WO2024090062A1

WO2024090062A1 - Antigen discovery method and antigen discovery system

Info

Publication number: WO2024090062A1
Application number: PCT/JP2023/033322
Authority: WO
Inventors: 裕一引地; 悠矢宮本; 慎弥渡辺; 基樹高木; 順一今井
Original assignee: チューニングフォーク・バイオ・インク; 公立大学法人福島県立医科大学
Priority date: 2022-10-27
Filing date: 2023-09-13
Publication date: 2024-05-02

Abstract

This antigen discovery method comprises a data acquisition step for acquiring measurement data, a data calculation step, and an antigen selection step. In the data calculation step, antigen data is calculated on the basis of the measurement data, in which the antigen data is associated with an antigen-antibody reaction between a first number of target protein molecules immobilized on a protein microarray and autoantibody molecules in a plurality of samples to be tested. In the antigen selection step, the antibody data are input to a machine learning model as input data, and a specific target protein having a correlation with a disease-associated antigen, which is selected from the first number of target protein molecules, is output for each of a plurality of diseases.

Description

Antigen discovery method and antigen discovery system

The present invention relates to an antigen discovery method and an antigen discovery system that uses a protein microarray to discover target proteins that are correlated with disease-related antigens associated with each of a number of diseases.

It is known that specimen samples derived from subjects with a disease can be analyzed using protein microarrays (see, for example, Patent Document 1). A protein microarray is a substrate such as a glass slide on which thousands to tens of thousands of target proteins are immobilized in an array of spots. By contacting the specimen sample with each target protein on the substrate, it is possible to obtain, as measurement data, feature quantities that serve as an index of binding properties associated with the antigen-antibody reaction between the autoantibodies in the specimen sample and each target protein.

Autoantibodies are produced against disease-related antigens that are related to a disease. Therefore, by analyzing measurement data regarding the antigen-antibody reaction between each target protein on the protein microarray substrate and the autoantibody, it is possible to search for specific target proteins that have a correlation with disease-related antigens from among the target proteins.

However, since there are multiple diseases, there is room for improvement in efficiently searching for specific target proteins that correlate with each disease-associated antigen associated with each disease.

Specific Publication No. 2021-521536

The object of the present invention is to provide an antigen discovery method and an antigen discovery system that can efficiently discover target proteins that are correlated with disease-related antigens that are related to each of a number of diseases, using a protein microarray in which a number of target proteins are fixed on a substrate.

　An antigen discovery method according to one aspect of the present invention is a method for discovering target proteins that are correlated with disease-related antigens associated with each of a plurality of diseases, using a protein microarray in which a first number of target proteins are fixed on a substrate. This antigen discovery method includes a data acquisition step of acquiring measurement data for each of the first number of target proteins, which is related to a predetermined feature amount when each of a plurality of specimen samples derived from a plurality of subjects is contacted with the first number of target proteins; a data calculation step of calculating, based on the measurement data, antibody data for each of the first number of target proteins, which is related to binding associated with an antigen-antibody reaction between an autoantibody in the plurality of specimen samples and the first number of target proteins, in association with diagnostic data showing diagnostic results for a plurality of diseases for each of the plurality of subjects; and an antigen selection step of generating a machine learning model using the antibody data for each of the first number of target proteins associated with the diagnostic data as input data, selecting a specific target protein that is correlated with the disease-related antigen from among the first number of target proteins based on the antibody data, corresponding to each of a plurality of diseases shown in the diagnostic data, according to the machine learning model, and outputting the selected specific target protein as output data.

An antigen search system according to another aspect of the present invention comprises a measurement device that uses a protein microarray having a first number of target proteins fixed on a substrate to measure a predetermined characteristic amount when each of a plurality of specimen samples derived from a plurality of specimens is contacted with the first number of target proteins, and outputs measurement data for each of the first number of target proteins indicating the measurement results, and an antigen search device that searches for target proteins that have a correlation with disease-associated antigens associated with each of a plurality of diseases. The antigen search device includes a data acquisition unit that acquires the measurement data, a data calculation unit that calculates antibody data for each of the first number of target proteins related to binding associated with an antigen-antibody reaction between the autoantibody in the multiple test specimens and the first number of target proteins based on the measurement data, in association with diagnostic data showing diagnostic results for multiple diseases for each of the multiple test specimens, and an antigen selection unit that generates a machine learning model using the antibody data for each of the first number of target proteins associated with the diagnostic data as input data, selects a specific target protein that has a correlation with the disease-related antigen from among the first number of target proteins based on the antibody data in accordance with the machine learning model, corresponding to each of the multiple diseases shown in the diagnostic data, and outputs the selected specific target protein as output data.

The objects, features and advantages of the present invention will become more apparent from the following detailed description and the accompanying drawings.

FIG. 1 is a block diagram of an antigen discovery system according to an embodiment of the present invention. FIG. 2 is a diagram explaining a measurement process performed by a measurement device provided in the antigen search system. FIG. 11 is a diagram showing an example of measurement data showing a measurement result by a measurement device. 1 is a flowchart of an antigen searching method performed by an antigen searching device provided in an antigen searching system. 11 is a diagram showing an example of antibody data calculated by a data calculation unit of the antigen searching device. FIG.

Below, an antigen search system according to an embodiment of the present invention and an antigen search method executed by an antigen search device of the antigen search system will be described with reference to the drawings.

[Overall configuration of antigen discovery system]
As shown in Fig. 1, the antigen search system 1 is a system that includes a measurement device 2 and an antigen search device 3. In the antigen search system 1, the measurement device 2 acquires measurement data D1 related to diseases of multiple subjects using a protein microarray 4 shown in Fig. 2, and the antigen search device 3 searches for proteins that have a correlation with disease-related antigens related to each of the multiple diseases.

The subject primarily refers to humans and other mammals. Other mammals include, for example, primates such as monkeys and chimpanzees, livestock animals such as cows, horses, pigs and sheep, pet animals such as dogs and cats, and laboratory animals such as mice, rats and rabbits. Of these, it is preferable that the subject is a human.

Diseases include neuromuscular diseases such as attention deficit hyperactivity disorder, dementia with Lewy bodies, mild cognitive impairment, major depressive disorder, and Parkinson's disease; malignant tumors such as brain tumors, gastric cancer, colon cancer, liver cancer, and lung cancer; digestive system diseases such as ulcerative colitis; respiratory system diseases such as idiopathic interstitial pneumonia; circulatory system diseases such as idiopathic dilated cardiomyopathy; blood system diseases such as idiopathic thrombocytopenic purpura; and sequelae of infectious diseases such as COVID-19. Of these, it is preferable to target multiple diseases as neuromuscular diseases.

[Measuring equipment]
The measurement device 2 is a device that acquires measurement data D1 regarding a reference sample SPR and a plurality of specimen samples SPS derived from a plurality of specimens, using a protein microarray 4 shown in FIG.

The protein microarray 4 is a matrix of a plurality of spots 42 arranged on a substrate 41 such as a slide glass. In the protein microarray 4, a first number of a plurality of target proteins TP, ranging from several thousand to several tens of thousands, are fixed on the spots 42 on the substrate 41. That is, the protein microarray 4 is a matrix of a first number of target proteins TP, ranging from several thousand to several tens of thousands, fixed on the spots 42 on the substrate 41. The target proteins TP are antigen proteins such as recombinant proteins expressed in a cell-free expression system such as wheat germ extract. Examples of target proteins TP include standard proteins such as A1BG (1alpha-1-B glycoprotein), A1CF (APOBEC1 complementation factor), and A2M (2alpha-2-macroglobulin), mutant proteins derived from the p53 gene, and fusion proteins derived from the ALK fusion gene.

The specimen sample SPS derived from a subject may include, for example, whole blood, serum, plasma, urine, prostatic fluid, tears, mucus ascites, oral fluid, saliva, semen, seminal plasma, mucus, stool, sputum, cerebrospinal fluid, bone marrow, lymph, etc., and these may be prepared with a diluent or the like. The specimen sample SPS includes samples derived from subjects without a disease and samples derived from subjects with a disease. The specimen sample SPS derived from a subject with a disease contains the autoantibody ABA produced against the disease-associated antigen AG (see Figure 4 below) associated with the disease. On the other hand, the specimen sample SPS derived from a subject without a disease does not contain the autoantibody ABA.

Disease-associated antigen AG is a protein that produces autoantibody ABA in the body of a subject due to a disease. Autoantibody ABA refers to an antibody against disease-associated antigen AG. Examples of autoantibody ABA include antibodies with globulin types (classes) of IgG, IgM, IgA, IgD, and IgE. Autoantibody ABA binds to target protein TP that is correlated with disease-associated antigen AG associated with a disease among the first number of target proteins TP fixed to spots 42 on substrate 41 of protein microarray 4 by antigen-antibody reaction with the target protein TP. On the other hand, autoantibody ABA does not bind to target protein TP that is not correlated with disease-associated antigen AG associated with a disease among the first number of target proteins TP, or even if it does bind, the binding strength is weak. That is, the autoantibody ABA exhibits high binding affinity to the target protein TP that correlates with the disease-associated antigen AG associated with the disease corresponding to the autoantibody ABA, but exhibits low binding affinity to the target protein TP that does not correlate with the disease-associated antigen AG.

The reference sample SPR is a sample containing a reference antibody ABR that can bind, through an antigen-antibody reaction, to any of the first number of target proteins TP fixed to the spots 42 on the substrate 41 of the protein microarray 4. An example of the reference antibody ABR is a Goat reference antibody.

The measuring device 2 measures a predetermined feature amount when each of a plurality of specimen samples SPS derived from a plurality of specimens is brought into contact with a first number of target proteins TP fixed to spots 42 on a substrate 41 of a protein microarray 4. The measuring device 2 is not particularly limited in its measurement method as long as it is capable of measuring a feature amount that is an index of binding associated with an antigen-antibody reaction between the first number of target proteins TP and the autoantibody ABA in the specimen sample SPS. The measurement method of the measuring device 2 is distinguished according to the labeling substance that is bound to the autoantibody ABA to measure the feature amount. Examples of the measurement method of the measuring device 2 include a method using a fluorescent substance as a labeling substance, a method using an enzyme as a labeling substance, and a method using RI (radioisotope) as a labeling substance.

In this embodiment, the measuring device 2 uses a fluorescent substance as a labeling substance and measures the fluorescence intensity as a feature quantity. Specifically, the measuring device 2 measures the fluorescence intensity of the fluorescence emitted from the first fluorescent substance in response to irradiation of a predetermined excitation light in the first secondary antibody AB1 that can bind to the reference antibody ABR in the reference sample SPR and is labeled with a first fluorescent substance. The measuring device 2 also measures the fluorescence intensity of the fluorescence emitted from the second fluorescent substance in response to irradiation of a predetermined excitation light in the second secondary antibody AB2 that can bind to the autoantibody ABA in the specimen sample SPS and is labeled with a second fluorescent substance. The first secondary antibody AB1 is, for example, an anti-Goat IgG antibody labeled with a first fluorescent substance that emits green fluorescence. The second secondary antibody AB2 is, for example, an anti-human IgG antibody, anti-human IgM antibody, anti-human IgA antibody, anti-human IgD antibody, anti-human IgE antibody, etc., labeled with a second fluorescent substance that emits red fluorescence.

As shown in FIG. 2, assume that only the reference sample SPR is contacted with the first number of target proteins TP fixed to spots 42 on the substrate 41 of the protein microarray 4. In this case, the reference antibody ABR in the reference sample SPR binds to all of the first number of target proteins TP through an antigen-antibody reaction (primary reaction). When the first secondary antibody AB1 and the second secondary antibody AB2 are contacted with all of the first number of target proteins TP in a state in which the reference antibody ABR is bound, the first secondary antibody AB1 binds to the reference antibody ABR in the secondary reaction, and the second secondary antibody AB2 floats away and is removed as it is since there is no binding target. In this case, the first secondary antibody AB1 labeled with the first fluorescent substance binds to each of the first number of target proteins TP via the reference antibody ABR, while the second secondary antibody AB2 is not bound.

When only the reference sample SPR is brought into contact with the first number of target proteins TP, the measuring device 2 outputs measurement data D1 of a negative control (NC) for each of the first number of target proteins TP as data showing the measurement result. The measurement data D1 of the NC includes a first fluorescence intensity value D11 indicating the intensity of the fluorescence emitted from the first fluorescent substance of the first secondary antibody AB1 capable of binding to the reference antibody ABR, and a second fluorescence intensity value D12 indicating the intensity of the fluorescence emitted from the second fluorescent substance of the second secondary antibody AB2 capable of binding to the autoantibody ABA in the specimen sample SPS. When only the reference sample SPR is brought into contact with the first number of target proteins TP, as described above, the first secondary antibody AB1 labeled with the first fluorescent substance is bound to each of the first number of target proteins TP via the reference antibody ABR, while the second secondary antibody AB2 is not bound. Therefore, in the measurement data D1 of the NC, ideally, for each of the first number of target proteins TP, the first fluorescence intensity value D11 indicates a value equal to or greater than a predetermined first reference intensity value, and the second fluorescence intensity value D12 indicates a value close to zero that is less than the predetermined second reference intensity value.

　Let us assume that the reference sample SPR and the specimen sample SPS are brought into contact with the first number of target proteins TP fixed to the spots 42 on the substrate 41 of the protein microarray 4. In this case, the reference antibody ABR in the reference sample SPR binds to all of the first number of target proteins TP by an antigen-antibody reaction (primary reaction). On the other hand, the autoantibody ABA in the specimen sample SPS binds to the target proteins TP that are correlated with the disease-associated antigen AG associated with the disease corresponding to the autoantibody ABA by an antigen-antibody reaction (primary reaction) among the first number of target proteins TP, but does not bind to the target proteins TP that are not correlated with the disease-associated antigen AG. When the first secondary antibody AB1 and the second secondary antibody AB2 are brought into contact with the first number of target proteins TP in a state in which the reference antibody ABR binds to all of the first number of target proteins TP and the autoantibody ABA binds to the target proteins TP that are correlated with the disease-associated antigen AG, the first secondary antibody AB1 binds to the reference antibody ABR and the second secondary antibody AB2 binds to the autoantibody ABA in the secondary reaction. In this case, a first secondary antibody AB1 labeled with a first fluorescent substance is bound to each of the first number of target proteins TP via a reference antibody ABR, and a second secondary antibody AB2 labeled with a second fluorescent substance is bound to a target protein TP that has a correlation with a disease-associated antigen AG via an autoantibody ABA.

When the reference sample SPR and the specimen sample SPS are brought into contact with the first number of target proteins TP, the measuring device 2 outputs, as data indicating the measurement results, measurement data D1 regarding the specimen sample SPS for each of the first number of target proteins TP in association with diagnostic data D2 (FIG. 1) indicating the diagnostic results regarding the disease of the specimen. The measurement data D1 regarding the specimen sample SPS includes a first fluorescence intensity value D11 indicating the intensity of fluorescence emitted from a first fluorescent substance of the first secondary antibody AB1 capable of binding to the reference antibody ABR, and a second fluorescence intensity value D12 indicating the intensity of fluorescence emitted from a second fluorescent substance of the second secondary antibody AB2 capable of binding to the autoantibody ABA in the specimen sample SPS. When the reference sample SPR and the specimen sample SPS are brought into contact with the first number of target proteins TP, as described above, the first secondary antibody AB1 labeled with a first fluorescent substance is bound to each of the first number of target proteins TP via the reference antibody ABR, and the second secondary antibody AB2 labeled with a second fluorescent substance is bound to the target protein TP correlated with the disease-related antigen AG via the autoantibody ABA. Therefore, in the measurement data D1 regarding the specimen sample SPS, ideally, for the first number of target proteins TP, the first fluorescence intensity value D11 indicates a value equal to or greater than the first reference intensity value for the target protein TP correlated with the disease-related antigen AG, and the second fluorescence intensity value D12 indicates a value equal to or greater than the second reference intensity value, and for the target protein TP not correlated with the disease-related antigen AG, the first fluorescence intensity value D11 indicates a value equal to or greater than the first reference intensity value, and the second fluorescence intensity value D12 indicates a value close to zero that is less than the second reference intensity value.

The specimen sample SPS derived from a subject without a disease does not contain the autoantibody ABA produced against the disease-related antigen AG. Therefore, in the measurement data D1 relating to the specimen sample SPS derived from a subject without a disease, ideally, for each of the first number of target proteins TP, the first fluorescence intensity value D11 indicates a value equal to or greater than the first reference intensity value, and the second fluorescence intensity value D12 indicates a value close to zero that is less than the second reference intensity value.

3 shows an example of the measurement data D1 output from the measurement device 2. In the example of FIG. 3, the measurement data D1 includes, for each of the first number of target proteins TP, a first fluorescence intensity value D11 and a second fluorescence intensity value D12 for NC, and a first fluorescence intensity value D11 and a second fluorescence intensity value D12 associated with diagnostic data D2 for a plurality of specimen samples SPS derived from a plurality of specimens. In this case, the measurement data D1 for the plurality of specimen samples SPS includes, as a data group of the first fluorescence intensity value D11 and the second fluorescence intensity value D12, a data group associated with the first diagnostic data D21 indicating a diagnosis result of no disease, and a data group associated with the second to sixth diagnostic data D22 to D26 indicating a diagnosis result of, for example, a disease related to a neuromuscular disease. For example, the second diagnostic data D22 is diagnostic data indicating a diagnosis result of having attention deficit hyperactivity disorder. The third diagnostic data D23 is diagnostic data indicating a diagnosis result of having Lewy body dementia. The fourth diagnostic data D24 is diagnostic data indicating a diagnosis result of mild cognitive impairment. The fifth diagnostic data D25 is diagnostic data indicating a diagnosis result of major depressive disorder. The sixth diagnostic data D26 is diagnostic data indicating a diagnosis result of Parkinson's disease.

[About the antigen search device]
The measurement data D1 output from the measurement device 2 is input to the antigen search device 3. The antigen search device 3 will be described with reference to FIG. 1 as well as FIG. 4 and FIG. 5. The antigen search device 3 is a computer equipped with a CPU (Central Processing Unit), a storage area such as a HDD (Hard Disk Drive) or a flash memory, and a RAM (Random Access Memory) used as a working area for the CPU. Based on the measurement data D1, the antigen search device 3 searches for a target protein TP that has a correlation with each disease-related antigen AG associated with each of a plurality of diseases (e.g., diseases related to neuromuscular diseases) from among a first number of target proteins TP fixed on the substrate 41 of the protein microarray 4. The antigen search device 3 executes an antigen search method for searching for a target protein TP that has a correlation with each disease-related antigen AG associated with each of a plurality of diseases by machine learning using a neural network. A neural network is designed to mimic the structure of the human brain, and is made up of multiple layers of logic circuits that mimic the functions of neurons (nerve cells) in the human brain.

The antigen search device 3 has, as its functional configuration, a data acquisition unit 31 that performs the data acquisition step S1 in the antigen search method, a data calculation unit 32 that performs the data calculation step S2 in the antigen search method, and an antigen selection unit 33 that performs the antigen selection step S3 in the antigen search method.

The data acquisition unit 31 performs the data acquisition step S1 by acquiring the measurement data D1 output from the measurement device 2. In this embodiment, as described above, the measurement data D1 includes, for each of the first number of target proteins TP fixed on the substrate 41 of the protein microarray 4, a first fluorescence intensity value D11 and a second fluorescence intensity value D12 for NC, and a first fluorescence intensity value D11 and a second fluorescence intensity value D12 associated with diagnostic data D2 for multiple specimen samples SPS derived from multiple specimens. The data acquisition unit 31 acquires the measurement data D1 (step S11) and outputs the acquired measurement data D1 to the data calculation unit 32 (step S12).

The data calculation unit 32 performs the data calculation step S2 by calculating antibody data D3 for each of the first number of target proteins TP, which relates to the binding associated with the antigen-antibody reaction between the autoantibody ABA in the multiple specimen samples SPS and the first number of target proteins TP, in association with the diagnostic data D2 for each of the multiple specimens, based on the measurement data D1. The data calculation unit 32 calculates the antibody data D3 in association with the diagnostic data D2 based on the measurement data D1 (step S21), and outputs the calculated antibody data D3 to the antigen selection unit 33 (step S22).

The antibody data D3 for each of the first number of target proteins TP shows different values according to the difference in binding affinity of the autoantibody ABA to each target protein TP, and shows a larger value as the binding affinity of the autoantibody ABA increases. As shown in FIG. 5, the antibody data D3 for each of the first number of target proteins TP includes a data group of first antibody data D31 associated with first diagnostic data D21 indicating a diagnosis result of no disease, and a data group of second antibody data D32 associated with second to sixth diagnostic data D22 to D26 indicating a diagnosis result of disease.

The first antibody data D31 associated with the first diagnostic data D21 is calculated based on the first fluorescence intensity value D11 and the second fluorescence intensity value D12 associated with the first diagnostic data D21 in the measurement data D1. The first antibody data D31 indicates a value smaller than the second antibody data D32 associated with the second to sixth diagnostic data D22 to D26 for each of the first number of target proteins TP.

The second antibody data D32 associated with the second to sixth diagnostic data D22 to D26 is calculated based on the first fluorescence intensity value D11 and the second fluorescence intensity value D12 associated with the second to sixth diagnostic data D22 to D26 in the measurement data D1. In the second antibody data D32 associated with the second to sixth diagnostic data D22 to D26, in the first number of target proteins TP, the value for the target protein TP that is correlated with each disease-related antigen AG associated with each disease indicated in the second to sixth diagnostic data D22 to D26 is greater than the values for the other target proteins TP.

The number of data groups in the first antibody data D31 associated with the first diagnostic data D21 matches the number of specimen samples SPS derived from the subjects of the diagnostic results shown in the first diagnostic data D21. The number of data groups in the second antibody data D32 associated with the second to sixth diagnostic data D22 to D26 matches the number of specimen samples SPS derived from each subject of the diagnostic results shown in the second to sixth diagnostic data D22 to D26. For example, assume that the number of specimen samples SPS derived from each subject of the diagnostic results shown in the first to sixth diagnostic data D21 to D26 is 10. In this case, the number of data groups of the first antibody data D31 associated with the first diagnostic data D21, the number of data groups of the second antibody data D32 associated with the second diagnostic data D22, the number of data groups of the second antibody data D32 associated with the third diagnostic data D23, the number of data groups of the second antibody data D32 associated with the fourth diagnostic data D24, the number of data groups of the second antibody data D32 associated with the fifth diagnostic data D25, and the number of data groups of the second antibody data D32 associated with the sixth diagnostic data D26 are each "10".

The antigen selection unit 33 performs the antigen selection step S3 by selecting a specific target protein TPS that has a correlation with the disease-related antigen AG from among the first number of target proteins TP based on the antibody data D3 corresponding to each of the multiple diseases indicated in the diagnostic data D2. Specifically, the antigen selection unit 33 uses the antibody data D3 for each of the first number of target proteins TP associated with the diagnostic data D2 as input data, selects a specific target protein TPS that has a correlation with the disease-related antigen AG from among the first number of target proteins TP based on the antibody data D3 corresponding to each of the multiple diseases indicated in the diagnostic data D2, and generates a machine learning model LM that outputs the selected specific target protein TPS as output data (step S31). When the input data is input, the antigen selection unit 33 analyzes the relationship between the diagnostic data D2 and the first number of target proteins TP based on the antibody data D3 according to the machine learning model LM, and selects a specific target protein TPS from among the first number of target proteins TP corresponding to each of the multiple diseases indicated in the diagnostic data D2 (step S33). The antigen selection unit 33 then outputs the specific target protein TPS selected for each of the multiple diseases indicated in the diagnostic data D2 as a target protein TP that has a correlation with each disease-related antigen AG related to each disease (step S34).

In the antigen search device 3, antibody data D3 relating to the binding associated with the antigen-antibody reaction between a first number of target proteins TP fixed on the substrate 41 of the protein microarray 4 and autoantibodies ABA in multiple specimen samples SPS is input as input data to the machine learning model LM, and a specific target protein TPS that is correlated with a disease-associated antigen AG selected from the first number of target proteins TP based on the antibody data D3 corresponding to each of the multiple diseases indicated in the diagnostic data D2 can be output as output data. This makes it possible to efficiently search for a specific target protein TPS that is correlated with each disease-associated antigen AG associated with each of the multiple diseases, compared to searching for target proteins TP individually corresponding to each of the multiple diseases.

As described above, the antibody data D3 for each of the first number of target proteins TP calculated by the data calculation unit 32 includes a data group of the first antibody data D31 associated with the first diagnostic data D21 indicating a diagnosis result of not having a disease, and a data group of the second antibody data D32 associated with the second to sixth diagnostic data D22 to D26 indicating a diagnosis result of having a disease. In this case, as shown in FIG. 4, the antigen selection unit 33 may perform a process of step S32 of extracting a second number of multiple target proteins TP, which is smaller than the first number, from the first number of target proteins TP before the process of step S33 of selecting a specific target protein TPS. The antigen selection unit 33 extracts a second number of target proteins TP from the first number of target proteins TP that satisfy the condition that the first antibody data D31 is equal to or smaller than a first threshold value and the second antibody data D32 is equal to or larger than a second threshold value that is larger than the first threshold value. In addition, when there are multiple data groups of the first antibody data D31 associated with the first diagnostic data D21, and there are multiple data groups of the second antibody data D32 associated with the second to sixth diagnostic data D22 to D26, respectively, the antigen selection unit 33 may extract a second number of target proteins TP from the first number of target proteins TP that satisfy the condition that the number of the first antibody data D31 equal to or greater than the first threshold is equal to or less than the first determination number, and the number of the second antibody data D32 equal to or greater than the second threshold is equal to or greater than the second determination number.

When the second number of target proteins TP is extracted from the first number of target proteins TP in step S32, the antigen selection unit 33 analyzes the relationship between the diagnostic data D2 and the second number of target proteins TP based on the antibody data D3 according to the machine learning model LM, and selects a specific target protein TPS from the second number of target proteins TP corresponding to each of the multiple diseases indicated in the diagnostic data D2 (step S33). In this case, when selecting a specific target protein TPS having a correlation with the disease-related antigen AG according to the machine learning model LM for each of the multiple diseases indicated in the diagnostic data D2, the specific target protein TPS is selected from the second number of target proteins TP that is less than the first number of target proteins TP fixed on the substrate 41 of the protein microarray 4. This makes it possible to shorten the time required to select a specific target protein TPS using the machine learning model LM, and to more accurately search for a specific target protein TPS corresponding to each of the multiple diseases.

Furthermore, when selecting a specific target protein TPS from the second number of target proteins TP in step S33, the antigen selection unit 33 may obtain the similarity between the antibody data D3 for each of the second number of target proteins TP, and classify the second number of target proteins TP into a plurality of groups corresponding to each of the plurality of diseases indicated in the diagnostic data D2 by hierarchical clustering HC based on the similarity. Then, the antigen selection unit 33 selects the target proteins TP belonging to each of the plurality of groups as the specific target protein TPS. In this case, the antigen selection unit 33 classifies the second number of target proteins TP into a plurality of groups by hierarchical clustering HC based on the similarity between the antibody data D3 for each of the second number of target proteins TP. Then, the antigen selection unit 33 can select the target proteins TP belonging to each of the plurality of groups as the specific target protein TPS having a correlation with each disease-related antigen AG related to each of the plurality of diseases.

The antigen selection unit 33 starts by assigning the antibody data D3 for each of the second number of target proteins TP to one cluster in the hierarchical clustering HC, and then recursively combines similar clusters of antibody data D3. Depending on the criteria for selecting the clusters to be combined, the hierarchical clustering HC includes methods such as the shortest distance method, the longest distance method, and the group average method. The similarity between clusters is defined as the similarity between the antibody data D3 across the clusters.

The antigen selection unit 33 combines the clusters into one cluster in order starting from the pair with the greatest similarity between them, and stops the combination when the similarity between all clusters falls below a predetermined similarity reference value. In the hierarchical clustering HC by the antigen selection unit 33, when the combination of clusters is stopped, the second number of target proteins TP are classified into a plurality of groups corresponding to each of a plurality of diseases. At this time, the antigen selection unit 33 sets a maximum limit number for the number of target proteins TP belonging to each group, and by considering the similarity to be zero when the total number of antibody data D3 contained in the two combined clusters is greater than the maximum limit number, it is possible to prevent the number of target proteins TP belonging to each group from exceeding the maximum limit number.

The antigen selection unit 33 may also use a genetic algorithm GA when generating the machine learning model LM in step S31 (step S35). In this case, the antigen selection unit 33 generates the machine learning model LM using the genetic algorithm GA with the goal that the diseases indicated in the diagnostic data D2 corresponding to the antibody data D3 of each target protein TP belonging to each of the multiple groups are the same. The antigen selection unit 33 uses the antibody data D3 for each of the second number of target proteins TP associated with the diagnostic data D2 as input data, and generates the machine learning model LM using the genetic algorithm GA with a specific target protein TPS belonging to each group classified by the hierarchical clustering HC as output data. By using the genetic algorithm GA, the antigen selection unit 33 can generate the machine learning model LM with the goal that the diseases indicated in the diagnostic data D2 corresponding to the antibody data D3 of each target protein TP belonging to each group classified by the hierarchical clustering HC are the same.

In the genetic algorithm GA, the antigen selection unit 33 evaluates, for example, the identity of each disease corresponding to each target protein TP belonging to each group classified by the hierarchical clustering HC, and gives a first reward if the diseases are the same, and gives a second reward lower than the first reward if the diseases are not the same. By repeatedly performing such a genetic algorithm GA, the antigen selection unit 33 can generate a machine learning model LM in which the diseases corresponding to each target protein TP belonging to each group classified by the hierarchical clustering HC are the same. As a result, when the antibody data D3 for each of the second number of target proteins TP is input as input data to the machine learning model LM generated using the genetic algorithm GA, the second number of target proteins TP can be classified by the hierarchical clustering HC into multiple groups corresponding to each of the multiple diseases indicated in the diagnostic data D2.

The measurement data D1 obtained when using a protein microarray 4 may have low reproducibility. For this reason, when using multiple protein microarrays 4 corresponding to multiple diseases in the antigen searching device 3 to search for a specific target protein TPS that correlates with a disease-associated antigen AG associated with each disease, it is necessary to take into account the measurement error between the measurement data D1 corresponding to each of the multiple diseases.

In the antigen search device 3 according to this embodiment, the data calculation unit 32 performs a predetermined standardization process on the antibody data D3 input as input data to the machine learning model LM to absorb the measurement error between the measurement data D1 corresponding to each of the multiple diseases indicated in the diagnostic data D2, thereby calculating the standardized antibody data D3 for each of the first number of target proteins TP. Specifically, the data calculation unit 32 performs a standardization process on the antibody data D3 for each of the multiple diseases indicated in the diagnostic data D2 using a value at least at one percentile, thereby calculating the standardized antibody data D3 for each of the first number of target proteins TP. The data calculation unit 32 can calculate the standardized antibody data D3 that appropriately absorbs the measurement error between the measurement data D1 corresponding to each of the multiple diseases by this percentile-based standardization process. In this case, the antigen selection unit 33 can input the standardized antibody data D3 as input data to the machine learning model LM, thereby outputting the accurate specific target protein TPS as output data corresponding to each of the multiple diseases indicated in the diagnostic data D2.

In addition, in the protein microarray 4, there may be variation in the amount of the first number of target proteins TP fixed on the substrate 41. For this reason, when searching for specific target proteins TP that are correlated with disease-associated antigens AG related to each of a plurality of diseases using the protein microarray 4 in the antigen searching device 3 corresponding to each of a plurality of diseases, it is necessary to take into account measurement errors caused by differences in the amount of target proteins TP fixed on the substrate 41 for the measurement data D1 corresponding to each of the plurality of diseases.

Therefore, in the antigen searching device 3 according to this embodiment, the data acquiring unit 31 acquires, as the measurement data D1 for each of the first number of target proteins TP fixed on the substrate 41 of the protein microarray 4, a first fluorescence intensity value D11 relating to the intensity of fluorescence emitted from the first fluorescent substance of the first secondary antibody AB1 bound to a predetermined reference antibody ABR bound to all of the first number of target proteins TP, and a second fluorescence intensity value D12 relating to the intensity of fluorescence emitted from the second fluorescent substance of the second secondary antibody AB2 bound to the autoantibody ABA bound to some of the target proteins TP among the first number of target proteins TP. Then, the data calculating unit 32 performs a standardization process using the first fluorescence intensity value D11 and the second fluorescence intensity value D12 for the antibody data D3 input as input data to the machine learning model LM, thereby calculating the standardized antibody data D3 for each of the first number of target proteins TP. This allows the data calculation unit 32 to calculate standardized antibody data D3 for the measurement data D1 corresponding to each of the multiple diseases, which appropriately absorbs measurement errors caused by differences in the amount of the target protein TP fixed on the substrate 41. In this case, the antigen selection unit 33 can input the standardized antibody data D3 as input data to the machine learning model LM, thereby outputting an accurate specific target protein TPS as output data corresponding to each of the multiple diseases indicated in the diagnosis data D2.

Next, an example of the standardization process in which the data calculation unit 32 calculates the antibody data D3 based on the measurement data D1 will be described in more detail, while comparing the measurement data D1 shown in FIG. 3 with the antibody data D3 shown in FIG. 5. The data calculation unit 32 calculates the standardized antibody data D3 based on the measurement data D1, according to the first to fifth standardization process steps shown below.

In the first standardization process step, the data calculation unit 32 divides, for each of the first number of target proteins TP, the first fluorescence intensity value D11 of the NC and the first fluorescence intensity values D11 associated with each of the first to sixth diagnostic data D21 to D26 by the average value of each of the first fluorescence intensity values D11. As a result, the data calculation unit 32 obtains a first standardized index value for each of the first number of target proteins TP in association with the NC and each of the first to sixth diagnostic data D21 to D26.

In the second standardization processing step, the data calculation unit 32 divides the second fluorescence intensity value D12 of the NC and the second fluorescence intensity value D12 associated with each of the first to sixth diagnostic data D21 to D26 by each of the first standardized index values for each of the first number of target proteins TP. As a result, the data calculation unit 32 obtains the second standardized index value for each of the first number of target proteins TP in association with the NC and each of the first to sixth diagnostic data D21 to D26.

In the third standardization processing step, the data calculation unit 32 subtracts the second standardized index value associated with NC from each of the second standardized index values associated with the first to sixth diagnostic data D21 to D26 for each of the first number of target proteins TP. In this way, the data calculation unit 32 obtains a third standardized index value for each of the first number of target proteins TP in association with each of the first to sixth diagnostic data D21 to D26.

In the fourth standardization process step, the data calculation unit 32 subtracts the value at the 25th percentile from each of the third standardized index values for the first number of target proteins TP for each of the first to sixth diagnostic data D21 to D26, and divides the subtracted value by the standard deviation. In this way, the data calculation unit 32 obtains a fourth standardized index value for each of the first number of target proteins TP in association with each of the first to sixth diagnostic data D21 to D26.

In the fifth standardization process step, the data calculation unit 32 divides each of the fourth standardized index values for the first number of target proteins TP by the value at the 75th percentile for each of the first to sixth diagnostic data D21 to D26, and multiplies the divided value by "10". In this way, the data calculation unit 32 obtains standardized antibody data D3 for each of the first number of target proteins TP in association with each of the first to sixth diagnostic data D21 to D26.

By performing standardization processing according to the first to fifth standardization processing steps described above, the data calculation unit 32 can calculate standardized antibody data D3 that appropriately absorbs the measurement error of the measurement data D1 caused by differences in the amount of the target protein TP fixed on the substrate 41 of the protein microarray 4 and the measurement error between the measurement data D1 corresponding to each of the first to sixth diagnostic data D21 to D26. In this case, the antigen selection unit 33 can input the standardized antibody data D3 as input data to the machine learning model LM, and output an accurate specific target protein TPS as output data corresponding to each of the multiple diseases indicated in the diagnostic data D2.

In the antigen search device 3 according to this embodiment, the antigen selection unit 33 may be configured to extract various proteins associated with each of a plurality of diseases based on information registered in existing databases such as OMIM, Genecards, and AAgAtlas. In this case, the antigen selection unit 33 can narrow down the proteins that are correlated with the disease-related antigen AG associated with each of a plurality of diseases to those that are common to the various proteins extracted for a specific target protein TPS output from the machine learning model LM.

The specific embodiments described above mainly include inventions having the following configurations:

According to this antigen search method, antibody data relating to the binding associated with the antigen-antibody reaction between a first number of target proteins fixed on a protein microarray substrate and autoantibodies in multiple test samples is input as input data into a machine learning model, and specific target proteins that are correlated with disease-associated antigens selected from the first number of target proteins based on the antibody data for each of the multiple diseases indicated in the diagnostic data can be output as output data. This makes it possible to efficiently search for specific target proteins that are correlated with each disease-associated antigen associated with each of the multiple diseases, compared to searching for target proteins individually for each of the multiple diseases.

In the above antigen search method, the antibody data may include a data group of first antibody data associated with the diagnostic data indicating a diagnosis result of not having a disease, and a data group of second antibody data associated with the diagnostic data indicating a diagnosis result of having a disease. In this case, the antigen selection step satisfies the condition that the first antibody data is equal to or less than a first threshold value and the second antibody data is equal to or greater than a second threshold value that is greater than the first threshold value, and a second number of target proteins less than the first number are extracted from the first number of target proteins, and the specific target protein is selected from the second number of target proteins.

In this embodiment, when selecting a specific target protein that is correlated with a disease-related antigen for each of a plurality of diseases indicated in the diagnostic data according to the machine learning model, the specific target protein is selected from a second number of target proteins that is less than the first number of target proteins immobilized on the substrate of the protein microarray. This makes it possible to shorten the time required to select a specific target protein using the machine learning model, and to more accurately search for a specific target protein corresponding to each of a plurality of diseases.

In the antigen search method described above, the antigen selection step may involve determining a similarity between the antibody data for each of the second number of target proteins, and classifying the second number of target proteins into a plurality of groups corresponding to each of a plurality of diseases indicated in the diagnostic data by hierarchical clustering based on the similarity, thereby selecting target proteins belonging to each of the plurality of groups as the specific target proteins.

In this embodiment, the second number of target proteins are classified into a plurality of groups by hierarchical clustering based on the similarity between the antibody data for each of the second number of target proteins. Then, the target proteins belonging to each of the plurality of groups can be selected as specific target proteins that are correlated with each disease-associated antigen associated with each of the plurality of diseases.

In hierarchical clustering, antibody data for each of the second number of target proteins is assigned to one cluster, and clusters with similar antibody data are recursively combined. Hierarchical clustering includes methods such as the shortest distance method, the longest distance method, and the group average method, depending on the criteria for selecting clusters to be combined. The similarity between clusters is defined as the similarity between antibody data across clusters. Clusters are combined in order from the pair with the greatest similarity to form one cluster, and the combination is stopped when the similarity between all clusters falls below a predetermined similarity reference value. In hierarchical clustering, when the combination of clusters is stopped, the second number of target proteins are classified into multiple groups corresponding to each of multiple diseases. In this case, a maximum limit is set for the number of target proteins belonging to each group, and if the total number of antibody data included in the two clusters to be combined is greater than the maximum limit, the similarity is considered to be zero, making it possible to prevent the number of target proteins belonging to each group from exceeding the maximum limit.

In the antigen search method described above, in the antigen selection step, the machine learning model may be generated using a genetic algorithm with the goal that the diseases indicated in the diagnostic data corresponding to the antibody data for each of the target proteins belonging to each of the multiple groups are identical.

In this embodiment, a machine learning model is generated using a genetic algorithm, in which antibody data for each of a second number of target proteins associated with diagnostic data is used as input data, and specific target proteins belonging to each group classified by hierarchical clustering are used as output data. By using a genetic algorithm, a machine learning model can be generated with the goal of identifying the same disease indicated by diagnostic data corresponding to each antibody data for each target protein belonging to each group classified by hierarchical clustering.

In the genetic algorithm, for example, the identity of each disease corresponding to each target protein belonging to each group classified by hierarchical clustering is evaluated, and if the diseases are the same, a first reward is given, and if the diseases are not the same, a second reward lower than the first reward is given. By repeatedly performing such a genetic algorithm, it is possible to generate a machine learning model in which the diseases corresponding to each target protein belonging to each group classified by hierarchical clustering are the same. In this way, when antibody data for each of the second number of target proteins is input as input data to the machine learning model generated using the genetic algorithm, the second number of target proteins can be classified by hierarchical clustering into multiple groups corresponding to each of the multiple diseases indicated in the diagnostic data.

In the above antigen search method, the data calculation step may involve performing a predetermined standardization process to absorb measurement errors between the measurement data corresponding to each of the multiple diseases indicated in the diagnostic data, thereby calculating standardized antibody data for each of the first number of target proteins.

When using protein microarrays, the measurement data may have low reproducibility. For this reason, when searching for specific target proteins that are correlated with disease-related antigens related to each of a plurality of diseases using multiple protein microarrays corresponding to each of a plurality of diseases, it is necessary to take into account the measurement error between the measurement data corresponding to each of the plurality of diseases. Therefore, for the antibody data input as input data to the machine learning model, a predetermined standardization process is performed to absorb the measurement error between the measurement data corresponding to each of the plurality of diseases, and standardized antibody data is calculated for each of the first number of target proteins. In this case, by inputting the standardized antibody data as input data to the machine learning model, it is possible to output accurate specific target proteins as output data corresponding to each of the plurality of diseases indicated in the diagnostic data.

In the above antigen discovery method, the data calculation step may perform the standardization process using a value at least at one percentile for the antibody data for each of the multiple diseases indicated in the diagnostic data, thereby calculating standardized antibody data for each of the first number of target proteins.

In this embodiment, standardization processing is performed on the antibody data for each of the multiple diseases indicated in the diagnostic data, using a value at least at one percentile. This percentile-based standardization processing makes it possible to calculate standardized antibody data that appropriately absorbs the measurement error between the measurement data corresponding to each of the multiple diseases.

In the above antigen search method, the measurement data may include a first fluorescence intensity value, the feature value being the intensity of fluorescence emitted from a first fluorescent substance in a first secondary antibody that is capable of binding to a predetermined reference antibody that binds to any of the first number of target proteins and is labeled with a first fluorescent substance, and a second fluorescence intensity value, the feature value being the intensity of fluorescence emitted from a second fluorescent substance in a second secondary antibody that is capable of binding to autoantibodies in the multiple test samples and is labeled with a second fluorescent substance. In this case, in the data calculation step, the standardization process is performed using the first fluorescence intensity value and the second fluorescence intensity value, thereby calculating the standardized antibody data for each of the first number of target proteins.

In protein microarrays, there may be variation in the amount of a first number of target proteins immobilized on a substrate. For this reason, when searching for specific target proteins that correlate with disease-associated antigens related to each of a number of diseases using protein microarrays corresponding to each of a number of diseases, it is necessary to take into account measurement errors caused by differences in the amount of target proteins immobilized on the substrate for the measurement data corresponding to each of a number of diseases.

Therefore, as measurement data for each of the first number of target proteins fixed on the substrate of the protein microarray, a first fluorescence intensity value related to the intensity of fluorescence emitted from a first fluorescent substance of a first secondary antibody bound to a predetermined reference antibody bound to all of the first number of target proteins, and a second fluorescence intensity value related to the intensity of fluorescence emitted from a second fluorescent substance of a second secondary antibody bound to an autoantibody in the test sample are obtained. Then, standardization processing is performed using the first fluorescence intensity value and the second fluorescence intensity value for the antibody data input as input data to the machine learning model, thereby calculating standardized antibody data for each of the first number of target proteins. This makes it possible to calculate standardized antibody data for measurement data corresponding to each of the multiple diseases, in which measurement errors caused by differences in the amount of target protein fixed on the substrate are appropriately absorbed. In this case, by inputting the standardized antibody data as input data to the machine learning model, it is possible to output accurate specific target proteins as output data corresponding to each of the multiple diseases indicated in the diagnostic data.

According to this antigen search system, the antigen selection unit of the antigen search device inputs antibody data relating to the binding associated with the antigen-antibody reaction between a first number of target proteins fixed on a protein microarray substrate and autoantibodies in multiple test samples as input data into a machine learning model, and is then able to output, as output data, specific target proteins that are correlated with disease-associated antigens selected from the first number of target proteins based on the antibody data, corresponding to each of the multiple diseases indicated in the diagnostic data. This makes it possible to efficiently search for specific target proteins that are correlated with each disease-associated antigen associated with each of the multiple diseases, compared to searching for target proteins individually corresponding to each of the multiple diseases.

As described above, according to the present invention, it is possible to provide an antigen discovery method and an antigen discovery system that can efficiently discover target proteins that are correlated with disease-related antigens that are associated with each of a plurality of diseases, using a protein microarray in which a plurality of target proteins are fixed onto a substrate.

Claims

1. An antigen discovery method for discovering a target protein having a correlation with a disease-associated antigen associated with each of a plurality of diseases, using a protein microarray having a first number of target proteins immobilized on a substrate, the method comprising:
a data acquiring step of acquiring measurement data for each of the first number of target proteins, the measurement data relating to a predetermined feature amount when each of a plurality of test specimen samples derived from a plurality of test specimens is contacted with the first number of target proteins;
a data calculation step of calculating antibody data for each of the first number of target proteins, which is related to binding properties associated with an antigen-antibody reaction between the autoantibody in the plurality of test specimens and the first number of target proteins, based on the measurement data, in association with diagnostic data indicating diagnostic results for a plurality of diseases for each of the plurality of test specimens;
an antigen selection step of generating a machine learning model using the antibody data for each of the first number of target proteins associated with the diagnostic data as input data, selecting a specific target protein that has a correlation with the disease-associated antigen from among the first number of target proteins based on the antibody data, corresponding to each of a plurality of diseases indicated in the diagnostic data, in accordance with the machine learning model, and outputting the selected specific target protein as output data.
the antibody data includes a data group of first antibody data associated with the diagnostic data indicating a diagnosis result of not having a disease, and a data group of second antibody data associated with the diagnostic data indicating a diagnosis result of having a disease,
2. The antigen searching method according to claim 1, wherein the antigen selection step satisfies the conditions that the first antibody data is equal to or less than a first threshold value and the second antibody data is equal to or greater than a second threshold value that is greater than the first threshold value, and a second number of target proteins less than the first number are extracted from the first number of target proteins, and the specific target protein is selected from the second number of target proteins.
The antigen search method according to claim 2, wherein in the antigen selection step, a similarity between the antibody data for each of the second number of target proteins is calculated, and the second number of target proteins are classified into a plurality of groups corresponding to each of a plurality of diseases indicated in the diagnostic data by hierarchical clustering based on the similarity, thereby selecting a target protein belonging to each of the plurality of groups as the specific target protein.
The antigen search method according to claim 3, wherein in the antigen selection step, the machine learning model is generated using a genetic algorithm with the goal that each disease indicated in the diagnostic data corresponding to each of the antibody data for each of the target proteins belonging to each of the multiple groups is the same.
The antigen search method according to claim 1, wherein in the data calculation step, a predetermined standardization process is performed to absorb measurement errors between the measurement data corresponding to each of the multiple diseases indicated in the diagnostic data, thereby calculating standardized antibody data for each of the first number of target proteins.
The antigen discovery method according to claim 5, wherein in the data calculation step, the standardization process is performed using at least one percentile value for the antibody data for each of the multiple diseases indicated in the diagnostic data, thereby calculating standardized antibody data for each of the first number of target proteins.
The measurement data is
a first secondary antibody capable of binding to a predetermined reference antibody that binds to any of the first number of target proteins and that is labeled with a first fluorescent substance, the first secondary antibody having an intensity of fluorescence emitted from the first fluorescent substance as the feature amount; and
a second fluorescent intensity value, the second fluorescent intensity value being determined as the characteristic quantity based on an intensity of fluorescent light emitted from a second fluorescent substance in a second secondary antibody capable of binding to an autoantibody in the plurality of test samples and labeled with a second fluorescent substance;
The antigen searching method according to claim 5, wherein in the data calculation step, the standardization process is performed using the first fluorescence intensity value and the second fluorescence intensity value, thereby calculating standardized antibody data for each of the first number of target proteins.
a measuring device that uses a protein microarray having a first number of target proteins immobilized on a substrate, measures a predetermined feature amount when each of a plurality of test specimen samples derived from a plurality of test specimens is contacted with the first number of target proteins, and outputs measurement data for each of the first number of target proteins that indicates the measurement results;
An antigen searching device that searches for a target protein having a correlation with a disease-related antigen associated with each of a plurality of diseases,
The antigen search device is
A data acquisition unit for acquiring the measurement data;
a data calculation unit that calculates antibody data for each of the first number of target proteins, which is related to binding properties associated with an antigen-antibody reaction between the autoantibody in the plurality of test specimens and the first number of target proteins, based on the measurement data, in association with diagnostic data indicating diagnostic results for a plurality of diseases for each of the plurality of test specimens;
an antigen selection unit that generates a machine learning model using the antibody data for each of the first number of target proteins associated with the diagnostic data as input data, selects a specific target protein that has a correlation with the disease-associated antigen from among the first number of target proteins based on the antibody data, corresponding to each of a plurality of diseases indicated in the diagnostic data, in accordance with the machine learning model, and outputs the selected specific target protein as output data.