, and a model generation step of generating the trained model by machine learning the correlation between the set of the marker data and the preparation data acquired in the training data acquisition step and the clinical data.

[0010] In order to solve the above-mentioned problems, the learning method of the present invention is a learning method for generating a trained model to be used in testing for a disease using a disease marker, and includes a training data acquisition step of acquiring marker data indicating the results of measuring a disease marker in a body fluid sample, preparation data indicating preparation conditions for the body fluid sample, and clinical data indicating the presence or absence of a disease in a subject from whom the body fluid sample was collected, and a model generation step of generating a trained model capable of determining the presence of the disease from the marker data and the preparation data by machine learning the correlation between the acquired set of marker data and preparation data and the clinical data.

[0011] In order to solve the above-mentioned problems, the learning device of the present invention is a learning device that generates a trained model to be used in disease testing using disease markers, and is configured to include a learning data acquisition unit that acquires marker data indicating the results of measuring a disease marker in a body fluid sample, preparation data indicating the preparation conditions of the body fluid sample, and clinical data indicating the presence or absence of a disease in a subject from whom the body fluid sample was collected, and a model generation unit that generates a trained model capable of determining the presence of the disease from the marker data and the preparation data by machine learning the correlation between the acquired set of marker data and preparation data and the clinical data.

[0012] In order to solve the above-mentioned problems, the testing device of the present invention is a testing device that performs disease testing using disease markers, and is configured to include a specimen data acquisition unit that acquires the marker data and preparation data in a body fluid sample derived from a subject, and a discrimination unit that uses the trained model generated by the above-mentioned learning device to discriminate the presence of the disease from the marker data and preparation data acquired by the specimen data acquisition unit. Effect of the Invention

[0013] According to one aspect of the present invention, by using preparation data indicating the sample preparation conditions in addition to marker data indicating the results of measuring disease markers in a sample, it is possible to perform highly accurate disease testing without placing an undue burden on medical facilities. [Brief description of the drawings]

[0014] [Figure 1] 1 is a functional block diagram showing a schematic configuration of an inspection device and a terminal device according to an embodiment of the present invention; [Diagram 2] 5 is a flowchart showing a process flow in a learning phase of a control unit of an inspection apparatus according to an embodiment of the present invention. [Diagram 3] 5 is a flowchart showing a process flow in an inspection phase of a control unit of an inspection device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] An embodiment of the present invention will be described below. The testing device in this embodiment is a device for testing a disease using a disease marker, and is a testing device that receives marker data indicating the result of measuring a disease marker in a body fluid sample and preparation data of the body fluid sample as input, and determines the presence or absence of a disease using a trained model that has been machine-learned to output the presence or absence of a disease.

[0016] 〔term〕 First, the terminology will be explained. In this specification, the term "body fluid sample" refers to a body fluid collected from a subject and used in a test. There are no particular limitations on the body fluid as long as it can be used in a test for measuring a disease marker, and examples of the body fluid include blood, serum, plasma, cerebrospinal fluid, urine, saliva, tears, tissue fluid, and lymphatic fluid. Among these, blood, serum, and plasma are preferably used.

[0017] In this specification, the term "disease marker" refers to a biomolecule whose presence or abundance is related to a specific disease. Examples of disease markers include DNA, RNA, and proteins. Among these, RNA is preferably used, and non-coding RNA (ncRNA) is more preferably used. ncRNA is roughly classified into small ncRNAs with a length of about 20 to 200 bases and long ncRNAs with a total length of several hundred to several hundred thousand bases. Examples of ncRNA include transfer RNA, ribosomal RNA, small nuclear RNA, small nucleolar RNA, signal recognition complex RNA, miRNA, piRNA, long non-coding RNA, circular RNA, and non-translated regions of mRNA, and miRNA is particularly preferably used. In addition, the disease is not particularly limited as long as the presence of a disease marker is known, and examples thereof include cancer, dementia, high blood pressure, heart disease, brain disease, hepatitis, infectious disease, and allergy. Examples of cancer include pancreatic cancer, biliary tract cancer, breast cancer, lung cancer, colon cancer, esophageal cancer, gastric cancer, liver cancer, prostate cancer, bladder cancer, brain cancer, blood cancer, ovarian cancer, and uterine cancer. Examples of the cancer include pancreatic cancer and biliary tract cancer.

[0018] In this specification, the term "marker data" refers to data indicating the presence or absence or the amount of one or more disease markers obtained by measuring a disease marker in a body fluid sample. When the actual detection target is DNA, RNA, or protein, the "abundance" can be rephrased as "expression amount". When used with respect to a disease marker, "measure" can be rephrased as "detect". The marker data in this embodiment is not limited to the result obtained by measuring one specific disease marker, but may include multiple measurement results or numerical data corresponding to each disease marker obtained by measuring multiple disease markers. The measurement results of multiple disease markers may be the result of simultaneously measuring each disease marker, or may be the result of measuring each disease marker independently. When multiple disease markers are used, the number is not limited, and may be, for example, 2 or more, 3 or more, 4 or more, 5 or more, 10 or more, 15 or more, 20 or more, 30 or more, or 40 or more. The measurement means for obtaining the marker data can be appropriately selected depending on the biomolecule to be measured and the required data. Examples of the measurement means include microarrays, various PCRs including qRT-PCR, various sequencings including next-generation sequencing, and ELISA. Among them, it is preferable to use a microarray, from the viewpoint that a plurality of disease markers can be measured simultaneously and a highly accurate test can be performed by using a plurality of disease markers. When converting the measurement results into numerical data, the conversion method is not particularly limited as long as a correlation is observed between the measurement results and the numerical data.

[0019] As used herein, "subject" refers to mammals such as primates, including humans and chimpanzees, pets such as dogs and cats, livestock animals such as cows, horses, sheep and goats, rodents such as mice and rats, and animals kept in zoos. A preferred subject is a human. A "healthy individual" is an individual of the same species as the subject of interest who is not affected by the disease of interest.

[0020] "Preparation data" refers to data indicating preparation conditions for a body fluid sample, and includes one or more pieces of information related to the body fluid sample other than the measurement results of disease markers obtained by measurement. In detail, examples of information included in the preparation data include information on the subject from which the body fluid sample was collected, information indicating the collection conditions of the body fluid sample, and information on the conditions of treatment performed on the body fluid sample before measuring disease markers in the body fluid sample. Differences in the subject or the condition of the subject, or differences in the collection or treatment conditions of the body fluid sample, can cause considerable variation in the profile of disease markers in the body fluid sample.

[0021] "Subject information" refers to information about the subject himself that is not directly related to the body fluid sample and the collection of the body fluid sample, or information about the subject's physical condition at the time of collection of the body fluid sample. Such information includes, for example, information about the subject's race, diet and eating time on the day or the day before the sample is collected, the time between the subject's last meal and the collection of the body fluid, and, if the subject is taking medication, the type of medication taken or the time between the administration of the medication and the collection of the body fluid, or other information about the subject's physical condition.

[0022] "Information indicating the collection conditions of the bodily fluid sample" refers to information regarding the conditions and the equipment used when collecting the bodily fluid. Such information includes the type of blood collection needle used when a needle is used to collect the bodily fluid, the type of blood collection tube used to collect the blood when the bodily fluid is blood or a blood-derived component, the type of coagulation promoter or coagulation inhibitor added to the blood collection tube, and the type of blood vessel when the blood was collected (capillary blood or venous blood), etc.

[0023] The "conditions of treatment applied to the body fluid sample" refer to the conditions of various treatments and operations applied to the body fluid sample before the measurement of the disease marker, and also include the conditions of storage of the body fluid sample. The conditions of storage include, for example, the time and temperature until freezing or transferring to a freezer, the temperature during freezing, the time of freezing, the time until decomposition enzymes in the body fluid sample, such as RNase, DNase, or Protease, are inactivated, the material of the sample storage container during freezing, and the use or non-use of packaging materials to cushion the impact of temperature changes at the start of freezing. In addition, the conditions of various treatments and operations that can be applied to the body fluid sample include the time from collection of the body fluid from the subject to centrifugation, and the centrifugal acceleration, when the body fluid is centrifuged or when the collected blood is centrifuged to obtain serum, which is the actual body fluid sample, etc. In addition, when the actual body fluid specimen is obtained by subjecting the collected body fluid to centrifugation, the time until the body fluid specimen is frozen or transferred to a freezer after centrifugation is also an example of the condition of the treatment applied to the body fluid specimen. Further, in order to measure a disease marker from a stored body fluid specimen, the time until the body fluid specimen is thawed from a frozen state and the temperature at which the body fluid specimen is thawed are also examples of the condition of the treatment applied to the body fluid specimen.

[0024] The preparation data may include at least one of the above information, and may include two or more, three or more, or four or more.

[0025] "Clinical data" refers to data indicating whether or not a subject from whom a body fluid sample was collected has a disease. In the learning data described below, the presence or absence of a disease may be diagnosed by a histopathological test.

[0026] [Configuration of the inspection device] Next, an inspection device according to this embodiment will be described with reference to Figs. 1 to 3. Fig. 1 is a functional block diagram showing a schematic configuration of an inspection device and a terminal device according to this embodiment. As shown in Fig. 1, the inspection device 100 is configured to communicate with a terminal device 200 including a display unit, an input device, and the like. However, the inspection device 100 is not limited to the configuration of this embodiment, and for example, the inspection device 100 may include a display unit, an input device, and the like by itself without communicating with the terminal device 200.

[0027] The testing device 100 includes a control unit 110, a storage unit 120, and a communication unit 130. The control unit 110 includes an acquisition unit (learning data acquisition unit, specimen data acquisition unit) 140, a model generation unit 150, and a discrimination unit 160. The acquisition unit 140 includes a marker data acquisition unit 141, a preparation data acquisition unit 142, and a clinical data acquisition unit 143.

[0028] The control unit 110 controls the inspection device 100 in an overall manner.

[0029] The storage unit 120 is a storage device that stores data necessary for processing by the inspection device 100. The storage unit 120 also stores a trained model 121. The storage unit 120 may be an external device of the inspection device 100. For example, the storage unit 120 may be a storage device such as a server that is communicably connected to the inspection device 100.

[0030] The marker data acquisition unit 141 acquires marker data indicating the results of measuring disease markers in a body fluid sample. The preparation data acquisition unit 142 acquires preparation data indicating the preparation conditions of the body fluid sample. Then, the clinical data acquisition unit 143 acquires clinical data indicating the presence or absence of a disease in the subject from whom the body fluid sample was collected. The acquisition unit 140 stores the marker data, preparation data, and clinical data for the same body fluid sample in the memory unit 120 in association with each other.

[0031] The model generation unit 150 performs machine learning using marker data, preparation data corresponding to the marker data, and clinical data corresponding to the marker data, and generates a trained model 121 that has learned the correlation between the set of marker data and preparation data corresponding to the marker data, and the clinical data. Various parameters that define the trained model 121 are stored in the storage unit 120. The specific algorithm for generating the trained model 121 is not particularly limited, and a neural network, a decision tree, a random forest, gradient boosting, bagging, a support vector machine (SVM), mapping, clustering, a self-organizing map (SOM), a local search method, an expectation maximization method, a multivariate adaptive regression spline method, a Bayesian network, a kernel density estimation, a principal component analysis, a Gaussian mixture model, construction of a sequential covering rule, a gradient boosting decision tree, and deep learning including a convolutional neural network, etc., can be used.

[0032] The discrimination unit 160 uses the trained model 121 stored in the storage unit 120 to discriminate whether or not the subject on which the input marker data and the corresponding preparation data are based suffers from a target disease. Specifically, the discrimination unit 160 inputs the marker data and the corresponding preparation data to the trained model 121 stored in the storage unit 120, and acquires a discrimination result output from the trained model 121. The acquired discrimination result is transmitted to the terminal device 200 via the communication unit 130. Note that the discrimination result output from the trained model 121 may be output as a numerical value within a certain numerical range. In this case, if the numerical value as the discrimination result exceeds a pre-specified threshold, the discrimination unit 160 discriminates that the specimen to be tested is positive (has a disease). On the other hand, if the numerical value is equal to or less than the threshold, the discrimination unit 160 discriminates that the specimen is negative (has no disease). The discrimination unit 160 may then transmit the result to the terminal device 200. Alternatively, the numerical value output from the trained model 121 may be directly transmitted to the terminal device 200, and the user may determine whether the sample is positive or negative based on a predetermined criterion. In addition, a plurality of thresholds designated in advance may be provided in stages. In this case, the discrimination unit 160 may determine the degree of disease risk of the sample to be tested in a plurality of stages, for example, depending on which of the plurality of thresholds the numerical value as the discrimination result exceeds. For example, a first threshold and a second threshold lower than the first threshold may be set as thresholds, and the discrimination unit 160 may determine the risk as high if the first threshold is exceeded, the discrimination unit 160 may determine the risk as medium if the numerical value is equal to or less than the first threshold and exceeds the second threshold, and the discrimination unit 160 may determine the risk as low if the numerical value is equal to or less than the second threshold.

[0033] In the present embodiment, the inspection device 100 is described in which the control unit 110 includes the model generating unit 150 and the acquisition unit 140 includes the clinical data acquisition unit 143. However, the inspection device 100 does not necessarily have to be provided with the model generating unit 150 and the clinical data acquisition unit 143. That is, the above-mentioned trained model 121 may be constructed by a learning device that includes at least the acquisition unit 140, which includes the marker data acquisition unit 141, the preparation data acquisition unit 142, and the clinical data acquisition unit 143 and exists independently of the inspection device 100, and the model generating unit 150. When the learning device exists independently of the inspection device 100, the inspection device 100 can use the trained model 121 by reading the trained model 121 stored in a storage medium. Alternatively, the inspection device 100 can use the trained model 121 by receiving the trained model 121 from another device via a wired or wireless network.

[0034] The terminal device 200 includes a communication unit 210, a control unit 220, an input device 230, and a display unit 240. The communication unit 210 is a communication interface that transmits and receives data to and from the inspection device 100 via wired or wireless communication. The control unit 220 performs overall control of the terminal device 200. The display unit 240 is a display that can display images, characters, and the like. The input device 230 accepts user input, and is realized by, for example, a touch panel, a mouse, a keyboard, and the like. When the input device 230 is a touch panel, the touch panel is provided on the display unit 240. A user can use the functions of the inspection device 100 via the terminal device 200.

[0035] [Operation of the inspection device] The inspection device 100 executes a learning phase (corresponding to a learning method) in which a trained model is generated, and an inspection phase (corresponding to an inspection method) in which inspection is performed using the trained model generated in the learning phase.

[0036] FIG. 2 is a flowchart showing the flow of processing by the control unit 110 in the learning phase. Note that some steps may be executed in parallel or in a different order. The control unit 110 acquires marker data obtained by measuring disease markers in a body fluid sample, which is used as learning data (S10; learning data acquisition step). The control unit 110 also acquires preparation data, which is information related to the preparation of the body fluid sample, which is used as learning data (S11; learning data acquisition step). The control unit 110 also acquires clinical data, which is used as learning data and indicates the presence or absence of a disease in a subject from whom the body fluid sample was collected (S12; learning data acquisition step). These learning data may be, for example, data that a user has stored in advance in the storage unit 120, or data that a user has uploaded to the testing device 100 by operating the terminal device 200. If the learning data is not stored in the storage unit 120, the control unit 110 stores each data acquired by the acquisition unit 140 in the storage unit 120. The model generation unit 150 generates a trained model 121 by machine learning the correlation between the set of the marker data and the preparation data and the clinical data using the marker data, the preparation data corresponding to the marker data, and the clinical data, and stores the trained model 121 in the storage unit 120 (S13; model generation step). This completes the learning phase.

[0037] FIG. 3 is a flowchart showing the flow of processing by the control unit 110 in the testing phase. Note that some steps may be executed in parallel or in a different order. The processing shown in FIG. 3 is started, for example, when a user performs an operation to instruct execution of a testing process. The control unit 110 acquires marker data as testing data (S20; specimen data acquisition step). The control unit 110 also acquires preparation data corresponding to the acquired marker data as testing data (S21; specimen data acquisition step). The testing data may be, for example, data that the user has stored in advance in the storage unit 120, or data that the user has uploaded to the testing device 100 by operating the terminal device 200. If the testing data is not stored in the storage unit 120, the control unit 110 stores each data acquired by the acquisition unit 140 in the storage unit 120. The discrimination unit 160 inputs the acquired marker data and preparation data to the trained model 121 (S22; discrimination step), and acquires a discrimination result output from the trained model 121 (S23; discrimination step). The discrimination unit 160 transmits output data indicating the obtained discrimination result to the terminal device 200 via the communication unit 130 (S24). This completes the inspection phase. The terminal device 200 causes the display unit 240 to display the output data received via the communication unit 210.

[0038] Before executing the above-mentioned process, the user needs to obtain "marker data," "preparation data," and "clinical data." "Preparation data" and "clinical data" are usually obtained at the medical institution where the specimen was collected. On the other hand, "marker data" is usually obtained at the medical institution where the specimen was collected or at a testing institution different from the medical institution. Therefore, if the user belongs to a testing institution, the user is provided with "preparation data" and "clinical data" in advance from the medical institution.

[0039] The learning phase process and the inspection phase process in the inspection device 100 do not necessarily have to be performed consecutively. That is, only the learning phase process may be performed in advance, and the inspection device 100 may wait until the inspection phase process is required. Also, the learning phase process may be performed in the above-mentioned learning device, and the inspection device 100 may perform only the inspection phase process. In this case, the inspection device 100 may acquire in advance from the learning device the trained model generated in the learning device.

[0040] As described above, the testing device 100 in this embodiment uses the marker data and the preparation data to determine the presence or absence of a disease, but it is also possible to use additional data (hereinafter, additional data). For example, a disease marker or test value that is conventionally used as an indicator of the presence of a disease that is different from the disease to be tested and is not directly related to the disease to be tested may be used in combination with the above-mentioned preparation data. As a more detailed example, when the testing device 100 is used to determine the presence or absence of pancreatic cancer or biliary tract cancer, the amount of blood cells (red blood cell amount, white blood cell amount, or platelet amount) contained in the blood of the subject, the amount of protein disease markers such as PSA and CYFRA, and test values ​​such as the amount of HDL cholesterol and the amount of LDL cholesterol may be used as additional data in combination with the preparation data. When such additional data is used, the control unit 110 acquires the additional data in the same manner as the preparation data. Then, the model generation unit 150 may generate a trained model by machine learning the correlation with clinical data after adding the additional data to a set of marker data and preparation data corresponding to the marker data. In the inspection phase, the control unit 110 also acquires additional data as inspection data. The discrimination unit 160 then uses the additional data as input to the trained model to acquire a discrimination result.

[0041] [Effects of this embodiment] As described above, in this embodiment, in a disease testing method using a trained model in which marker data is input and the presence or absence of a disease is output, the trained model is created using an index (prepared data) unrelated to the presence or absence of a disease as one piece of input information. Surprisingly, the accuracy of the discrimination result indicating the presence or absence of a disease can be improved by using an index unrelated to the presence or absence of a disease.

[0042] The acquisition of preparation data does not impose additional burdens, such as strict adherence to detailed protocols, on the medical site that actually samples the specimen. In other words, it is sufficient to simply record what protocol or conditions were used for sampling, while allowing for a range of possible specimen collection conditions, and provide the information to the user who actually performs the test. In addition, this information can be easily acquired or confirmed even after the marker data has been acquired. Therefore, the burden on the medical site is extremely small when the test method according to this embodiment is implemented. In addition, even if disease markers that can fluctuate during a relatively short operation time are included in the analysis target, highly accurate discrimination results can be obtained without excluding them.

[0043] In particular, when a test is performed using a plurality of disease markers, the state of the disease marker fluctuation that may occur due to the difference in parameters corresponding to the preparation data may differ for each disease marker. Therefore, when using a plurality of disease markers to improve the test accuracy, it is necessary to reduce the disease marker fluctuation, and more strict compliance with the protocol is required. However, according to this embodiment, since it is only necessary to record information, it can be more suitably applied to the test using a plurality of disease markers.

[0044] [Software implementation example] The control blocks of the inspection device 100 (the control unit 110, in particular the acquisition unit 140, the model generation unit 150 and the discrimination unit 160) may be realized by a logic circuit (hardware) formed in an integrated circuit (IC chip) or the like, or may be realized by software.

[0045] In the latter case, the inspection device 100 includes a computer that executes instructions of a program, which is software that realizes each function. The computer includes, for example, at least one processor (control device) and at least one computer-readable recording medium that stores the program. The object of the present invention is achieved by having the processor read the program from the recording medium and execute it in the computer. The processor may be, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The recording medium may be a "non-transient tangible medium", such as a ROM (Read Only Memory), a tape, a disk, a card, a semiconductor memory, a programmable logic circuit, or the like. The device may further include a RAM (Random Access Memory) that expands the program. The program may be supplied to the computer via any transmission medium (such as a communication network or a broadcast wave) that can transmit the program. Note that one aspect of the present invention may also be realized in the form of a data signal embedded in a carrier wave, in which the program is embodied by electronic transmission.

[0046] 〔summary〕 The testing method of the present invention is a testing method for testing for a disease using a disease marker, and includes a specimen data acquisition step of acquiring marker data indicating the results of measuring a disease marker in a body fluid sample collected from a subject, and preparation data indicating the preparation conditions of the body fluid sample, and a step of determining the presence or absence of a disease in the subject by inputting the marker data and preparation data acquired in the specimen data acquisition step into a trained model that has been machine-learned to determine the correlation between a set of marker data indicating the results of measuring a disease marker in a body fluid sample and the preparation data of the body fluid sample, and the presence or absence of a disease in the subject from whom the body fluid sample was collected.

[0047] In addition, in one embodiment of the testing method of the present invention, the preparation data includes at least one piece of information selected from information on the subject, information indicating the collection conditions of the body fluid sample, and condition information on the treatment performed on the body fluid sample before measuring the disease marker.

[0048] In one embodiment of the testing method according to the present invention, the conditions of the treatment performed on the body fluid sample are at least one condition selected from the group consisting of the time until the body fluid sample is frozen and stored, the temperature during the frozen storage, the time for which the body fluid sample is frozen and stored, and the time until the body fluid sample is centrifuged.

[0049] In one aspect of the testing method of the present invention, the information indicating the collection conditions of the body fluid sample is at least one piece of information selected from the thickness of the needle used to collect the body fluid sample, the type of blood collection tube used to collect the body fluid sample, and the time between the subject's last meal and the collection of the body fluid sample.

[0050] In one aspect of the testing method according to the present invention, the information on the subject is information on the race of the subject.

[0051] In one aspect of the testing method according to the present invention, the body fluid sample is blood, serum, plasma, cerebrospinal fluid, urine, saliva, tears, tissue fluid, or lymphatic fluid.

[0052] In a further aspect of the testing method according to the present invention, the body fluid sample is blood, serum or plasma.

[0053] In one embodiment of the testing method according to the present invention, the disease marker is miRNA.

[0054] In one aspect of the testing method according to the present invention, the marker data is data obtained from a microarray, PCR or sequencing.

[0055] Furthermore, one aspect of the testing method according to the present invention is configured to further include a training data acquisition step of acquiring marker data indicating the results of measuring a disease marker in a body fluid sample, preparation data indicating preparation conditions for the body fluid sample, and clinical data indicating the presence or absence of a disease in a subject from whom the body fluid sample was collected, and a model generation step of generating the trained model by machine learning the correlation between the set of the marker data and the preparation data acquired in the training data acquisition step and the clinical data.

[0056] The learning method of the present invention is a learning method for generating a trained model to be used in testing for a disease using a disease marker, and includes a training data acquisition step of acquiring marker data indicating the results of measuring a disease marker in a body fluid sample, preparation data indicating preparation conditions for the body fluid sample, and clinical data indicating the presence or absence of a disease in a subject from whom the body fluid sample was collected, and a model generation step of generating a trained model capable of determining the presence of the disease from the marker data and the preparation data by machine learning the correlation between the acquired set of marker data and preparation data and the clinical data.

[0057] The learning device of the present invention is a learning device that generates a trained model used in testing for a disease using a disease marker, and is configured to include a learning data acquisition unit that acquires marker data indicating the results of measuring a disease marker in a body fluid sample, preparation data indicating the preparation conditions of the body fluid sample, and clinical data indicating the presence or absence of a disease in a subject from whom the body fluid sample was collected, and a model generation unit that generates a trained model capable of determining the presence of the disease from the marker data and the preparation data by machine learning the correlation between the acquired set of marker data and preparation data and the clinical data.

[0058] The inspection device according to the present invention is an inspection device that performs disease inspection using disease markers, and includes a specimen data acquisition unit that acquires the marker data and the preparation data in a body fluid specimen derived from a subject, and the learning device described above. A discrimination unit that discriminates the presence or absence of the disease from the marker data and the preparation data acquired by the specimen data acquisition unit using the learned model generated by the above.

[0059] The inspection device and the learning device according to each aspect of the present invention may be realized by a computer. In this case, an inspection program or a learning program that realizes the inspection device or the learning device by operating the computer as each part (software element) included in the inspection device or the learning device, and a computer-readable recording medium on which it is recorded are also within the scope of the present invention.

[0060] Examples are shown below to explain the embodiments of the present invention in more detail. Of course, the present invention is not limited to the following examples, and it goes without saying that various aspects are possible for the details. Furthermore, the present invention is not limited to the above-described embodiments, various modifications are possible within the scope shown in the claims, and embodiments obtained by appropriately combining the disclosed technical means are also included in the technical scope of the present invention.

Example

[0061] [Reference Example 1] [Sampling of Specimen] Informed consent was obtained from 48 patients with pancreatic cancer or biliary tract cancer diagnosed by histopathological examination and 48 healthy subjects, and their blood was collected, and serum was obtained by centrifugation. The obtained serum was transferred to a freezer at -80°C within 2 hours after blood collection and stored at -80°C until use. There was a certain variation among the specimens in the time required from blood collection to centrifugation and the time required from centrifugation to storage at -80°C.

[0062] [Extraction of total RNA] As samples, 300 μL of each serum obtained from each of the 96 individuals was used. Total RNA was obtained from each serum using the RNA extraction reagent in the 3D-Gene® RNA extraction reagent from liquid sample kit (Toray Industries, Inc., Japan) according to the protocol specified by the company. Hereinafter, the sample group in Reference Example 1 is referred to as sample group 1.

[0063] <Measurement of gene expression levels> The total RNA obtained from each serum of the above 96 people was fluorescently labeled with miRNA in the total RNA using 3D-Gene (registered trademark) miRNA Labeling kit (Toray Industries, Inc.) according to the protocol specified by the company. As an oligo DNA chip, 3D-Gene (registered trademark) Human miRNA Oligo chip (Toray Industries, Inc.) equipped with a probe having a sequence complementary to miRNA registered in miRBase release 21 was used, and hybridization and post-hybridization washing were performed under stringent conditions according to the protocol specified by the company. The DNA chip was scanned using a 3D-Gene (registered trademark) scanner (Toray Industries, Inc.), and the image was acquired, and the fluorescence intensity was quantified using 3D-Gene (registered trademark) Extraction (Toray Industries, Inc.).

[0064] The expression levels of the detected genes were calculated as follows using the digitized fluorescence intensities. First, excluding the top 5% and bottom 5% of the signal intensities of multiple negative control spots respectively, the [mean value + 2 × standard deviation] was calculated, and genes showing signal intensities greater than this value were considered detected. Also, from the signal intensities of the detected genes, the mean value of the signal intensities of the negative control spots excluding the top 5% and bottom 5% respectively was subtracted, and the subtracted value was converted to a base-2 logarithmic value as the gene expression level. Data normalization was performed using the mean values of three miRNAs reported as endogenous controls: miR-149-3p, miR-2861, miR-4463 (Non-Patent Document: Shimomura et al., Cancer Science, 2016, Vol. 107, pages 326 - 34) to normalize the data of different specimens. For genes not detected as described above, they were replaced with a base-2 logarithmic value of 0.1. In this way, the signal values of the miRNA gene expression levels for the above 96 sera were obtained.

[0065] Calculations and statistical analyses using the digitized miRNA gene expression levels were performed using R language 3.5.3 (R Foundation for Statistical Computing, https: / / www.R-project.org), caret package 6.0 - 84, and xgboost package 0.82.1.

[0066] [Reference Example 2] [Sample Collection] Informed consent was obtained from 24 healthy individuals, and their respective blood samples were collected. Serum was obtained by centrifugation. The obtained serum was transferred to a -80°C freezer within 6 hours after blood collection and stored at -80°C until use. Note that there was a certain variation among the samples in the time required from blood collection to centrifugation and the time required from centrifugation to storage at -80°C.

[0067] [Extraction of total RNA] As specimens, 300 μL of serum obtained from each of the 24 individuals was used, and total RNA was obtained in the same manner as in Reference Example 1. Hereinafter, the specimen group in Reference Example 2 is referred to as specimen group 2.

[0068] <Measurement of gene expression levels> Signal values ​​of gene expression levels of miRNA were obtained in the same manner as in Reference Example 1 using total RNA obtained from the serum of each of the above 24 individuals.

[0069] [Example 1] <Verification of pancreatic and biliary cancer discrimination performance using machine learning 1: Using 45 types of miRNA> In this example, a gradient boosting decision tree was used as an example of machine learning. Machine learning was performed using clinical data indicating whether a patient was suffering from cancer or healthy, the signal value of the gene expression level of each miRNA, the time required from blood collection to centrifugation, and the time required from centrifugation to storage at -80°C, and a discrimination model (trained model) for classifying patients as suffering from cancer or healthy was created and verified. In this example, 45 types of miRNAs that were shown to be associated with pancreatic cancer or biliary tract cancer in a prior study were arbitrarily selected as the miRNAs to be used.

[0070] The following stepwise procedure was taken to distinguish between pancreatic cancer and biliary tract cancer. Specifically, the data from sample group 1 was divided 6:4 to create a training sample group and a validation sample group. A discrimination model was obtained from the training sample group using gradient boosting decision trees, and this was verified using the validation sample group. As a result of the verification, the sensitivity for determining pancreatic cancer and biliary tract cancer as positive was 94.4%, the specificity for determining healthy individuals as negative was 95.2%, and the accuracy was 94.9%. Here, "sensitivity" refers to the proportion of positives correctly classified as positive. High sensitivity enables early detection of disease. Also, "specificity" refers to the proportion of negatives correctly classified as negative. High specificity prevents unnecessary additional testing, patient transfer, and treatment due to misclassification of healthy individuals as having the disease, leading to reduced burden on patients and reduced medical costs. Accuracy indicates the proportion of correct discrimination results for all samples, and is the first indicator for evaluating test performance.

[0071] [Example 2] <Verification of pancreatic and biliary cancer discrimination performance using machine learning 2: Using 78 types of miRNA> A discrimination model was created and verified in the same manner as in Example 1, except that 78 types of miRNAs that had been shown to be associated with pancreatic cancer or biliary tract cancer in a prior study were arbitrarily selected as the miRNAs to be used. As a result of the verification, the sensitivity for determining pancreatic cancer and biliary tract cancer as positive was 88.9%, the specificity for determining healthy individuals as negative was 95.2%, and the accuracy was 92.3%.

[0072] [Example 3] <Verification of pancreatic and biliary cancer discrimination performance using machine learning 3: Using 45 types of miRNA> In this example, the discrimination model created in Example 1 was verified using data from sample group 2 instead of data from sample group 1. As a result of the verification, the specificity for determining healthy individuals as negative was 87.5%.

[0073] [Example 4] <Verification of pancreatic and biliary cancer discrimination performance using machine learning 4: Using 78 miRNAs> In this example, the discrimination model created in Example 2 was verified using data from sample group 2 instead of data from sample group 1. As a result of the verification, the specificity for determining healthy individuals as negative was 100.0%.

[0074] [Comparative Example 1] <Verification of pancreatic and biliary cancer discrimination performance using machine learning 5: Using 45 types of miRNA> In this comparative example, a gradient boosting decision tree was used as an example of machine learning. Machine learning was performed using clinical data indicating whether a patient was suffering from cancer or was healthy, and the signal value of the gene expression level of each miRNA, and a discrimination model for classifying patients as suffering from cancer or being healthy was created and verified. The miRNAs used were the same 45 types as in Example 1.

[0075] The following stepwise procedure was taken to distinguish between pancreatic cancer and biliary tract cancer. That is, the data of sample group 1 was divided 6:4 to create a learning sample group and a validation sample group. A discrimination model was obtained from the learning sample group using a gradient boosting decision tree, and verified from the validation sample group. As a result of the verification, the sensitivity for determining pancreatic cancer and biliary tract cancer as positive was 88.9%, the specificity for determining healthy subjects as negative was 95.2%, and the accuracy was 92.3%. That is, it was shown that the sensitivity and accuracy were inferior to those of Example 1.

[0076] [Comparative Example 2] <Verification of pancreatic and biliary cancer discrimination performance using machine learning 6: Using 78 miRNAs> A discrimination model was created and verified in the same manner as in Comparative Example 1, except that the miRNAs used were changed to 78 types, the same as in Example 2. As a result of the verification, the sensitivity for determining pancreatic cancer and biliary tract cancer as positive was 77.8%, the specificity for determining healthy subjects as negative was 90.5%, and the accuracy was 86.4%. In other words, it was shown that the sensitivity, specificity, and accuracy were all inferior to those of Example 2.

[0077] [Comparative Example 3] <Verification of pancreatic and biliary cancer discrimination performance using machine learning 7: Using 45 types of miRNA> In this comparative example, the discrimination model created in Comparative Example 1 was verified using data from Sample Group 2 instead of data from Sample Group 1. As a result of the verification, the specificity for determining healthy subjects as negative was 33.3%. In other words, it was shown that the specificity was inferior to that of Example 3.

[0078] [Comparative Example 4] <Verification of pancreatic and biliary cancer discrimination performance using machine learning 8: Using 78 miRNAs> In this comparative example, the discrimination model created in Comparative Example 2 was verified using data from Sample Group 2 instead of data from Sample Group 1. As a result of the verification, the specificity for determining healthy subjects as negative was 79.2%. In other words, it was shown that the specificity was inferior to that of Example 3.

[0079] The above results are shown in Tables 1 and 2.

[0080] [Table 1]

[0081] [Table 2] [Industrial Applicability]

[0082] The present invention can be used in disease testing using disease markers. [Explanation of symbols]

[0083] 100 Inspection Equipment 110 Control section 120 Storage section 121 trained models 130 Communications Department 140 Acquisition Department 141 Marker Data Acquisition Unit 142 Preparation Data Acquisition Unit 143 Clinical Data Acquisition Department 150 Model Generation Unit 160 Discrimination part

Claims

1. A method for testing for diseases including pancreatic cancer and biliary tract cancer using a disease marker including miRNA, comprising: a sample data acquisition step of acquiring marker data indicating the results of measuring the disease markers in a body fluid sample including blood, serum, or plasma collected from a subject, and preparation data indicating the preparation conditions of the body fluid sample; a discrimination step of discriminating whether or not the subject has a disease by inputting the marker data and the preparation data acquired in the specimen data acquisition step into a trained model that has been machine-learned to determine a correlation between a set of marker data indicating a result of measuring a disease marker in a body fluid sample and preparation data indicating preparation conditions of the body fluid sample, and the presence or absence of a disease in the subject from which the body fluid sample was collected, the preparation data includes condition information of a treatment performed on the body fluid sample before the measurement of the disease marker; The testing method, wherein the conditions of the treatment performed on the body fluid sample are at least one condition selected from the group consisting of the time until the body fluid sample is frozen and stored, the temperature during the frozen storage, the time for which the body fluid sample is frozen and stored, and the time until the body fluid sample is centrifuged.

2. The testing method according to claim 1 , wherein the preparation data further includes at least one piece of information selected from information on the subject and information indicating collection conditions for the body fluid sample.

3. 3. The testing method of claim 2, wherein the information indicating the collection conditions of the body fluid sample is at least one piece of information selected from the thickness of the needle used to collect the body fluid sample, the type of blood collection tube used to collect the body fluid sample, and the time between the subject's last meal and the collection of the body fluid sample.

4. The testing method according to claim 2 , wherein the information on the subject is information on the race of the subject.

5. The testing method according to claim 1 , wherein the marker data is data obtained from a microarray, PCR or sequencing.

6. a learning data acquisition step of acquiring marker data indicating the results of measuring disease markers in a body fluid sample, preparation data indicating preparation conditions of the body fluid sample, and clinical data indicating the presence or absence of a disease in a subject from whom the body fluid sample was collected; The method according to any one of claims 1 to 5, further comprising a model generation step of generating the trained model by machine learning a correlation between the set of the marker data and the preparation data acquired in the training data acquisition step and the clinical data.

7. A learning method for generating a trained model for use in testing for diseases including pancreatic cancer and biliary tract cancer using a disease marker including miRNA, comprising: marker data showing the results of measuring the disease markers in a body fluid sample including blood, serum, or plasma; a learning data acquisition step of acquiring preparation data indicating preparation conditions of the body fluid sample and clinical data indicating the presence or absence of a disease in the subject from whom the body fluid sample was collected; and a model generation step of generating a trained model capable of discriminating the onset of the disease from the marker data and the preparation data by machine learning a correlation between the acquired set of the marker data and the preparation data and the clinical data, the preparation data includes condition information of a treatment performed on the body fluid sample before the measurement of the disease marker; A learning method, wherein the conditions of the treatment performed on the body fluid sample are at least one condition selected from the time until the body fluid sample is frozen and stored, the temperature during freezing and storage, the time for which the body fluid sample is frozen and stored, and the time until the body fluid sample is centrifuged.

8. A learning device that generates a trained model used in testing for diseases including pancreatic cancer and biliary tract cancer using a disease marker including miRNA, a learning data acquisition unit that acquires marker data indicating the results of measuring the disease markers in a body fluid sample including blood, serum, or plasma, preparation data indicating preparation conditions for the body fluid sample, and clinical data indicating the presence or absence of a disease in a subject from whom the body fluid sample was collected; A model generation unit that generates a trained model capable of discriminating the onset of the disease from the marker data and the preparation data by performing machine learning on a correlation between the acquired set of the marker data and the preparation data and the clinical data, the preparation data includes condition information of a treatment performed on the body fluid sample before the measurement of the disease marker; A learning device, wherein the conditions of the treatment performed on the body fluid sample are at least one condition selected from the time until the body fluid sample is frozen and stored, the temperature during freezing and storage, the time for which the body fluid sample is frozen and stored, and the time until the body fluid sample is centrifuged.

9. A testing device for testing for diseases including pancreatic cancer and biliary tract cancer using a disease marker including miRNA, a specimen data acquisition unit for acquiring the marker data and the preparation data for a body fluid specimen including blood, serum, or plasma derived from a subject; A discrimination unit that discriminates the onset of the disease from the marker data and the preparation data acquired by the specimen data acquisition unit using the trained model generated by the learning device according to claim 8, the preparation data includes condition information of a treatment performed on the body fluid sample before the measurement of the disease marker; A testing device in which the conditions of the treatment performed on the body fluid sample are at least one condition selected from the following: the time until the body fluid sample is frozen and stored, the temperature during freezing and storage, the time for which the body fluid sample is frozen and stored, and the time until the body fluid sample is centrifuged.

10. 9. A learning program for causing a computer to function as the learning device according to claim 8, the learning program causing a computer to function as the learning data acquisition unit and the model generation unit.

11. 10. A testing program for causing a computer to function as the testing apparatus according to claim 9, the testing program causing a computer to function as the sample data acquisition unit and the discrimination unit.

Images