DISEASE DETECTION SUPPORT DEVICE, DISEASE DETECTION SUPPORT METHOD, AND DISEASE DETECTION SUPPORT PROGRAM
The disease determination support system improves diagnostic accuracy by analyzing biomarkers through deep learning and CNNs, addressing the inadequacies of existing methods in disease detection.
Patent Information
- Application Number
- JP2021526141
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2019-06-11
- Filing Date
- 2020-06-11
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2040-06-11
AI Technical Summary
Existing methods for determining disease morbidity lack sufficient accuracy and practicality for clinical use, particularly in early detection of conditions like cancer.
A disease determination support system utilizing deep learning to analyze multiple biomarkers in biological samples, converting analysis results into images, and employing convolutional neural networks (CNN) to improve diagnostic accuracy by emphasizing differences in glycopeptide abundance between healthy and diseased individuals.
Enhances the accuracy of disease detection, particularly in early stages, by leveraging glycopeptide analysis and deep learning techniques, enabling effective clinical application.
Smart Images

Figure 0007589948000001 
Figure 0007589948000002 
Figure 0007589948000003
Abstract
Description
[Technical field]
[0001] The present technology relates to a determination support device, a determination support method, and a determination support program for supporting the determination of the presence or absence of a disease state based on characteristics of a biological sample or the like. [Background technology]
[0002] The desire to be healthy is a universal wish, regardless of country or culture. However, it is difficult to say that medical services, including tests and treatments, which have been technologically advanced and improved in quality every day, are adequately addressing this need. For example, in the case of diseases such as cancer, tumors are often discovered late or in advanced cases with metastasis because there are few subjective symptoms even when they occur. This places a heavy burden on the patient and those around them, and the social losses are also extremely large, so there is a need for technological developments that enable early detection. In addition, from the perspective of controlling social security costs, which have been increasing in recent years, it is becoming increasingly important to be able to accurately grasp the prevention of lifestyle-related diseases and the state of disease.
[0003] In order to investigate the disease state, specimen tests that analyze biological samples such as blood tests, which are simpler and less expensive than image diagnosis, are desirable, and especially for detecting the early stages of disease without subjective symptoms, methods that are less physically and financially burdensome are desirable. However, there are only a few types of biomarkers that can be used to determine the disease state, and many of them are not sufficiently accurate. It has often been very difficult to detect and determine the early stages of a disease using only specimen tests that use conventionally known biomarkers.
[0004] Taking cancer as an example, it is known that the sugar chain structure of glycoproteins contained in the serum of cancer patients changes with canceration, and it has been reported that structural changes in the sugar chains of blood glycoproteins associated with ovarian cancer can also be detected with CA125, an existing ovarian cancer marker. Patent Document 1 describes that a glycoprotein having a sugar chain added to an asparagine residue at a specific position, or a fragment thereof having a sugar chain, can be used as a marker for distinguishing epithelial ovarian cancer.
[0005] In addition, CA125, a marker used to diagnose ovarian cancer, also shows high values in endometriosis. Patent Document 2 proposes an ovarian cancer marker glycoprotein that can be easily distinguished from endometriosis, and a method for detecting ovarian cancer using the same.
[0006] Furthermore, Patent Document 3 proposes a method in which preprocessed values of biomarker concentrations are input to a trained CNN to generate an output value corresponding to the presence or absence or severity of a disease. [Prior art documents] [Patent documents]
[0007] [Patent Document 1] JP 2012-145500 A [Patent Document 2] JP 2015-125093 A [Patent Document 3] JP 2007-52774 A Summary of the Invention [Problem to be solved by the invention]
[0008] Although various methods have been proposed so far as methods for determining the disease state, few of them have sufficient accuracy and are actually practical in clinical practice. Therefore, the present invention aims to provide a determination support device, method, and program that can improve the accuracy of disease state determination. [Means for solving the problem]
[0009] The disease assessment support device of the present invention comprises a pre-processing unit that rearranges the analysis results obtained by analyzing multiple predetermined types of biomarkers in a biological sample derived from a subject to be assessed (hereinafter simply referred to as "subject") in a predetermined order and converts the rearranged analysis results into an image, and a assessment support unit that outputs information indicating the possibility that the subject is suffering from the disease using the above image and a trained model that has learned by deep learning the relationship between the image obtained by rearranging the analysis results of multiple biomarkers in biological samples derived from a subject with a specific disease and a subject without the disease and then converting the information into an image and the disease state of the subject.
[0010] The type of the biomarker may be any one of a protein, a nucleic acid, a peptide, a sugar chain, and a lipid, or a combination of two or more of these.
[0011] The predetermined type of biomarker may also include a plurality of glycopeptides generated by fragmenting glycoproteins with proteases. The analysis results may be expressed by analyzing the plurality of glycopeptides multiple times with a mass spectrometer, selecting the glycopeptides having a reproducibility equal to or higher than a predetermined standard, and using the peaks of the selected glycopeptides. These glycopeptides may be detected by any method, such as immunoassay, liquid chromatography (LC), mass spectrometry (MS), lectin array, electrophoresis, etc., but liquid chromatography-mass spectrometry (LC-MS) is particularly preferable. When using a mass spectrometer, the same sample is measured multiple times in advance, and glycopeptides based on the obtained peak information with high reproducibility are selected, thereby making it possible to appropriately learn the characteristics and improve the accuracy of cancer determination.
[0012] The target disease is not particularly limited, but may be cancer, the cure rate of which is dramatically improved by early detection. In the step of learning about cancer and non-cancer using a deep learning method, the subjects include cancer patients and non-cancer patients, and the step of rearranging the analysis results in a predetermined order may be determined based on the characteristics of the relative abundance of glycopeptides obtained from the blood of cancer patients to glycopeptides obtained from a predetermined control sample. By determining the order in this manner, glycopeptides that are particularly different in abundance between cancer patients and non-cancer patients can be highlighted, and the differences in characteristics between cancer patients and non-cancer patients can be efficiently learned and judgment support can be provided.
[0013] The step of rearranging the analysis results in a predetermined order may be determined based on the degree of similarity in relative abundance of a plurality of glycopeptides obtained by principal component analysis, cluster analysis, or factor analysis. Specifically, by using the data rearranged in the order determined in this manner, the differences in characteristics between cancer patients and non-cancer patients can be efficiently learned and the diagnosis can be supported.
[0014] The step of visualizing the sorted numerical data may convert the numerical data into an image based on a certain criterion. Specifically, the method may be to change the color intensity or the type of color according to the magnitude of the numerical value. When multiple numerical data are used, it is preferable to use a combination of multiple colors by, for example, associating each type of numerical data with one of the three primary colors of light, thereby improving the accuracy of the judgment. By making the input data to the deep learning image data, it becomes possible to easily use various existing deep learning methods implemented as image classifiers.
[0015] The preprocessing unit may convert a value based on the analysis result of a tumor marker different from the predetermined type of biomarker into a color different from the color assigned to the predetermined type of biomarker among the three primary colors and add it to the image, or convert it into a color in a region different from the predetermined type of biomarker and add it to the image, and the judgment support unit may output information indicating the possibility that the subject is suffering from a disease using a trained model that has learned the relationship between the image created based on the analysis result of the predetermined type of biomarker and the tumor marker and the disease state by deep learning. The accuracy of judgment can be improved by using the analysis result of a general tumor marker.
[0016] Deep learning may be performed using a neural network or a convolutional neural network (CNN). A convolutional neural network can be suitably used for learning processing of images, for example. In particular, transfer learning using a model with prior learning is preferable.
[0017] The contents described in the Means for Solving the Problems can be combined as much as possible without departing from the problem and technical idea of the present invention. The contents of the means for solving the problem can be provided as an apparatus such as a computer or a system including multiple apparatuses, a method executed by a computer, or a program executed by a computer. The program can also be executed on a network. A recording medium holding the program may be provided. Effect of the Invention
[0018] According to the present invention, it is possible to provide a judgment support device, method, and program that can improve the accuracy of judging the morbid state of a disease. [Brief description of the drawings]
[0019] [Figure 1]FIG. 1 is a diagram illustrating an example of machine learning of characteristics of cancer that appear in blood and a determination support for the presence or absence of cancer using a determination support model obtained by machine learning, according to this embodiment. [Diagram 2] FIG. 2 is a functional block diagram illustrating an example of a learning device. [Diagram 3] FIG. 3 is a process flow diagram showing an example of a process according to the present embodiment. [Figure 4] FIG. 4 is a process flow diagram showing an example of the model creation process according to this embodiment. [Diagram 5] FIG. 5 is a functional block diagram illustrating an example of a judgment support device. [Figure 6] FIG. 6 is a process flow diagram illustrating an example of the determination support process. [Figure 7] FIG. 7 shows the breakdown and uses of serum samples. [Figure 8] FIG. 8 shows an example of the peaks of measured glycopeptide fragments. [Figure 9] FIG. 9 is a diagram showing the results of the principal component analysis. [Figure 10] FIG. 10 is a diagram showing an example of the created image. [Figure 11] FIG. 11 shows the results of the ROC analysis. [Figure 12] FIG. 12 is a diagram for explaining the evaluation of Examples 1-4. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0020] Hereinafter, an example of an embodiment of the present invention will be described with reference to the drawings. In particular, the description will focus on support for diagnosis of ovarian cancer using glycopeptides, but the present invention is not limited thereto.
[0021] In the present invention, a disease refers to a state in which the normal state of the body is impaired and the life-sustaining functions are inhibited or altered, and refers to a state in which a completely good physical, mental, and social state is broken down. These may be synonymous disease names classified by the International Classification of Diseases (hereinafter referred to as ICD) or the like. In addition, even if there are no abnormalities in specimen tests or obvious symptoms, these may be targets, since they may be in the early stages of a disease state.
[0022] The biological sample used in the specimen test refers to a sample containing components derived from a living organism, and may be, for example, whole blood, plasma, serum, blood cells, urine, stool, saliva, sputum, semen, tears, nasal discharge, vaginal, nasal, rectal, pharyngeal, spinal fluid, and urethral swabs, excretions, and secretions, as well as biopsy tissue samples, or combinations thereof. The types of biological samples may be appropriately selected and used in consideration of ease of handling, such as specimen collection and pretreatment. Those skilled in the art can appropriately set conditions for the pretreatment method depending on the type of biological sample and the type of biomarker to be treated.
[0023] A biomarker refers to an index for objectively evaluating a physical condition, and in the present invention, a biomarker that is particularly used for the diagnosis of a disease may be used. The type of biomarker used may be used alone or in combination. For example, the present invention may be used with biomarkers such as SCC, CEA, SLX, CYFRA, NSE, ProGRP, AFP, PIVKA-II, CA19-9, PSA, CA15-3, NCC-ST-439, STN, ElastaseI, βHCG, CA125, HE4, and SLX alone or in combination. When used in the determination of ovarian cancer, it may be used in combination with CA125 or HE4, which are generally recognized for the diagnosis of ovarian cancer. In addition to analysis results obtained by specimen tests such as biochemical tests, blood tests, and tumor markers, diagnostic imaging data such as CT (Computed Tomography), MRI (Magnetic Resonance Imaging), and PET (Positron Emission Tomography) are also included, as well as vital signs used in daily medical examinations such as body temperature and pulse rate. Any substance contained in a biological sample may be the subject of analysis, and preferably proteins, nucleic acids (DNA (Deoxyribonucleic acid), RNA (Ribonucleic acid)), peptides, sugar chains, lipids, and the like can be used. Furthermore, a combination of multiple types of these substances may be used.
[0024] Any means may be used to measure and analyze these biomarkers. Those skilled in the art can select an appropriate analysis method and set the conditions according to the purpose of analysis, such as the type and concentration of the biomarker to be analyzed. For example, instrumental analysis such as mass spectrometers and chromatography, analysis methods using immune reactions such as ELISA, latex agglutination, immunoturbidimetric assay, and flow cytometer methods, test methods that measure absorbance such as enzyme methods, ultraviolet spectrophotometry, enzyme immunoassays, and in the case of luminescence measurement, chemiluminescence immunoassays and electrochemiluminescence immunoassays, and genetic analysis methods such as TaqMan (registered trademark) PCR, Invader (registered trademark), Sniper, SNPIT, Pyrominisequencing, DHPLC, NanoChip, LAMP, hybridization assay, and sequencing can be used. In addition, when using indicators such as vital signs, physiological function tests such as pulse, body temperature, blood pressure, electrocardiogram, electroencephalogram, ultrasound, and respiratory function tests can also be used. Further, radiological examinations such as X-ray examinations, CT examinations, MRI examinations, nuclear medicine examinations (RI examinations using radioisotopes), and endoscopy examinations may be used. In the present invention, in the support for the diagnosis of ovarian cancer using glycopeptides, analysis by a mass spectrometer is preferably used, but may be appropriately selected and used. When using the analysis results by a mass spectrometer, it can be performed using a liquid chromatograph (LC) device and a mass spectrometer (MS). The liquid chromatograph device and the mass spectrometer may be connected in series, or each may be an independent device. For example, an LC-MS system consisting of a liquid chromatograph device and a mass spectrometer connected in series can be used. By using the LC-MS system, the components separated by liquid chromatography can be continuously analyzed by mass analysis.
[0025] In the cancer diagnosis assistance described as an example of this embodiment, glycoproteins in the blood, or their degradation products, glycopeptides (collectively referred to as "glycoproteins"), are used to perform machine learning to determine the relationship between the presence or absence of cancer, and the created classifier is used to output information to assist in the diagnosis of disease, indicating the degree of possibility of cancer from a given sample.
[0026] In this embodiment, the term "cancer" refers to any malignant neoplasm, including brain, nerve, and eye cancers such as brain tumors and gliomas, mouth and throat cancers such as tongue cancer, nasopharyngeal cancer, oropharyngeal cancer, hypopharyngeal cancer, laryngeal cancer, and thyroid cancer, chest cancers such as lung cancer, thymoma, thymic cancer, mesothelioma, and breast cancer, digestive tract cancers such as esophageal cancer, gastric cancer, colon cancer (colon cancer and rectal cancer), and gastrointestinal stromal tumor (GIST), and liver cancers such as hepatocellular carcinoma, bile duct cancer, gallbladder cancer, and pancreatic cancer. -Gallbladder and pancreatic cancer, renal cell carcinoma, renal pelvis and ureter cancer, bladder cancer, and other urinary tract cancers; prostate cancer, testicular (testicular) tumors, breast cancer, cervical cancer, uterine cancer (endometrial cancer), ovarian cancer, vaginal cancer, vulvar cancer, basal cell carcinoma, squamous cell carcinoma, malignant melanoma (skin), skin cancers such as cutaneous lymphoma, soft tissue sarcoma, and bone and muscle cancers; acute myeloid leukemia, acute lymphocytic leukemia / These include, but are not limited to, leukemias such as lymphoblastic lymphoma, chronic myeloid leukemia, chronic lymphocytic leukemia / small lymphocytic lymphoma, myelodysplastic syndrome, adult T-cell leukemia / lymphoma, Hodgkin's lymphoma, non-Hodgkin's lymphoma, follicular lymphoma, MALT lymphoma, lymphoplasmocytic lymphoma, mantle cell lymphoma, diffuse large B-cell lymphoma, peripheral T-cell lymphoma, Burkitt's lymphoma, extranodal NK / T-cell lymphoma, nasal type and other skin lymphomas, acute lymphocytic leukemia / lymphoblastic lymphoma, chronic lymphocytic leukemia / small lymphocytic lymphoma, adult T-cell leukemia / lymphoma, and other malignant lymphomas, multiple myeloma, cancer of unknown primary origin, hereditary tumors, familial tumors, etc. Also, the present invention can be implemented for multiple cancers, rather than for a single type of cancer.
[0027] Furthermore, a glycoprotein refers to a glycoprotein having at least an N-linked or O-linked glycan, and a glycopeptide refers to a peptide having a molecular weight of 10,000 or less in the natural state and having an N-linked or / and O-linked glycan, or a peptide having at least an N-linked or O-linked glycan among fragments of glycoproteins degraded with proteases such as trypsin and lysyl endopeptidase.
[0028] FIG. 1 is a diagram showing an example of machine learning of cancer characteristics appearing in blood and cancer presence / absence determination by a determination model obtained by machine learning according to the present embodiment. In the machine learning according to the present embodiment, first, proteins (e.g., glycoproteins) 11 contained in blood 1 of cancer patients and non-cancer patients (also called "healthy subjects") are reduced and fragmented to obtain peptides (e.g., glycopeptides) 12. Then, the peptides 12 contained in blood 1 are analyzed using a mass spectrometer 2. The mass spectrometer 2 performs, for example, liquid chromatography mass spectrometry (LC-MS) or liquid chromatography-tandem mass spectrometry (LC-MS / MS) to output a mass spectrum 21. A person skilled in the art can set appropriate conditions for the ionization method and individual measurement conditions in mass spectrometry according to the type of sample to be measured and the type of biomarker to be measured.
[0029] Also, peaks may be selected and rearranged from the mass spectrum 21 based on a predetermined rule (including those based on information such as "peak height" and "peak area"). In the example of FIG. 1, for peaks of cancer patients compared to healthy subjects, information on intensity and reproducibility is used, and the data is arranged two-dimensionally based on a predetermined rule, and an image 22 is created as a two-dimensional code in which a color of a predetermined density is assigned to one pixel or a predetermined region according to the peak intensity. As the information on the peaks to be used, peak intensity may be used alone or in combination with other information, for example, information on reproducibility. When information on reproducibility is used, it is preferable to create a model with higher accuracy by targeting peptides with high reproducibility. In addition, deep learning (also called "deep learning") is performed using a CNN (convolutional neural network) with the image 22 as an input, and a trained model 23 for outputting the possibility of cancer from the image 22 is created. As described above, the relationship between the expression pattern of peptides, and therefore the characteristics of the expression pattern of proteins in the blood, and the presence or absence of cancer is machine-learned.
[0030] In the judgment support process, for a subject whose disease status is unknown, a mass spectrum 21 of peptides 12 is obtained from blood 1 in the same manner as in machine learning. An image 22 is also created using the peaks of peptides 12 selected in the machine learning. Then, with the image 22 as input, information indicating the degree of possibility of cancer is output using a trained model 23. A user such as a doctor can refer to the output information and use it to diagnose the subject.
[0031] In the example of FIG. 1, one learning / diagnosis support device 3 is shown, but machine learning and cancer diagnosis support may be performed by different devices. In addition, some processes, such as machine learning using images 22 and output of the possibility of morbidity, may be performed by different devices. The different devices may be connected via a network and provide so-called cloud services. Below, the learning device and the diagnosis support device will be described separately.
[0032] <Learning device> 2 is a functional block diagram showing an example of a learning device 3. The learning device 3 is a computer and includes a communication I / F 31, a storage device 32, an input / output device 33, and a processor 34, and these components are connected via a bus 35.
[0033] The communication I / F 31 is, for example, a wired network card or a wireless communication module, and communicates with other computers based on a predetermined protocol. For example, the communication I / F 31 transmits and receives data to and from other computers via a communication network such as the Internet or a LAN (Local Area Network).
[0034] The storage device 32 is, for example, a main storage device such as a random access memory (RAM) or a read only memory (ROM), or an auxiliary storage device such as a hard disk drive (HDD), a solid state drive (SSD), an embedded multi-media card (eMMC), or a flash memory. The main storage device temporarily holds data intermediately generated in the processing described below, and secures a working area for the processor 14. The auxiliary storage device stores the program according to this embodiment and other data.
[0035] The input / output device 33 is a user interface such as an input device such as a keyboard, a mouse, an output device such as a monitor, an input / output device such as a touch panel, etc. The learning device 3 accepts a user's operation via the input / output device 33 and executes the process according to this embodiment.
[0036] The processor 34 is an arithmetic processing device such as a CPU (Central Processing Unit), and performs processing described below by executing a program according to this embodiment. In the example of Fig. 2, functional blocks are shown in the processor 34. That is, the processor 34 functions as a peak selection unit 341, a preprocessing unit 342, a deep learning unit 343, and a model verification unit 344 by executing the program according to this embodiment.
[0037] The peak selection unit 341 analyzes the blood of a cancer patient using the analyzer 2, and selects multiple peptide peaks to be used in machine learning processing from the output mass spectrum 21. The preprocessing unit 342 categorizes multiple peptides based on information such as intensity and reproducibility, for example, using principal component analysis (PCA) or cluster analysis, and rearranges them according to the similarity of fluctuation based on the result. For example, peptides whose amount changes with canceration or peptides whose amount does not change may be rearranged in order of similarity of the degree of fluctuation. In this embodiment, the peak intensities of the cancer patient relative to the peak intensities obtained from the blood of a control sample (a mixture of blood of a healthy person and a cancer patient, analyzed simultaneously at each test) are rearranged, and image data is created by rearranging rectangular regions to which the peak intensity ratios are assigned in two dimensions. The deep learning unit 343 uses each region of the image data as an input, performs deep learning using the presence or absence of cancer as a teacher value, and creates a classifier (also called a "trained model") that classifies the presence or absence of cancer of a subject based on the image data. The model verification unit 344 determines the presence or absence of cancer using the created trained model and blood (also called a "test sample") from a cancer patient different from the blood (also called a "learning sample") used to create the trained model, and verifies the accuracy.
[0038] In the case of analysis using a mass spectrometer as a means for analyzing biomarkers, mass spectrometry is used, but results obtained by other analysis methods can be used in the same way. In that case, the selection of data to be used, the procedure for quantification if necessary, data sorting, and image data creation can also be performed according to the same procedure. As described above, the learning device 3 creates a model for outputting information indicating the possibility of morbidity for a specific disease.
[0039] <Processing> 3 is a process flow diagram showing an example of the process according to this embodiment. In this embodiment, the process can be divided into a model creation process (FIG. 3: S1) executed by a learning device and a judgment support process (FIG. 3: S2) executed by a judgment support device. Note that the model creation process and the judgment support process do not need to be executed consecutively, and for example, only a device, method, or program that executes the judgment support process may be provided.
[0040] <Model creation process> FIG. 4 is a process flow diagram showing an example of a model creation process according to this embodiment. In this embodiment, blood 1 from each of a cancer patient and a healthy person (collectively referred to as "subject") is used as a specimen, and it is preferable to separate serum and plasma to use as a specimen. Then, proteins are extracted from the specimen, reduced, and fragmented into peptides (FIG. 4: S11). The process of extracting proteins from the specimen involves, for example, adding a solvent in an amount 2 to 10 times that of the specimen. The solvent may be any solvent that precipitates proteins. For example, the solvent is preferably acetone, methanol, ethanol, trichloroacetic acid, an aqueous hydrochloric acid solution, or the like, and a mixture of acetone and trichloroacetic acid is particularly preferable. The precipitated protein is denatured, reduced and alkylated, and fragmented into peptides using a protease. The protease may be any protease that decomposes proteins into peptide fragments. For example, the protease is preferably trypsin or lysyl endopeptidase, or both of these.
[0041] The peptides after decomposition are then analyzed by a mass spectrometer 2 to obtain a mass spectrum 21 (FIG. 4: S12). The peptides to be analyzed are preferably concentrated into glycopeptides using a lectin column or ultrafiltration filter, but may also include peptides without glycans. The mass spectrometer 2 may be any type capable of simultaneously analyzing peptides. For example, LC-MS is preferred, and quadrupole, TOF, triple Q, orbitrap, etc. are particularly preferred.
[0042] Furthermore, the peak selection unit 341 of the learning device 3 in Fig. 2 selects the above peak as a predetermined standard to be learned from the peaks of multiple peptides contained in the mass spectrum 21 in Fig. 1 (Fig. 4: S13). In this step, first, the peak intensity value obtained from the mass spectrometer 2 is normalized by a predetermined method. The normalization method may be any method that can represent the expression level of glycoproteins, and is preferably, for example, an internal standard method that uses a ratio to the peak intensity of a predetermined internal standard, or a method that uses a ratio to the glycopeptide peak intensity contained in a control sample obtained by mixing the sera of multiple cancer patients or healthy individuals.
[0043] Furthermore, when creating a model in S1 of FIG. 3, the peptides used in the calculation can be selected by analyzing the same sample multiple times in advance with the mass spectrometer 2 of FIG. 1, and selecting the peak intensities of peptides with high reproducibility of the analysis results (for example, those with a coefficient of variation below a certain level). By targeting peptides with high reproducibility, it is possible to create a more accurate model, which is preferable. For example, it is possible to select those with a coefficient of variation (CV) value of the peak intensities of peptides measured multiple times that is below a certain threshold, and the selected peaks can also be selected using their intensity values.
[0044] The peak selection unit 341 stores the list of selected peptides in the storage device 32 as a list of peptides to be selected in the judgment support process.
[0045] In addition, the preprocessing unit 342 of the learning device 3 categorizes the peptide peaks using a predetermined method and rearranges them in an order in which those with similar fluctuation patterns are arranged in close proximity (FIG. 4: S14). In this step, the categorization is performed using the relative abundance of each peptide, which is expressed as the ratio of the peak intensity of the peptide of the cancer patient to the peak intensity of the corresponding peptide of the control sample. For example, principal component analysis may be performed and the peptides may be rearranged based on the values of the first principal component (PC1) and the second principal component (PC2), or the peptides may be categorized using a classification method such as cluster analysis using k-means, Euclidean distance, Mahalanobis distance, etc., or factor analysis, and rearranged based on the degree of similarity. In other words, the similarity of the fluctuation patterns means that the characteristics of the change in abundance (the degree of increase or decrease of the peak) between the control sample and the patient sample are similar.
[0046] Principal component analysis is a method for reducing the dimensionality of data by using synthetic variables (principal components) that have low correlation and large overall variation from among many correlated variables. The first principal component is set to maximize the variance of the data, and the second and third principal components are selected to maximize the variance under the constraint that they are orthogonal to the principal components previously determined.
[0047] The preprocessing unit 342 also converts the data of the rearranged peptides into a predetermined image data. Although the conversion method is not particularly specified, for example, a method is conceivable in which the maximum value of the peak intensity is black, the minimum value of the peak intensity is white, and the intermediate value is converted into gray with a stepwise difference in density according to the intensity ratio. Then, an image is generated in which rectangular regions painted in colors according to the peak intensity are arranged vertically and horizontally. As described above, the type and density of the color of each region can be determined based on, for example, the range of values to which the peak intensity belongs. By generating an image, for example, it becomes possible to use an existing machine learning system that has already learned the characteristics of the image. In addition, it is also possible to create an image by using multiple colors and their shades according to the amount of information of the analysis results to be used.
[0048] Biomarkers other than peptide data may be used in combination. For example, when the present invention is used to determine ovarian cancer, CA125 and HE4, which are generally known as biomarkers for assisting in the determination of ovarian cancer, may be used in combination. In that case, in addition to the above, one of the three primary colors may be selected for each biomarker, and the color may be converted into a shade according to the concentration range of the biomarker. For example, the concentration range of CA125 obtained by immunoassay is quantized, and converted into 256 shades of red by assigning a shade of color according to the magnitude of the concentration. Then, for example, the converted red color is added to the entire image created from the peptide data. In addition, when HE4 is combined, the concentration range of HE4 is converted into 256 shades of green as a color that is further different from that of HE4. Then, for example, the converted green color is added to the entire image created from the peptide data. Note that image data created by rearranging peptide data is created in blue, for example, and one image is created by mixing the three colors of red, green, and blue (RGB).
[0049] In addition, for types of biomarkers other than peptide data, such as tumor markers such as CA125 and HE4, they may be embedded in the image as one rectangular region without being colored differently from peptide data. The positions at which other types of biomarkers are embedded may be determined in a predetermined position, and they may be rearranged based on a predetermined rule, as with peptide data. In addition, for other types of biomarkers, for example, when a predetermined type of biomarker is obtained by analysis using a mass spectrometer, it is sufficient that the other types of biomarkers are glycopeptides detectable by mass spectrometry, and are not limited to a specific substance and can be appropriately selected and used by those skilled in the art. Depending on the type, the biomarkers may be fragmented in some way, principal component analysis may be performed, and an image may be created in which the rectangular regions are rearranged based on the results of the principal component analysis. In this case, an image is created by assigning different primary colors to each type of biomarker.
[0050] The pre-processing unit 342 stores information indicating the order of the peptides in the storage device 32 for use in the judgment support process described below.
[0051] Then, the deep learning unit 343 of the learning device 3 performs machine learning of the relationship between the peak intensity of the rearranged peptides and the presence or absence of a predetermined cancer by deep learning using a predetermined CNN, and creates a classifier (FIG. 4: S15). Deep learning is a type of machine learning method, and uses a multi-layered CNN. CNN can be suitably used for image recognition. Note that a program for performing machine learning can be created using an existing programming language such as MATLAB (registered trademark) or Python. In addition, an existing CNN such as AlexNet or VGG16 may be used, and parameters for cancer determination support may be optimized by performing transfer learning based on any existing classifier.
[0052] Furthermore, the model verification unit 344 of the learning device 3 uses the created trained model to determine the presence or absence of cancer in the test specimen and verifies the accuracy (FIG. 4: S16). In this step, for example, a receiver operating characteristic (ROC) analysis is performed, and if a predetermined criterion is met, it is determined that the determination accuracy of the created trained model is sufficient. Through the above-described learning process, a model for supporting the determination of the presence or absence of cancer from the characteristics of glycoproteins and the like contained in blood can be generated.
[0053] <Judgment support device> Fig. 5 is a functional block diagram showing an example of a judgment support device. The judgment support device 3 is also a computer, and includes a communication I / F 31, a storage device 32, an input / output device 33, and a processor 34, and these components are connected via a bus 35. The components corresponding to those in the learning device 3 shown in Fig. 2 are denoted by the same reference numerals, and the description thereof will be omitted.
[0054] 5 also shows functional blocks in the processor 34. That is, the processor 34 executes the program according to this embodiment to function as a peak selection unit 345, a preprocessing unit 346, and a judgment support unit 347. In the judgment support process, for example, proteins are fragmented and mass analyzed for blood of a subject whose cancer incidence is unknown, and the obtained mass spectrum 21 is input to the judgment support device 3.
[0055] The peak selection unit 345 extracts the same peptide peak as that selected in S13 of the learning process from the peaks of multiple peptides included in the mass spectrum 21, and measures its intensity. When creating a model, in addition to the peak intensity of the peptide being equal to or higher than a predetermined standard, for example, peptides with high reproducibility, such as peptides with a coefficient of variation equal to or lower than a predetermined threshold, may be extracted. It is assumed that a list of peptides to be selected is stored in advance in the storage device 32. The preprocessing unit 346 rearranges the peptide peaks according to the similarity of the fluctuations in the same order as the rearrangement in S14 of the learning process. It is assumed that information indicating the order of the peptides is stored in advance in the storage device 32. The judgment support unit 347 inputs information representing the rearranged peptide peaks to the trained deep learning model created in S15 of the learning process, and outputs information indicating the degree of possibility of being affected by cancer.
[0056] FIG. 6 is a process flow diagram showing an example of the judgment support process. In the judgment support process, for example, blood from a subject whose cancer incidence is unknown is used as a sample, and as in the learning process, it is preferable to separate serum and plasma to use as the sample. Also, as in the learning process, proteins are extracted from the sample and reductively alkylated to fragment them into peptides (FIG. 6: S21). This step is the same as S11 in FIG. 4. Furthermore, the peptides after the decomposition are analyzed by the mass spectrometer 2 to obtain a mass spectrum 21 (FIG. 6: S22). This step is the same as S12 in FIG. 4.
[0057] Then, the peak selection unit 345 of the judgment support device 3 extracts the same peptide peaks as those selected in S13 of the learning process (FIG. 6: S23). It is assumed that a list of peptides to be selected is stored in advance in the storage device 32. The list can be represented by the retention time of liquid chromatography and the ion mass-to-charge ratio (m / z) obtained from the mass spectrometer.
[0058] Furthermore, the pre-processing unit 346 of the judgment support device 3 rearranges the peptide peaks in the same order as rearranged in S14 of the learning process, and images them (FIG. 6: S24). Note that information indicating the peptide order is also assumed to be stored in advance in the storage device 32. For example, information indicating the peptide order can be represented by associating each peptide included in the above-mentioned list with coordinates on a two-dimensional code, for example. Also, in this step, similar to S14 of the learning process, the rearranged peptide peaks are converted into predetermined image data. The method of creating the image is similar to S14 of FIG. 4.
[0059] Then, the judgment support unit 347 of the judgment support device 3 inputs information on the peptide peaks rearranged in S24 into the model created in S15 of the learning process, and outputs information indicating the degree of possibility of having cancer (FIG. 6: S25). In this step, for example, image data created in S24 is input into the trained model of the neural network in which parameters such as weighting between neurons have been adjusted in S15 of the learning process. Also, information indicating the degree of possibility of having cancer is output to the user via the input / output device 33, for example, a numerical value close to 1 if the possibility of having cancer is high, and a numerical value close to 0 if the possibility is low.
[0060] <Effects> According to the above-mentioned learning device and judgment support device, it is possible to output information for supporting the judgment of the presence or absence of cancer based on the characteristics of proteins contained in blood. It is known that the sugar chain structure of glycoproteins contained in serum changes with carcinogenesis, and the above-mentioned learning device and judgment support device can be suitably used for, for example, ovarian cancer judgment support. As one of the embodiments shown here, it was an unexpected effect that ovarian cancer could be judged using a substance that is not generally used as a tumor marker, or a substance that has not been reported to be a tumor marker. In addition, information according to the analysis results of other types of biomarkers may be further included in the image. For example, the accuracy of judgment can be improved by using the analysis results of general tumor markers.
[0061] Moreover, instead of using a specific cancer marker, information representing peptide peaks is selected according to a predetermined criterion, rearranged, and then machine-learned, thereby enabling accurate learning and determination of cancer characteristics. In particular, by selecting peptides to be used for learning and determination support based on the relative abundance represented by the ratio of peak intensities obtained from a cancer patient's specimen to those obtained from a control specimen in which serum from multiple healthy people is mixed, accurate learning and determination support of cancer characteristics can be achieved. In addition, by classifying peptides into similar ones by, for example, principal component analysis, and rearranging them based on the classification results, efficient learning and determination support of cancer characteristics can be achieved.
[0062] In another embodiment of the present invention, a combination of multiple biomarkers may be used, or medical information not based on analysis results may be used. For example, the medical information may be appropriately selected and used from among age, menopausal status, lifestyle habits, interview results, doctor's findings, medical record information such as nursing records, prescriptions, and hospital visit history, medical receipt information, and health check information.
[0063] When using this information, it is necessary to digitize it. The means can be appropriately constructed according to the classification criteria and digitization criteria according to the intended judgment support, etc., and can be used in combination with the analysis results. For example, age may be used as it is, or it can be redigitized and used according to a predetermined criterion. When using receipt information and examination information, the presence or absence of drugs can be used as an indicator, and in the case of lifestyle habits such as the presence or absence of menopause and the presence or absence of smoking, it can be digitized as 0 or 1. Also, the digitization can be set according to the progress of the disease for the interview results and the doctor's findings. The course and results of treatment may be digitized in accordance with the RECIST guideline (Response Evaluation Criteria in Solid Tumors), which is a new guideline for the determination of the treatment effect of solid cancer. When other test items are cited and used from the medical record information, they can be digitized in the same way.
Example
[0064] <Measurement conditions> Using a liquid chromatograph (Agilent HP1200, manufactured by Agilent technologies) and a mass spectrometer (Q-TOF 6520, manufactured by Agilent technologies), the glycopeptides obtained from the serum sample were measured under the following conditions. The column of the liquid chromatograph used was Inertsil ODS4 (inner diameter 1.5 mm, length 100 mm, particle size 2 μm). As the eluent, solution A: 0.1% formic acid aqueous solution and solution B: 0.1% formic acid, 90% acetonitrile aqueous solution were used. The eluent was linearly changed from 10% to 56% of solution B over 40 minutes, and then the ratio of solution B was maintained at 56% for another 10 minutes. The column oven temperature was 40 °C and the flow rate was 0.1 ml / min. Mass spectrometry was performed in negative mode, and the measurement was carried out at a capillary voltage of 4000 V, a nebulizer gas flow rate of 45 psi, and a dry gas flow rate of 10 L / min (350 °C). The collision energy of the MSMS measurement using the mass spectrometer for the purpose of peptide identification was optimized between 20 eV and 70 eV according to each peptide.
[0065] <ROC analysis> The AUC (Area Under the Curve) value was calculated as follows. The samples to be compared were divided into, for example, two groups (Group A (healthy subjects, non-cancer patients in this embodiment) and Group B (patient group, ovarian cancer patients in this embodiment)), and the cutoff (threshold) of the marker to be used for calculating the AUC value was changed from 0 to infinity. The sensitivity (positive rate of ovarian cancer patients) and 1-specificity (negative rate of non-cancer patients) were plotted to create an ROC curve. The ROC curve is drawn in a square of 1 vertical x 1 horizontal, and when sensitivity = 1 and specificity = 1 (i.e., when the ovarian cancer patient group can be completely distinguished from non-cancer patients), it is a line passing through the upper left vertex. The AUC (Area Under Curve) value is the area of the lower right part of the square divided by the ROC curve (when sensitivity = 1 and specificity = 1, AUC is 1).
[0066] <Example 1> Figure 7 shows the breakdown and uses of serum samples. Serum samples obtained with the consent of patients were classified into the following groups. Group 1: Healthy subjects (Non-EOC) 254 people Group 2: Stage 1 ovarian cancer group (EOC Stage I) 97 patients Of group 1, 152 were used for training (Non-EOC Training) and 102 were used for validation (Non-EOC Test). Also, of group 2, 58 were used for training (EOC Training) and 39 were used for validation (EOC Test).
[0067] Next, 80 μL of acetone containing 10% trichloroacetic acid was added to 20 μL of serum from each patient, and the mixture was centrifuged at 12,000 rpm for 20 minutes at 4°C in a centrifuge (Himacc CT1, Hitachi Koki) to precipitate proteins. After removing the supernatant, 200 μL of a denaturing agent containing urea (0.4 g of urea, 500 μL of 1M Tris-HCl buffer (pH 8.5), 50 μL of 0.1M EDTA aqueous solution, 20 μL of 1M TCEP aqueous solution, and 190 μL of water) was added to the precipitate to denature the proteins, which were then subjected to reductive alkylation with 45 mg of iodoacetamide. After removing the denaturant and reducing agent, trypsin was added to fragment the protein into peptides, and the peptide fragments were analyzed using a liquid chromatography (Agilent HP1200, Agilent Technologies)-mass spectrometry (Q-TOF 6520, Agilent Technologies) (also referred to as "LC-MS") under the above-mentioned conditions to analyze the structure of the glycopeptide fragments contained in each serum. FIG. 8 is a diagram showing an example of the peak intensity of the measured glycopeptide fragments. The table shown in FIG. 8 includes the identification information and peak intensity of the glycopeptide. Then, 1712 glycopeptide peaks were calculated and standardized by setting the peak of the control sample (mixed serum of 10 ovarian cancer patients) to 1000.
[0068] Next, the peaks of glycopeptides detected by mass spectrometry were imaged. First, the same sample was repeatedly measured, and samples with a CV value of 30% or less and a peak intensity of 1000 or more were extracted. In addition, the peak intensity ratio between the peak intensity of 1712 glycopeptides contained in the serum of a cancer patient and the peak intensity of a control sample in which the serum of multiple healthy people was mixed was calculated, and the relative abundance of each glycopeptide in each patient was calculated. Next, principal component analysis was performed on the relative abundance of the 1712 glycopeptides, and the loading values of principal component 1 and principal component 2 were obtained. Figure 9 shows the results of the principal component analysis.
[0069] The 1712 glycopeptides were sorted in descending order of principal component 1, and classified into 42 groups, with 42 glycopeptides per group. The peak intensities of the glycopeptides in each group were sorted in descending order of principal component 2. The 42 groups classified according to principal component 1 were assigned to columns 1 to 42 of the spreadsheet software, the peak intensities of the glycopeptides sorted by principal component 2 were assigned to rows 1 to 42 of the spreadsheet software, and each cell of the spreadsheet software was colored according to the peak intensity. Specifically, the range of the standardized peak intensity was divided into 14 intervals, and each interval was assigned a grayscale color with 14 gradations of density. FIG. 10 shows an example of the created image.
[0070] <Example 2> The images created in Example 1 were trained using "CNN (Alexnet) without pre-training." The "CNN (Alexnet) without pre-training" is configured using only the Alexnet framework, with no pre-training (i.e., parameter optimization). Deep learning was performed using training samples (58 EOC subjects and 152 Non-EOC subjects). In addition, the created model was used to judge validation samples (39 EOC subjects and 102 Non-EOC subjects). As a result, the AUC was 0.853, proving the usefulness of the present invention.
[0071] <Improvement of decision support system> In order to further improve the judgment accuracy of the present invention, the learning model used for deep learning was changed to a model with pre-learning and improvement was attempted.
[0072] <Example 3> The images created in Example 1 were then trained using a pre-trained convolutional neural network to try to determine ovarian cancer. The pre-training was to classify 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into 1000 different classes, including animals, flowers, and food. Of the 25 layers of Alexnet that had been trained on these, 23 layers were used as is, and the remaining three layers were initialized. Then, training (parameter optimization) using the images created in Example 1 was performed on only the remaining three layers. Deep learning was performed using training samples (58 EOC and 152 Non-EOC), and the created model was used to determine validation samples (39 EOC and 102 Non-EOC). As a result, an AUC value of 0.881 was obtained.
[0073] <Example 4> In addition to the information on glycopeptides detected by mass spectrometry, CA125 and HE4, known ovarian cancer markers, were combined and used to support the diagnosis. The samples used, the target patient groups, the mass spectrometry measurements, the principal component analysis, the deep learning method, and the procedure up to the diagnosis were the same as in Example 1. The pre-learning model used in the deep learning was the same as that used in Example 3.
[0074] The concentrations of ovarian cancer markers CA125 and HE4 were measured using the CLIA method (serum test by LSI Medience). The concentration range of CA125 was made to correspond to each of 256 shades of red, and the measured values were converted into colors. Similarly, the concentration range of HE4 was made to correspond to each of 256 shades of green, and the measured values were converted into colors. The created image data was subjected to deep learning using CNN. The method of deep learning and the procedure up to the judgment were performed in the same manner as in Example 3. As a result, the ROC was 0.954, and the diagnostic performance was significantly improved compared to Example 3, which did not include information on CA125 and HE4 (Figure 12, p<10 -7 12. The bar graphs shown in FIG. 12 represent the evaluations of Examples 1 to 4 from the left.
[0075] That is, the present invention has realized a highly accurate determination of the presence or absence of cancer. Ovarian cancer is considered to be a cancer that is difficult to detect because there are no specific symptoms until it reaches near stage III, and various symptoms only appear when it spreads beyond the ovaries. All subjects who were the subjects of determination in the above embodiment were patients with stage I ovarian cancer. As described above, ovarian cancer is considered to be a cancer that is difficult to detect, and in particular, the frequency of ovarian cancer being discovered in asymptomatic women in mass screening is said to be 1 in 10,000 (0.01%). However, since the present invention was able to determine stage I cancer patients with high accuracy, it can be said that sufficient determination accuracy has been achieved in clinical practice.
[0076] From the above, it has been demonstrated that the present invention provides a diagnosis support system with sufficient diagnosis accuracy in clinical practice.
[0077] <Other> The above-described embodiment and modifications are merely examples, and the present invention is not limited to the above-described configurations. Furthermore, the contents described in the embodiment and modifications can be combined as much as possible without departing from the objectives and technical ideas of the present invention.
[0078] The present invention also includes a method and a computer program for executing the above-mentioned processing, and a computer-readable recording medium having the program recorded thereon. The recording medium having the program recorded thereon enables the above-mentioned processing by causing a computer to execute the program.
[0079] Here, a computer-readable recording medium refers to a recording medium that stores information such as data and programs through electrical, magnetic, optical, mechanical, or chemical action and can be read by a computer. Among such recording media, those that can be removed from a computer include flexible disks, magneto-optical disks, optical disks, magnetic tapes, memory cards, etc. Furthermore, recording media that are fixed to a computer include HDDs, SSDs (Solid State Drives), ROMs, etc. [Explanation of symbols]
[0080] 1. Blood (serum) 11 Proteins (glycoproteins) 12 Peptides (glycopeptides) 2 Mass spectrometer 21 Mass Spectrum 22 Images (peptide expression patterns after sorting) 23 Trained Model (Classifier) 3. Learning and Judgment Support Device 341, 345 Peak Selection Section 342, 346 Pretreatment section 343 Deep Learning Department 344 Model Verification Department 345 Judgment Support Department
Claims
1. a pre-processing unit that rearranges analysis results obtained by analyzing a predetermined type of biomarker in a biological sample derived from a subject into a predetermined order and converts the rearranged analysis results into an image; a determination support unit that outputs information indicating the possibility that the subject is affected by the disease, using a trained model that has learned by deep learning the relationship between the image and the analysis results of the predetermined type of biomarkers in biological samples derived from subjects with a specific disease and subjects without the disease, which are rearranged in the predetermined order and then converted into an image, and the disease state; and Equipped with the disease is cancer, the predetermined type of biomarkers includes a plurality of glycopeptides produced by fragmenting a glycoprotein with a protease; The analysis results are expressed using peak intensities of the selected glycopeptides by analyzing the glycopeptides multiple times using a mass spectrometer and selecting glycopeptides having reproducibility equal to or higher than a predetermined standard. Morbidity determination support device.
2. The type of the biomarker is any one of a protein, a nucleic acid, a peptide, a sugar chain, and a lipid, or a combination of two or more of these. The disease diagnosis support device according to claim 1.
3. The subject includes a cancer patient and a non-cancer patient, The predetermined order is determined based on a relative abundance profile of glycopeptides obtained from the blood of the cancer patient relative to glycopeptides obtained from a predetermined control sample. The disease diagnosis support device according to claim 1 or 2.
4. The predetermined order is determined based on the degree of similarity in the relative abundance of a plurality of glycopeptides, which is determined by principal component analysis, cluster analysis, or factor analysis. The disease diagnosis support device according to claim 3.
5. The preprocessing unit creates the image that two-dimensionally represents information indicating the magnitude of the peak intensities such that the peak intensities of glycopeptides with high similarity are arranged in close proximity to each other. The disease diagnosis support device according to claim 4.
6. The image is generated by determining the color of an area corresponding to each of the plurality of glycopeptides according to a numerical value of an analysis result. The disease diagnosis support device according to any one of claims 1 to 5.
7. The pre-processing unit determines a value based on an analysis result of a tumor marker different from the predetermined type of biomarker. converting the three primary colors into a color different from the color assigned to the predetermined type of biomarker and adding it to the image; or and adding the predetermined type of biomarker to the image in a different color from the other area. The judgment support unit outputs information indicating the possibility that the subject is affected by the disease using a trained model that has learned by deep learning the relationship between an image created based on the analysis results of the predetermined type of biomarker and the tumor marker and a state of the disease. The disease diagnosis support device according to claim 6.
8. The deep learning is performed using a convolutional neural network. The disease diagnosis support device according to any one of claims 1 to 7.
9. The computer rearrange the analysis results obtained by analyzing a predetermined type of biomarker in a biological sample derived from a subject into a predetermined order, and convert the rearranged analysis results into an image; outputting information indicating the possibility that the subject is suffering from the disease, using a trained model that has learned by deep learning the relationship between the image and the analysis results of the predetermined type of biomarkers in biological samples derived from subjects with a specific disease and subjects without the disease, which are rearranged in the predetermined order and then converted into an image, and the disease state; the disease is cancer, the predetermined type of biomarkers includes a plurality of glycopeptides produced by fragmenting a glycoprotein with a protease; The analysis results are expressed using peak intensities of the selected glycopeptides by analyzing the glycopeptides multiple times using a mass spectrometer and selecting the glycopeptides having reproducibility equal to or higher than a predetermined standard. Morbidity determination support method.
10. rearrange the analysis results obtained by analyzing a predetermined type of biomarker in a biological sample derived from a subject into a predetermined order, and convert the rearranged analysis results into an image; The image and analytical results of the predetermined type of biomarkers in biological samples derived from subjects with a specific disease and subjects without the disease are rearranged in the predetermined order and then converted into an image, and information indicating the possibility that the subject is affected by the disease is output using a trained model that has learned by deep learning the relationship between the image and the disease state. The process is executed by a computer, the disease is cancer, the predetermined type of biomarkers includes a plurality of glycopeptides produced by fragmenting a glycoprotein with a protease; The analysis results are obtained by analyzing the plurality of glycopeptides multiple times using a mass spectrometer, and determining whether the plurality of glycopeptides meet a predetermined standard. The glycopeptides having the reproducibility above are selected, and expressed using the peak intensity of the selected glycopeptides. Disease diagnosis support program.
Citation Information
Patent Citations
Computer assisted method for diagnosing disease
JP2007052774A
Marker for differentiation of epithelial ovarian cancer
JP2012145500A
Ovarian cancer marker and ovarian cancer detection method
JP2015125093A
Genetic information analysis system and genetic information analysis method
JP2018092515A
Method and apparatus for predicting health data value through generation of health data pattern
US20170147777A1
