WO2022039092A1

WO2022039092A1 - Learning model generation method, program, and calculation device

Info

Publication number: WO2022039092A1
Application number: PCT/JP2021/029682
Authority: WO
Inventors: 孝章赤池; 智史高木
Original assignee: 孝章赤池; バイオ・アクセラレーター株式会社
Priority date: 2020-08-17
Filing date: 2021-08-11
Publication date: 2022-02-24
Also published as: JP2023145811A

Abstract

This learning model generation method is for creating a learning model for making a diagnosis of a degree of infection by a pathogen. The learning model generation method involves: acquiring a plurality of teaching data items that each include information of a protein obtained from a sample collected from a sample provider and a degree of infection at which the sample provider is infected with the pathogen; and generating, by using the teaching data items, a learning model to which the information of the protein is to be inputted and from which the degree of infection by the pathogen is outputted.

Description

Learning model generation method, program, arithmetic unit

The present invention relates to a learning model generation method, a program, and an arithmetic unit.

There is an increasing demand for a method for accurately determining the presence or absence of infection with a pathogen. Patent Document 1 describes a method for diagnosing a microbial infection in an organism, wherein the microbial infection is at least partially present as a basic body in the cell material of the organism, and the diagnostic method is a method of diagnosing the cell material from the organism. A step of obtaining a sample and a step of processing the sample to obtain a test composition rich in the basic body as long as the sample contains the basic body, and a predetermined maximum number of basic body bodies at most. Disclosed is a method comprising the step of subjecting a volume of the test composition comprising the test composition to a mass analysis method to identify the presence of the microbial infection.

Japan Special Table 2020-510829 Gazette

In the invention described in Patent Document 1, since the pathogen itself is the detection target, there is room for improvement in the detection accuracy of the infection to the pathogen.

The method for generating a learning model according to the first aspect of the present invention is a method for creating a learning model for diagnosing the degree of infection with a pathogen, and includes information on proteins obtained from a sample collected from a sample provider and the sample. A plurality of teacher data including the degree of infection with the pathogen of the donor are acquired, and the information of the protein is input using the teacher data, and a learning model is generated in which the degree of infection with the pathogen is output.
The program according to the second aspect of the present invention acquires evaluation data which is information on the protein of the sample collected from the evaluation subject, inputs the information on the protein obtained from the sample collected from the sample provider, and provides the sample. The acquired evaluation data is input to a learning model trained using teacher data that outputs the degree of infection of the pathogen of the person, and the degree of infection of the pathogen of the evaluation target person is output. Let the computer perform the process.
The arithmetic unit according to the third aspect of the present invention inputs the reading unit that reads the evaluation data, which is the information on the protein of the sample collected from the evaluation subject, and the information on the protein obtained from the sample collected from the sample provider. A storage unit that stores a learning model trained using teacher data that outputs the degree of infection of the pathogen of the sample provider, and inputting the evaluation data into the learning model, the evaluation target person It is provided with a calculation unit that outputs the degree of infection with the pathogen.

According to the present invention, it is possible to create a learning model with high detection accuracy of infection with a pathogen.

Overall configuration diagram of the infection determination system in the first embodiment The figure which shows an example of the teacher data in 1st Embodiment Overall configuration diagram of the infection determination system in the second embodiment Overall configuration diagram of the infection determination system in the third embodiment The figure which shows the analysis result in 4th Embodiment The figure which shows the measurement result in 4th Embodiment Configuration diagram of arithmetic unit in the fourth embodiment The figure which shows an example of the teacher data in 4th Embodiment

This specification focuses on the presence of proteins in determining infection with a pathogen. Pathogens replicate genetic information such as DNA and RNA in the host's body, but produce various proteins for self-replication. In order to confirm the existence of a pathogen, it is effective to detect the genetic information of the replicated pathogen as in the prior art. However, in order to confirm that the pathogen does not exist, it is necessary to confirm that the gene for the pathogen does not exist, and if the sample does not contain the pathogen for some reason or if the pathogen contains less than the detection sensitivity. Is mistakenly determined not to exist. On the other hand, the number of proteins produced by pathogens and proteins produced by pathogens in host cells is overwhelmingly large compared to genetic information, and can be a feature indicating the presence of pathogens.

However, it is very difficult for humans to grasp the relationship with pathogens because the types of proteins are enormous and easily changed by other influences. Therefore, in the present specification, the relationship between the degree of infection with a pathogen and a protein is learned by machine learning, and inference is performed using the learning model obtained by this learning.

-First embodiment-
Hereinafter, the first embodiment of the learning model generation method will be described with reference to FIGS. 1 and 2.

FIG. 1 is an overall configuration diagram of an infection determination system S for determining infection with a pathogen P. The pathogen P is not particularly limited and may be a virus, a bacterium, or a fungus. The infection determination system S includes an arithmetic unit 10, a pretreatment device 92, a mass spectrometer 93, and a gene analysis device 94. The arithmetic unit 10 includes an arithmetic unit 1, a communication unit 2, and a storage unit 3. The storage unit 3 is a non-volatile storage device such as a hard disk drive. The storage unit 3 stores the teacher data 31, the learning model 32, the learning model program 32A, and the evaluation data 33.

The arithmetic unit 1 is, for example, a central processing unit (not shown), and performs learning processing and inference processing described later by expanding and executing a program stored in a read-only memory (not shown) in a volatile memory (not shown). .. However, the arithmetic unit 1 may be realized by using a hardware circuit or a reconfigurable logic circuit, or may be realized by a combination thereof. The arithmetic unit 1 performs learning processing and inference processing. The learning process is the creation and updating of the learning model 32 using the teacher data 31. The inference process is the evaluation of the evaluation data 33 using the learning model 32. These processes will be described later.

The communication unit 2 is a communication module capable of communicating with the mass spectrometer 93, the gene analysis device 94, and the examination terminal 95. The communication unit 2 may be directly connected to the mass spectrometer 93, the gene analysis device 94, and the examination terminal 95, or may be connected via a communication network, for example, the Internet. Since the communication unit 2 acquires the evaluation data 33 from the mass spectrometer 93, the gene analysis device 94, and the examination terminal 95, it can also be called a “reading unit” that performs a process of reading the evaluation data 33 from the outside.

The teacher data 31 includes a plurality of sets of teacher protein information 311, teacher genome information 312, teacher medical examination information 313, and infection confirmation information 314. The teacher protein information 311 is the identification information of the protein detected by the mass spectrometer 93 and the quantitative information of the detected protein, and is created by the mass spectrometer 93. Quantitative information on a protein is either the specific amount of protein detected and the weight ratio of that protein to the total amount of total protein detected. The teacher genome information 312 is information created by the gene analysis device 94 and indicating whether or not the sample contains the known genome information of the pathogen P.

Teacher consultation information 313 is various objective and subjective information such as body temperature, cough, headache, myalgia, and fatigue. The infection confirmation information 314 is the presence or absence of infection with the pathogen P, the presence or absence of infectivity when infected, and the number of days until the infectivity is generated when there is no infectivity. Each infection confirmation information 314 is associated with teacher protein information 311, teacher genomic information 312, and teacher consultation information 313. The teacher data 31 is derived from the teacher sample 911 and the group 90 collected from the group 90 as described later.

FIG. 2 is a diagram showing an example of teacher data 31. The teacher protein information 311 describes the name of the protein and the weight ratio of the protein to the total amount of the protein. As described above, the weight may be described instead of the ratio, or the number of moles or the number of molecules may be described. In the example shown in FIG. 2, the presence or absence of the pathogen P is described in the teacher genomic information 312 under the assumption that the pathogen P has only DNA. When the pathogen P has DNA and RNA, the teacher genomic information 312 is provided with a DNA column and an RNA column independently. The teacher consultation information 313 of FIG. 2 describes body temperature, cough, and headache. The infection confirmation information 314 of FIG. 2 describes that the person is infected and currently has no infectivity, but it is about 3 days before the infectivity develops.

The learning model 32 includes one or more models such as a predictive model, a regression model, a stochastic model, an artificial neural network, and a long short-term memory (LSTM) neural network. For example, regression models include decision trees, classifiers, linear regression models, and so on. For example, probabilistic models include support vector machines, Markov models and hidden Markov models. Artificial neural networks include recurrent neural networks and the like. The learning model 32 may include a model for determining the presence or absence of infection with the pathogen P and a model for determining the presence or absence of infectivity for the pathogen P. The learning model 32 is created and updated using the teacher data 31. That is, the learning model 32 also includes various parameters required for the calculation.

The learning model 32 is a learning model in which learning is performed by inputting teacher protein information 311, teacher genome information 312, and teacher examination information 313 and outputting infection confirmation information 314 in the learning phase. Further, in the inference phase, the learning model 32 infers the degree of infection by inputting the evaluation protein information 331, the evaluation genome information 332, and the evaluation consultation information 333 described below.

The learning model program 32A is a program created for the convenience of executing inference processing using the learning model 32 and executed by the arithmetic unit 1. The learning model 32 is premised on being updated, and in order to execute the inference process, it is necessary to convert it into a format suitable for execution in the arithmetic unit 1. The learning model program 32A cuts out and clearly indicates a format suitable for this execution. The learning model program 32A may be optimized according to the hardware configuration of the arithmetic unit 1.

The evaluation data 33 is data to be inferred using the learning model 32. The evaluation data 33 includes the evaluation protein information 331, the evaluation genome information 332, and the evaluation consultation information 333. The evaluation protein information 331 is protein information obtained by processing the evaluation sample 912 collected from the evaluation subject 90A, and the type of information is the same as that of the teacher protein information 311. The evaluation genome information 332 is information indicating whether or not the pathogen P genomic information is included, which is obtained by processing the evaluation sample 912 collected from the evaluation subject 90A, and the type of information is the teacher genome information 312. Is the same as. The evaluation medical examination information 333 is information obtained by a medical examination for the evaluation target person 90A, and the type of information is the same as that of the teacher medical examination information 313.

Comparing the teacher data 31 and the evaluation data 33, the teacher data 31 is mainly generated using the teacher sample 911 collected from the group 90, whereas the evaluation sample 912 is an evaluation sample collected from the evaluation subject 90A. The difference is that it is generated using 912. Further, the information corresponding to the infection confirmation information 314 of the teacher data 31 is not included in the evaluation data 33. The information corresponding to the infection confirmation information 314 in the evaluation data 33 is output by inference using the learning model 32.

The group 90 is a group of humans of various genders and ages, including a person infected with the pathogen P targeted by the present embodiment. Each person belonging to the population 90 will have a teacher sample 911 taken from his or her body for a predetermined period, eg, every 3 months, in a predetermined cycle, eg, every day. The teacher specimen 911 is nasal mucosal fluid, pharyngeal mucosal fluid, saliva, exhaled breath condensate, and blood. It is sufficient that at least one person belonging to the population 90 is infected with the pathogen P during at least a part of the above-mentioned predetermined period. The mucous membrane is, for example, the mucous membrane of the eyes, ears, nose, mouth and the like. The group 90 can also be referred to as a "sample provider" in the sense that the learning model 32 provides sample data for learning.

The evaluation target 90A is a human whose infection with the pathogen P is unknown. For the evaluation target person 90A, the evaluation sample 912 is collected at least once. The evaluation sample 912 is nasal mucosal fluid, pharyngeal mucosal fluid, saliva, exhaled breath condensate, and blood. In the following, the teacher sample 911 and the evaluation sample 912 are collectively referred to as a "sample" 91.

The pretreatment device 92 is a device that performs pretreatment for detecting proteins, DNA, and RNA contained in the sample 91. Since the content of processing by the pretreatment device 92 is known, it will not be described in detail in this specification. The pretreatment device 92 may be fully automatic, or may be partially or wholly operated manually.

The mass spectrometer 93 is a device that measures the name and content ratio of the protein contained in the sample 91 by a known method. The mass spectrometer 93 may be, for example, either a liquid chromatograph mass spectrometer or a gas chromatograph mass spectrometer, or may include both. The mass spectrometer 93 transmits the measurement result to the storage unit 3 as teacher protein information 311 or evaluation protein information 331. The mass spectrometer 93 stores the measurement result as the teacher protein information 311 when the sample 91 is the teacher sample 911, and stores the measurement result as the evaluation protein information 331 when the sample 91 is the evaluation sample 912. The mass spectrometer 93 may be fully automated, or may be partially or wholly operated manually.

The gene analysis device 94 analyzes the gene contained in the sample 91 by using one or more known measurement methods such as a nucleic acid amplification method, a Q probe method, and a Tm analysis method. The gene analysis device 94 transmits information indicating whether or not the DNA or RNA of the pathogen P is detected from the sample 91 to the storage unit 3 as the teacher genomic information 312 or the evaluation genomic information 332. The gene analysis apparatus 94 stores the measurement result as the teacher genome information 312 when the sample 91 is the teacher sample 911, and stores the measurement result as the evaluation genome information 332 when the sample 91 is the evaluation sample 912. The gene analyzer 94 may be fully automatic, or may be partially or wholly operated manually.

The medical examination terminal 95 is a computer operated by the expert 96, and has a human interface and a communication module for communicating with the storage unit 3. The medical examination terminal 95 is, for example, a general-purpose personal computer or a smartphone. Expert 96 examines each of the groups 90 and creates teacher examination information 313. However, the teacher consultation information 313 can be added or modified later. Details will be described later.

(Creation of teacher data)
The method of creating the teacher data 31, that is, the method of creating the teacher protein information 311, the teacher genome information 312, and the teacher examination information 313 will be described. First, teacher specimens 911 such as nasal mucosal fluid, pharyngeal mucosal fluid, saliva, exhaled condensate, and blood are collected from each of the population 90. The teacher sample 911 is sent to the mass spectrometer 93 and the gene analyzer 94 after being processed by the pretreatment device 92. Specifically, the sample obtained by the known protein extraction operation is sent to the mass spectrometer 93, and the sample obtained by the known RNA or DNA extraction operation is sent to the gene analyzer 94.

The mass spectrometer 93 writes the name of the detected protein and the weight ratio of the detected protein in the storage unit 3 as teacher protein information 311. The gene analysis apparatus 94 writes in the storage unit 3 whether or not the teacher sample 911 contains the DNA or RNA of the pathogen P as the teacher genomic information 312.

Each person belonging to the group 90 receives a medical examination by an expert 96 at the same timing as the collection of the sample 91. The expert 96 confirms various states including symptoms when infected with the pathogen P known in the medical examination, and creates the teacher medical examination information 313 and the infection confirmation information 314 in the storage unit 3 using the medical examination terminal 95. However, the infection confirmation information 314 is also added or corrected after the fact. Subsequent additions and corrections of the infection confirmation information 314 will be described with specific examples.

For example, when a symptom observed when infected with the pathogen P (hereinafter referred to as "infectious symptom") is first confirmed for an individual belonging to the group 90 by a medical examination by a specialist 96, the following treatment is performed. .. That is, it is described in the infection confirmation information 314 up to the previous day that the individual was infected. Further, the number of days until the outbreak of infectivity is estimated from the measurement of infectivity or the known characteristics regarding the infection of the pathogen P, and the number of days until the outbreak of infectivity is added to the infection confirmation information 314 up to the previous day.

Teacher data 31 is created by the above processing. One teacher data 31 is created every time one person collects one teacher sample 911 in this way. Therefore, if 100 people collect teacher sample 911 every day for 30 days, a total of 3,000 teacher data 31 will be created. The individual teacher data 31 are created.

(Creation of evaluation data)
The method of creating the evaluation data 33 is substantially the same as the method of creating the teacher data 31. However, since the evaluation data 33 does not include the information corresponding to the infection confirmation information 314 of the teacher data 31, no ex post facto correction is necessary as in the teacher data 31.

According to the first embodiment described above, the following effects can be obtained.
(1) The method for creating the learning model 32 for diagnosing the degree of infection with the pathogen P, which is executed in the infection determination system S, is a method for creating a protein obtained from a teacher sample 911 collected from a group 90, which is a plurality of sample providers. A plurality of teacher data 31 including the teacher protein information 311 which is information and the infection confirmation information 314 which is the degree of infection to the pathogen of the sample provider are acquired, and the protein information is input using the teacher data 31 to enter the pathogen. A learning model 32 is generated, which outputs the degree of infection to. Therefore, it is possible to create a learning model 32 capable of more sensitive detection than in the case of detecting the genomic information of the pathogen itself of the pathogen P. Specifically, it is as follows.

When detecting the genomic information of the pathogen P as in Patent Document 1 described above, the detection is impossible unless the collected sample 91 contains the genomic information of the pathogen P. Therefore, if there is a problem in the collection method of the sample 91 and the pathogen P is collected from a place where the pathogen P does not exist, or if the pathogen P is not collected by chance, the genomic information of the pathogen P is not detected and there is no erroneous infection. It may be judged. However, in the present embodiment, since the protein produced by the pathogen infected with the pathogen P and the protein produced by the pathogen P in the host cell are learned by machine learning, the learning model 32 to be created is more sensitive than the conventional method. Infection to P can be detected.

(2) The learning model 32 further inputs the teacher examination information 313. Therefore, non-protein information such as body temperature, cough, headache, myalgia, and fatigue can be included in the learning and reasoning elements.

(3) The learning model 32 further inputs the teacher genomic information 312 regarding the pathogen P. Therefore, the learning model 32 created can more reliably detect infection with the pathogen P.

(4) The degree of infection is the presence or absence of infection with a pathogen.

(5) The degree of infection is the presence or absence of infection with a pathogen and the presence or absence of infectivity.

(6) The degree of infection is the presence or absence of infection with a pathogen, the presence or absence of infectivity, and the number of days remaining until the person has infectivity.

(7) The learning model program 32A acquires evaluation data 33, which is information on the protein of the sample 91 collected from the evaluation target person 90A, from the storage unit 3, and obtains it from the teacher sample 911 collected from the sample provider, that is, the group 90. The acquired evaluation data 33 is input to the learning model 32 trained using the teacher data 31 which inputs the information of the protein to be obtained and outputs the degree of infection to the pathogen P of the population 90, and the evaluation subject 90A. The arithmetic unit 10 is made to execute a process of outputting the degree of infection with the pathogen P.

(8) The arithmetic unit 10 is obtained from a reading unit that reads evaluation data 33, which is information on the protein of the evaluation sample 912 collected from the evaluation target person 90A, that is, a communication unit 2, and a teacher sample 911 collected from the group 90. A storage unit 3 for storing a learning model 32 trained using teacher data 31 in which teacher protein information 311 which is protein information is input and infection confirmation information 314 which is the degree of infection to a pathogen of a population 90 is output. , A calculation unit 1 for inputting evaluation data 33 into the learning model 32 and outputting the degree of infection of the evaluation target person 90A with the pathogen.

(Modification 1)
The evaluation target 90A does not have to be examined by the expert 96. In that case, the evaluation data 33 does not include the evaluation consultation information 333. Further, in this case, the group 90 does not have to be examined by the expert 96, and the teacher data 31 does not include the teacher examination information 313.

(Modification 2)
The infection determination system S does not have to include the gene analysis device 94. In this case, the teacher data 31 does not include the teacher genome information 312, and the evaluation data 33 does not include the evaluation genome information 332.

(Modification 3)
In the first embodiment described above, the infection confirmation information 314, that is, the degree of infection includes not only the presence or absence of infection but also the presence or absence of infectivity and the number of days until the occurrence of infectivity. However, the infection confirmation information 314, that is, the degree of infection may be the presence or absence of infection and the presence or absence of infectivity, or may be only the presence or absence of infection. The severity of the symptoms may be added to the degree of infection.

(Modification example 4)
In the first embodiment described above, the sample 91 contained nasal mucosal fluid, pharyngeal mucosal fluid, saliva, exhaled breath condensate, and blood. However, the sample 91 may contain at least one of nasal mucosal fluid, pharyngeal mucosal fluid, saliva, exhaled breath condensate, and blood.

(Modification 5)
In the first embodiment described above, learning and reasoning did not include information on the type of sample 91. However, the output of the mass spectrometer 93 and the gene analyzer 94 with the information of the type of the sample 91 may be used as the teacher data 31. Specifically, even in the case of the sample 91 collected at the same time by the same person, the information on the protein contained in the nasal mucosal fluid, the information on the protein contained in the pharyngeal mucosal fluid, the information on the protein contained in the saliva, and the exhaled breath. The information on the protein contained in the condensate and the information on the protein contained in the blood are treated as different information.

According to this modification 5, the following effects can be obtained.
(9) The teacher sample 911 contains two or more types of nasal mucosal fluid, pharyngeal mucosal fluid, blood, exhaled condensate, and saliva, and is treated as information on proteins different for each type of sample 91. Therefore, the learning model 32 can increase the types of information used in the calculation and improve the accuracy of inference.

(Modification 6)
Further, in the modification 5, the learning model 32 may be created for each type of the sample 91, and the inference result may be performed for each type of the sample 91. Specifically, four learning models 32 corresponding to each of nasal mucosal fluid, pharyngeal mucosal fluid, saliva, exhaled fluid, and blood are created, and learning and inference are performed on each of them. And the result of inference is also performed for each type of sample 91. According to this modification 6, a human can infer the state of infection by referring to the inference result for each type of the sample 91. Therefore, it becomes easier to give meaning to the diagnosis result as compared with the case where the inference is performed only by the learning model 32. For example, if it is inferred from the exhaled breath condensate of the evaluation subject 90A that there is an infection, it can be inferred that the infection has spread to the lungs and the condition is serious.

(Modification 7)
In the first embodiment described above, the population 90 has a teacher sample 911 collected at predetermined cycles over a predetermined period. However, the number of times the teacher sample 911 is collected from each individual belonging to the group 90 may be one. However, even in this case, the progress after collection is observed, and the infection confirmation information 314 is added or corrected as necessary.

(Modification 8)
In the first embodiment described above, the population 90 is a human population. However, the population 90 may be non-human organisms such as cows, pigs, chickens, fish, insects and the like.

(Modification 9)
A learning model 32 corresponding to each of the plurality of pathogens P may be created, and the degree of infection to the plurality of pathogens P may be estimated at the same time in the inference phase.

-Second embodiment-
A second embodiment of the method of generating a learning model will be described with reference to FIG. In the following description, the same components as those in the first embodiment are designated by the same reference numerals, and the differences will be mainly described. The points not particularly described are the same as those in the first embodiment. The present embodiment is different from the first embodiment in that the learning process and the inference process are mainly executed in different devices.

FIG. 3 is a configuration diagram of the infection determination system S1 in the second embodiment. The infection determination system S1 is different from the first embodiment in that the learning device 10A and the inference device 10B are provided instead of the arithmetic unit 10.

The learning device 10A includes a learning calculation unit 1A, a learning communication unit 2A, and a learning storage unit 3A. The teacher data 31 and the learning model 32 are stored in the learning storage unit 3A. The learning calculation unit 1A is, for example, a central processing unit (not shown), and performs learning processing by expanding a program stored in a read-only memory (not shown) into a volatile memory (not shown) and executing the program. However, the learning calculation unit 1A may be realized by using a hardware circuit or a reconfigurable logic circuit, or may be realized by a combination thereof.

The learning communication unit 2A is a communication module capable of communicating with the mass spectrometer 93, the gene analysis device 94, the examination terminal 95, and the inference device 10B. The learning communication unit 2A may be directly connected to the mass spectrometer 93, the gene analysis device 94, the examination terminal 95, and the inference device 10B, or may be connected via a communication network, for example, the Internet. The learning storage unit 3A is a non-volatile storage device such as a hard disk drive.

The inference device 10B includes an inference calculation unit 1B, an inference communication unit 2B, and an inference storage unit 3B. The learning model program 32A and the evaluation data 33 are stored in the inference storage unit 3B. The inference calculation unit 1B is, for example, a central processing unit (not shown), and performs inference processing by expanding and executing the learning model program 32A in a volatile memory (not shown). However, the inference calculation unit 1B may be realized by using a hardware circuit or a reconfigurable logic circuit, or may be realized by a combination thereof. That is, the learning model program 32A may be a computer program read by the processor or circuit information for reconstructing a logic circuit.

The inference communication unit 2B is a communication module capable of communicating with the mass spectrometer 93, the gene analysis device 94, the examination terminal 95, and the learning device 10A. The inference communication unit 2B may be directly connected to the mass spectrometer 93, the gene analysis device 94, the examination terminal 95, and the learning device 10A, or may be connected via a communication network, for example, the Internet. Since the inference communication unit 2B acquires the evaluation data 33 from the mass spectrometer 93, the gene analysis device 94, and the examination terminal 95, it can also be called a “reading unit” that performs a process of reading the evaluation data 33 from the outside. The inference storage unit 3B is a non-volatile storage device such as a hard disk drive.

The mass spectrometer 93 and the gene analysis device 94 output the processing result to the learning device 10A when the teacher sample 911 is the processing target, and output the processing result to the inference device 10B when the evaluation sample 912 is the processing target. do. The examination terminal 95 outputs the result to the learning device 10A when the group 90 is examined, and outputs the result to the inference device 10B when the evaluation target person 90A is examined. The learning device 10A creates a learning model program 32A and transmits it to the inference device 10B each time the learning model 32 is created and updated. The inference device 10B executes the learning model program 32A to infer the evaluation data 33.

According to the second embodiment described above, learning and inference can be performed by separate devices.

(Modification 1 of the second embodiment)
In the second embodiment, the learning model program 32A is transmitted from the learning device 10A to the inference device 10B via the communication line. That is, in the second embodiment, the learning model program 32A is transmitted using a communication medium, that is, a network such as wired, wireless, or optical, or a carrier wave or a digital signal propagating in the network. However, the learning device 10A and the inference device 10B are provided with input / output interfaces (not shown), and the learning model program 32A is transmitted via a recording medium in which the input / output interfaces can be used, such as a USB memory, a hard disk drive, or a CD-R. May be done.

(Modification 2 of the second embodiment)
In the second embodiment, the learning device 10A creates a learning model program 32A and transmits it to the inference device 10B. However, the learning device 10A may transmit the learning model 32 as it is to the inference device 10B, and create the learning model program 32A using the learning model 32 received by the inference device 10B.

-Third embodiment-
A third embodiment of the method of generating a learning model will be described with reference to FIG. In the following description, the same components as those in the first embodiment are designated by the same reference numerals, and the differences will be mainly described. The points not particularly described are the same as those in the first embodiment. In the present embodiment, the evaluation sample 912 is mainly different from the first embodiment.

FIG. 4 is a configuration diagram of the infection determination system S2 according to the third embodiment. In the present embodiment, the learning model 32 created by the process described in the first embodiment is applied to the environment E, and the presence of the pathogen in the environment E is evaluated. The environment E may be a closed space or an open space. For example, the environment E is a lobby of a building, an office, a hotel room, a living room at home, a vehicle, a train, a passenger plane, a gymnasium, a competition space of a stadium, a viewing space of a stadium, and a street.

In the present embodiment, the evaluation sample 912 can be extracted from the environment E by using various methods. For example, the liquid obtained by repeatedly passing the air in the environment E through a filter having a fine mesh and bringing a predetermined liquid into contact with the filter can be used as the evaluation sample 912. Further, the condensed liquid obtained by compressing the air in the environment E may be used as the evaluation sample 912, or the air in the environment E may be pumped into the liquid for a predetermined time to use the liquid as the evaluation sample 912. The processing method of the acquired evaluation sample 912 is the same as that of the first embodiment.

According to the present embodiment, for example, the evaluation sample 912 is extracted from the air in the office, the evaluation data 33 is created by processing with the pretreatment device 92 or the like, and the presence or absence of the pathogen P in the office is used by using the arithmetic unit 10. Can be estimated. Therefore, since the presence or absence of the pathogen P can be determined for any environment E, it can contribute to the prevention of infection of the infectious disease P for an unspecified number of people. When the pathogen P is detected in the air of the environment E, each person included in the environment E is set as the evaluation target person 90A, and the evaluation sample 912 and the evaluation sample 912 described in the first embodiment are obtained and evaluated. Evaluation may be performed.

(Modified example of the third embodiment)
The inference in the third embodiment may be executed using the learning model 32 generated as follows. That is, the teacher data 31 does not include the teacher examination information 313, but only the teacher protein information 311, the teacher genome information 312, and the infection confirmation information 314. Then, for example, a liquid obtained by treating the air in the room where the infected person infected with the pathogen P and having infectivity stays, that is, a sample 91 derived from the infected person infected with the pathogen is a teacher sample in which the pathogen P is present. It is set to 911. Further, for example, the liquid obtained by treating the air in the room where an uninfected person or a non-infected person who is not infected with the pathogen P stays is referred to as a teacher sample 911 in which the pathogen P does not exist.

The learning model 32 in this modification may estimate the presence or absence of the pathogen P in the evaluation sample 912, or may estimate the certainty of the presence of the pathogen P. In this modification, a plurality of teacher data 31 including protein information obtained from a sample derived from an infected person infected with a pathogen are acquired, protein information is input using the teacher data 31, and the presence or absence of pathogen P is output. The learning model 32 is generated.

Further, a learning model 32 may be created for a plurality of pathogens P, and the type of pathogen P contained in the sample may be inferred. That is, a plurality of teacher data 31 including the protein information obtained from the sample 91 collected from the sample and the type of the pathogen P in the sample are acquired, and the protein information is input using the teacher data 31 and included in the sample. A learning model 32 may be generated with the type of pathogen as an output. In this case, in the inference phase, the type of pathogen P contained in the sample 91 is output by inputting the protein information obtained by processing the sample 91 acquired from the environment E into the learning model 32.

-Fourth Embodiment-
A fourth embodiment of the method of generating a learning model will be described with reference to FIGS. 5 to 8. In the following description, the same components as those in the first embodiment are designated by the same reference numerals, and the differences will be mainly described. The points not particularly described are the same as those in the first embodiment. This embodiment differs from the first embodiment mainly in that ions described later are included in the teacher data.

(Test data)
We analyzed sulfur metabolites in exhaled breath using 12 healthy subjects who were not infected with the new corona and 12 subjects who were infected with the new corona. The presence or absence of new corona infection was based on PCR tests. Of the 12 people infected with the new corona, 2 were mildly ill, 10 were moderately ill, and 1 of them later became severely ill. The mild, moderate, and severe classifications were based on saturated oxygen levels, respiratory symptoms, and chest CT images. This classification is described in, for example, Severe Acute Respiratory Syndrome (COVID-19), Medical Guide, Edition 4.1.

For the collection of exhaled breath, an exhaled breath collector (manufactured by GL Sciences Co., Ltd.) and an exhaled breath recovery device (manufactured by GL Sciences Co., Ltd.) were used. The exhaled breath collection device is cooled to -20 degrees Celsius in advance, and the subject breathes for 5 to 10 minutes through the exhaled breath collection device (mouthpiece type, mask type) to rapidly cool the exhaled aerosol and exhale. About 1 ml of the condensate was recovered.

Β- (4-hydroxyphenyl) ethyliodoacetamide (HPE-IAM), which is an electrophile alkylating agent, was added to 25 μl of the exhaled breath condensate so as to be 5 mM, and an alkylation reaction was caused at 37 degrees Celsius for 30 minutes. Then, formic acid was added to 1% to stabilize the sulfur metabolite as an HPE-IAM adduct. The electrophile alkylating agent alkalizes the electron pair of sulfur and suppresses the decomposition of sulfur metabolites.

Furthermore, for each of the plurality of sulfur metabolites, a stable isotope-labeled HPE-IAM adduct standard was added so that the concentration in the solution was 10 nM, and this was used as a mass spectrometric sample. Known literature (Ida T, Sawa T, Ihara H, Tsuchiya Y, Watanabe Y, Kumagai Y, Suematsu M, Motohashi H, Fujii S, Matsunaga T, Yamamoto M, Ono K, Devarie-Baez NO, Xian M, Fukuto JM , Akaike T. Reactive cysteine persulfides and S-polythiolation regulate oxidative stress and redox signaling. Proc Natl Acad Sci USA 111: _7606-7611 ⁽ 2014).) The HPE ^- IAM adducts of HS ₂ O _3- ⁾ and hydrogen disulfide ion (HS _2- ) were quantitatively analyzed using a mass analyzer. That is, 35 μl of the above mass spectrometric sample / standard was injected into a high performance liquid chromatograph mass spectrometer: LCMS ^TM -8060 (Shimadzu), and quantitatively analyzed under the MRM (multiple reaction monitoring) conditions shown in Table 1.

FIG. 5 shows the data of the quantitative analysis results of these 3 ions for 12 healthy subjects and 12 new corona-infected subjects. The concentration of various sulfur metabolites contained in the exhaled breath condensate was generally lower than that in the living body. It is considered that this is due to the fact that sulfur metabolites in the living body are diluted in the exhaled breath and that the efficiency of exhaled breath recovery is not always good. In addition, the levels of sulfite ion (HSO _3- ⁾ , thiosulfate ion (HS ₂ O _3- ⁾ , ^and hydrogen sulfide ion (HS _2- ) are significantly higher in moderately ill patients than in healthy subjects. It was confirmed that.

X, one of the 12 new corona-infected individuals mentioned above, has transitioned from moderate to severe. FIG. 6 shows the measurement results of sulfate ion (HSO _3- ⁾ and thiosulfate ion (HS ₂ O _3- ⁾ when X becomes severe. It was found that the production level of thiosulfate ion (HS ₂ O _3- ⁾ was significantly higher than that of healthy subjects even in moderate and severe cases. The aforementioned X showed a marked increase in sulfite ion (HSO _3- ⁾ in specimens collected before it became severe, but not in moderately ill patients who did not become severe. .. Therefore, sulfite ion (HSO _3- ⁾ is a promising biomarker for assessing the transition risk of aggravation of pneumonia in coronavirus infections.

Since active sulfur has antioxidant activity, that is, it is oxidized by oxidative stress, active oxygen, etc., the aggravation of pneumonia, that is, the exacerbation of oxidative stress, thereby changing the profile of sulfur metabolites toward oxidation. It can be said that it shifts.

FIG. 7 is a configuration diagram of the arithmetic unit 10C according to the fourth embodiment. The teacher data 31 further includes teacher ion information 315. The evaluation data 33 further includes evaluation ion information 335.

The teacher ion information 315 is the identification information and detection of an ion containing at least one of sulfate ion (HSO _3- ⁾ , thiosulfate ion (HS ₂ O _3- ⁾ , and hydrogen sulfide ion (HS _2- ⁾ . Quantitative information on ions, produced by the mass spectrometer 93. The quantitative information of ions is, for example, the concentration of each ion. The evaluation ion information 335 is ion information obtained by processing the evaluation sample 912 collected from the evaluation target person 90A, and the type of information is the same as that of the teacher ion information 315.

FIG. 8 is a diagram showing an example of teacher data 31 in the present embodiment. The difference from the teacher data 31 shown in FIG. 2 in the first embodiment is that the teacher ion information 315 is added.

The teacher ion information 315 and the evaluation ion information 335 were obtained by the above-mentioned method using the breath recovery device, the breath recovery device, and the mass spectrometer described in the present embodiment. The teacher protein information 311 and the teacher genome information 312, the evaluation protein information 331, and the evaluation genome information 332 may be obtained through the same processing as the teacher ion information 315 and the evaluation ion information 335, or the teacher ion information 315 and the evaluation. It may be obtained by a treatment different from that of the ion information 335, that is, the treatment in the first embodiment.

In the learning phase, the learning model 32 in the present embodiment learns by inputting teacher ion information 315 and outputting infection confirmation information 314 in addition to teacher protein information 311, teacher genomic information 312, and teacher examination information 313. It is a learning model that is performed. Further, in the inference phase, the learning model 32 infers the degree of infection by inputting the evaluation ion information 335 in addition to the evaluation protein information 331, the evaluation genome information 332, and the evaluation consultation information 333.

The learning model program 32A created for the convenience of executing the inference process using the learning model 32 includes the evaluation protein information 331, the evaluation genome information 332, and the evaluation consultation information 333, similarly to the learning model 32 in the inference phase. In addition, the evaluation ion information 335 is used as an input to infer the degree of infection.

According to the fourth embodiment described above, the following effects can be obtained.
(10) The teacher data includes at least one information of sulfite ion, thiosulfate ion, and hydrogen sulfide ion obtained from the teacher sample 911 as ion information. In the generation of the learning model 32, protein information and ion information are input, and the degree of infection with a pathogen is output. Therefore, a learning model 32 can be generated using at least one of sulfate ion, thiosulfate ion, and hydrogen sulfide ion, which are promising biomarkers, and the degree of infection with the pathogen in the evaluation sample 912 can be accurately inferred.

In each of the above-described embodiments and modifications, the configuration of the functional block is only an example. Several functional configurations shown as separate functional blocks may be integrally configured, or the configuration represented by one functional block diagram may be divided into two or more functions. Further, a configuration in which a part of the functions of each functional block is provided in another functional block may be provided.

The above-mentioned embodiments and modifications may be combined. Although various embodiments and modifications have been described above, the present invention is not limited to these contents. Other aspects considered within the scope of the technical idea of the present invention are also included within the scope of the present invention.

1 ... Calculation unit 1A ... Learning calculation unit 1B ... Reasoning calculation unit 3 ... Storage unit 3A ... Learning storage unit 3B ... Reasoning storage unit 10 ... Calculation device 10A ... Learning device 10B ... Reasoning device 31 ... Teacher data 32 ... Learning model 32A ... Learning model program 33 ... Evaluation data 90A ... Evaluation target person 311 ... Teacher protein information 312 ... Teacher genomic information 313 ... Teacher consultation information 314 ... Infection confirmation information 331 ... Evaluation protein information 332 ... Evaluation genome information 333 ... Evaluation consultation information 91 ... Specimen 911 ... Teacher sample 912 ... Evaluation sample

Claims

It is a method of creating a learning model for diagnosing the degree of infection with a pathogen.
Obtain multiple teacher data including protein information obtained from the sample collected from the sample provider and the degree of infection of the sample provider with the pathogen.
A method for generating a learning model, which uses the teacher data to input information on the protein and output the degree of infection with the pathogen.
In the method for generating a learning model according to claim 1,
The learning model is a method of generating a learning model in which medical examination information is further input.
In the method for generating a learning model according to claim 1 or 2.
The learning model is a method of generating a learning model in which genomic information about the pathogen is further input.
In the method for generating a learning model according to any one of claims 1 to 3.
A method for generating a learning model, wherein the sample contains two or more of nasal mucosal fluid, pharyngeal mucosal fluid, blood, exhaled condensate, and saliva, and is treated as information on proteins that differ depending on the type of the sample.
In the method for generating a learning model according to any one of claims 1 to 4.
The degree of infection is a method for generating a learning model, which is the presence or absence of infection with the pathogen.
In the method for generating a learning model according to any one of claims 1 to 4.
The degree of infection is a method for generating a learning model, which is the presence or absence of infection with the pathogen and the presence or absence of infectivity.
In the method for generating a learning model according to any one of claims 1 to 4.
The degree of infection is a method for generating a learning model, which is the presence or absence of infection with the pathogen, the presence or absence of infectivity, and the number of days remaining until the patient has infectivity.
Obtain evaluation data, which is information on the protein of the evaluation sample collected from the evaluation subject,
The acquired evaluation data is applied to a learning model trained using teacher data in which information on proteins obtained from a teacher sample collected from a sample provider is input and the degree of infection with the pathogen of the sample provider is output. A program that causes a computer to perform a process of inputting and outputting the degree of infection of the evaluated person with the pathogen.
In the program according to claim 8,
The learning model is a program in which medical examination information is further input.
In the program according to claim 8 or 9.
The learning model is a program that further inputs genomic information about the pathogen.
In the program according to any one of claims 8 to 10.
The evaluation sample and the teacher sample include two or more of nasal mucosal fluid, pharyngeal mucosal fluid, blood, exhaled condensate, and saliva.
In the program according to any one of claims 8 to 11.
The degree of infection is the presence or absence of infection with the pathogen, a program.
In the program according to any one of claims 8 to 11.
The degree of infection is the presence or absence of infection with the pathogen and the presence or absence of infectivity, a program.
In the program according to any one of claims 8 to 11.
The degree of infection is the presence or absence of infection with the pathogen, the presence or absence of infectivity, and the number of days remaining until the patient has infectivity.
A reading unit that reads evaluation data including information on the protein of the evaluation sample collected from the evaluation target, and
A storage unit that stores a learning model trained using teacher data that inputs protein information obtained from a teacher sample collected from a sample provider and outputs the degree of infection of the sample provider with a pathogen.
An arithmetic unit including an arithmetic unit that inputs the evaluation data into the learning model and outputs the degree of infection of the evaluation target person with the pathogen.
A method of creating a learning model for diagnosing the presence of pathogens.
Obtain multiple teacher data including information on proteins obtained from specimens derived from infected persons infected with pathogens.
A method for generating a learning model, which uses the teacher data to input information on the protein and output the presence or absence of the pathogen as an output.
A method of creating a learning model for diagnosing the presence of pathogens.
Obtain multiple teacher data including protein information obtained from the sample collected from the sample and the type of pathogen in the sample.
A method for generating a learning model, which uses the teacher data to input information on the protein and output the type of pathogen contained in the sample.
Obtain evaluation data, which is information on the protein of the collected sample,
The acquired evaluation data is input to the learning model trained by inputting information on proteins obtained from a sample derived from an infected person infected with a pathogen and using teacher data that outputs the presence or absence of the pathogen. A program that causes a computer to execute a process that outputs the presence or absence of the pathogen in a sample.
In the method for generating a learning model according to any one of claims 1 to 4.
In the teacher data, at least one information of sulfate ion (HSO 3- ) , thiosulfate ion (HS 2 O 3- ) , and hydrogen sulfide ion (HS 2- ) obtained from the sample is used as ion information. Included,
A method for generating a learning model, in which information on the protein and information on the ions are input and the degree of infection with the pathogen is output in the generation of the learning model.
In the arithmetic unit according to claim 15,
In the evaluation data, at least one information of sulfate ion (HSO 3- ) , thiosulfate ion (HS 2 O 3- ) , and hydrogen sulfide ion (HS 2- ) obtained from the evaluation sample is ion information. Included as
The learning model is further input with at least one piece of information about sulfate ion (HSO 3- ) , thiosulfate ion (HS 2 O 3- ) , and hydrogen sulfide ion (HS 2- ) obtained from the teacher sample. The resulting arithmetic unit.