Disease evaluation index calculation system, method, and program
Through the disease evaluation index calculation system, the correlation model and the underlying variable score estimation model are used to solve the problem of calculating the risk of intestinal bacterial clump disease without detailed disease conditions, and a simple disease risk assessment is achieved.
Patent Information
- Application Number
- CN202380036331.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-05-30
- Filing Date
- 2023-05-26
- Publication Date
- 2024-12-06
AI Technical Summary
The prior art cannot calculate disease evaluation indicators for intestinal bacterial clumps without requiring subjects to answer detailed disease conditions.
A disease evaluation index calculation system was developed, which calculates user-specific versus specific by inputting disease that users expect to study, storing and parsing intestinal bacterial clump data and questionnaire results data from subjects, using association models and potential variable score estimation models. Risk of disease.
Even if you do not answer the detailed disease condition, you can provide accurate intestinal bacterial cluster data to calculate disease evaluation indicators, and simply provide users with disease risk evaluation reports.
Smart Images

Figure CN119096306A_ABST
Abstract
Description
[Technical field]
[0001] The embodiment of the present invention relates to a technology for calculating a disease evaluation index using intestinal bacterial flora, and more particularly to a technology for providing an evaluation report showing the risk of a specific disease to a subject who requests bowel movement. [Background technology]
[0002] A technology for calculating disease evaluation indexes using intestinal bacterial flora of living things (such as humans) has been developed. Patent document 1 discloses a technology for providing an evaluation report showing the risk of a specific disease to a subject who takes the time to collect stool and answers a detailed questionnaire on disease conditions, etc.
[0003] [Prior art literature]
[0004] [Patent Literature]
[0005] Patent Document 1: Patent No. 6533930
[0006] [Summary of the Invention]
[0007] [Problems to be solved by the invention]
[0008] However, before calculating the disease evaluation index, it is necessary to prepare a disease risk coefficient table. The disease risk coefficient table refers to a path between a specified intestinal bacterial flora and a specified disease and health state when a specified attribute is specified.
[0009] The present invention has been completed after intensive research focusing on such a problem, and its purpose is to provide a technology for calculating a disease evaluation index using the intestinal flora data of the user even if the user does not answer detailed disease conditions.
[0010]
Methods for solving the problem
[0011] In order to solve the above-mentioned problems, the first invention is a disease evaluation index calculation system, which includes an input unit for inputting one or more diseases that a user wishes to study, a first database for storing intestinal flora related data as a result of analyzing the stool of multiple subjects, a second database for storing questionnaire result data regarding the above-mentioned diseases given to the above-mentioned subjects, an extraction unit for extracting first intestinal flora data of healthy people and second intestinal flora data of people suffering from the above-mentioned diseases using specified extraction conditions, and inputting the above-mentioned first intestinal flora data and the above-mentioned second intestinal flora data, as well as the above-mentioned questionnaire result data, to generate a first observation variable regarding one or more bacteria showing the difference between the above-mentioned first intestinal flora data and the above-mentioned second intestinal flora data, and a second observation variable indicating the condition of suffering from the above-mentioned disease. a first preparation part of a correlation model including one or more latent variables related to the first observed variable and the second observed variable, parameters from the one or more latent variables to the first and second observed variables, and / or parameters between the multiple latent variables; a second preparation part of a score estimation model using the first observed variable as an explanatory variable and using the score of the latent variable as a target variable; an estimation part of inputting the intestinal flora data of the user into the explanatory variable of the score estimation model and estimating the score of the latent variable; and a calculation part of calculating the risk of the user for the above-mentioned disease using parameters from the one or more latent variables to the second observed variable, and / or parameters between the multiple latent variables and the estimated score.
[0012] The second invention is a disease evaluation index calculation system according to the first invention, wherein the parameter from the one or more latent variables to the second observed variable is a path coefficient from the one or more latent variables to the second observed variable.
[0013] The third invention is the disease evaluation index calculation system described in the first invention, wherein the first observed variable is a bacterium having a significant difference in abundance between the first intestinal bacterial flora data and the second intestinal bacterial flora data.
[0014] A fourth invention is the disease evaluation index calculation system according to the first invention, wherein when the second observed variable is a binary categorical variable, the calculation unit uses a logistic regression model.
[0015] The fifth invention is a disease evaluation index calculation system according to the first invention, wherein a person suffering from the above-mentioned disease includes a person currently suffering from the disease and / or a person who suffered from the disease in the past.
[0016] The sixth invention is the disease evaluation index calculation system according to the first invention, further comprising a selection unit for selecting the one or more bacteria using the effect amount as an index.
[0017] The seventh invention is a disease evaluation index calculation system recorded in the first invention, wherein the second production unit extracts a measurement equation model showing the influence from the latent variable to the first observed variable from the association model, and sets the same values as the values of the parameters of the measurement equation model as the values of the parameters of the score estimation model.
[0018] The eighth invention is the disease evaluation index calculation system described in the seventh invention, wherein the second creation unit further comprises a learning unit for learning the values of the parameters of the score estimation model using the first intestinal bacterial flora data and the second intestinal bacterial flora data.
[0019] The 9th invention is a disease evaluation index calculation system recorded in the 1st invention wherein, when there are multiple diseases, the first creating unit creates multiple association models for each of the diseases, the second creating unit creates multiple score estimation models for each of the diseases, and the calculation unit calculates the risk of each of the diseases, or the risk of the multiple diseases.
[0020] The tenth invention is a disease evaluation index calculation system recorded in the first invention, which is a system in which, when there are multiple diseases, the second observed variables regarding the multiple diseases become multiple, the first creation unit adds other latent variables regarding the multiple second observed variables to the association model, the calculation unit estimates the scores of the other latent variables from the estimated scores, and uses the parameters from the other latent variables to the second observed variables and the scores of the other latent variables to calculate the risk of the user for the multiple diseases.
[0021] The 11th invention is a method for calculating disease evaluation indicators using a computer, wherein the computer accepts input of one or more diseases that a user wishes to study, uses a first database storing intestinal flora-related data as a result of analyzing the stool of multiple subjects, a second database storing questionnaire result data regarding the above-mentioned diseases given to the above-mentioned subjects, and extracts first intestinal flora data of healthy people and second intestinal flora data of people suffering from the above-mentioned diseases under specified extraction conditions, inputs the above-mentioned first intestinal flora data and the above-mentioned second intestinal flora data, as well as the above-mentioned questionnaire result data, and creates a first observation variable regarding one or more bacteria showing the difference between the above-mentioned first intestinal flora data and the above-mentioned second intestinal flora data, and a first observation variable indicating the condition of suffering from the above-mentioned disease. A second observed variable of the above-mentioned situation, one or more latent variables related to the above-mentioned first observed variable and the above-mentioned second observed variable, and a correlation model composed of parameters from the above-mentioned one or more latent variables to the above-mentioned first and second observed variables, and / or parameters between the above-mentioned multiple latent variables, using the above-mentioned first observed variable as an explanatory variable, to prepare a score estimation model with the score of the above-mentioned latent variable as the target variable, input the intestinal flora data of the above-mentioned user into the above-mentioned explanatory variable of the above-mentioned score estimation model, estimate the score of the above-mentioned latent variable, and use the parameters from the above-mentioned one or more latent variables to the above-mentioned second observed variable, and / or parameters between the above-mentioned multiple latent variables and the above-mentioned estimated score to calculate the risk of the above-mentioned user for the above-mentioned disease.
[0022] The 12th invention is a disease evaluation index calculation program that causes a computer to execute the following steps: an input step of accepting input of one or more diseases that a user wishes to study, using a first database storing data on intestinal flora as a result of analyzing the stool of a plurality of subjects, a second database storing questionnaire result data on the above-mentioned diseases given to the above-mentioned subjects, an extraction step of extracting first intestinal flora data of healthy people and second intestinal flora data of people suffering from the above-mentioned diseases under specified extraction conditions, inputting the above-mentioned first intestinal flora data and the above-mentioned second intestinal flora data, as well as the above-mentioned questionnaire result data, a first observation variable regarding one or more bacteria showing the difference between the above-mentioned first intestinal flora data and the above-mentioned second intestinal flora data, and a second observation variable indicating the condition of suffering from the above-mentioned diseases , one or more latent variables related to the above-mentioned first observed variable and the above-mentioned second observed variable, and a first preparation step of preparing a correlation model composed of parameters between the above-mentioned first and second observed variables and / or parameters between the above-mentioned multiple latent variables from the above-mentioned one or more latent variables, a second preparation step of preparing a score estimation model with the score of the above-mentioned latent variable as the target variable using the above-mentioned first observed variable as an explanatory variable, an estimation step of inputting the intestinal flora data of the above-mentioned user into the above-mentioned explanatory variable of the above-mentioned score estimation model to estimate the score of the above-mentioned latent variable, and a calculation step of calculating the risk of the above-mentioned user for the above-mentioned disease using parameters from the above-mentioned one or more latent variables to the above-mentioned second observed variable and / or parameters between the above-mentioned multiple latent variables and the above-mentioned estimated score.
[0023] [Effects of the Invention]
[0024] According to the present invention, even if the user does not answer detailed disease conditions, a technology for calculating a disease evaluation index using the user's intestinal bacterial flora data can be provided.
[0025] [Brief description of the attached drawings]
[0026]
Figure 1
[0027]
Figure 2
[0028]
Figure 3
[0029]
Figure 4
[0030]
Figure 5
[0031]
Figure 6
[0032]
Figure 7
[0033]
Figure 8
[0034]
Fig. 9
[0035]
Fig.10
[0036]
Fig.11
[0037]
Fig.12
[0038] Embodiments of the present invention will be described with reference to the accompanying drawings. In addition, the same reference numerals are given to common parts in each figure, and repeated descriptions are omitted.
[0039] (Disease Evaluation Index Calculation System)
[0040] Figure 1 This is an overall schematic diagram of a disease evaluation index calculation system according to an embodiment of the present invention. The disease evaluation index calculation system includes a stage of creating an intestinal bacterial flora database (DB) from stool sampling kits submitted by a plurality of subjects, a stage of creating a model for one or more diseases (specific diseases), and a stage of calculating an evaluation index for the risk of a specific disease.
[0041] (Intestinal bacterial flora DB preparation stage)
[0042] In the intestinal bacterial flora DB preparation stage, a large number of subjects, more than 10,000 people, excrete their stool in the toilet and collect it using a stool collection kit. Next, the extraction company that receives the stool collection kit inputs the stool collection kit into the intestinal bacterial DNA extraction device 100, and outputs a DNA solution related to the intestinal bacterial flora.
[0043] The analysis provider who receives the DNA solution inputs the DNA solution into the intestinal bacterial flora analysis device 200 to analyze the intestinal bacterial flora. The analysis provider stores the analysis result data in the intestinal bacterial flora DB 300. The analysis first generates an ASV (Amplicon Sequence Variants) sequence unique to the intestinal bacterial flora of each subject. Next, the intestinal bacterial flora data of each subject is assigned a genus name using a publicly available external DNA sequence database (e.g., the SILVA rRNA database).
[0044] The intestinal bacterial flora DB 300 is associated with the ID (identification number) information of each subject and the intestinal bacterial flora data of the subject. In addition, although the description is separated into extraction companies and analysis companies, the same company may perform extraction and analysis operations.
[0045] Next, the majority of subjects who submitted the stool collection kit enter the questionnaire using their ID information and submit it to the questionnaire collection company. The questionnaire collection company stores the questionnaire results in the questionnaire DB 400 .
[0046] The questionnaire survey included detailed information on the subjects' attributes (gender and age), body mass index (BMI), bowel movement frequency, lifestyle (frequency of drinking, smoking, and exercise), disease status, sleep status, depression self-rating scale (CES-D), whether or not Helicobacter pylori treatment was used, hospitalization or surgery experience, and the use of prescription or commercially available drugs. In addition, for female subjects only, the subjects were asked about their menstrual status and whether they were pregnant or breastfeeding. Disease status was evaluated for 15 categories (atopic dermatitis, bone and / or joint disease, bronchial asthma, diabetes, dyslipidemia, gastrointestinal disease, heart disease, hypertension, kidney disease, liver disease, lower back and / or joint pain, and the other diseases). Others) and the name of the disease and whether they are currently being treated were investigated. That is, people with diseases include those who are currently suffering and / or those who have suffered in the past. Taking prescription drugs and commercial drugs For 16 categories of drugs (gastric / duodenal ulcer / reflux esophagitis, hypertension treatment drugs, hyperlipidemia treatment drugs, diabetes treatment drugs, sleep drugs, analgesics / antipyretics, allergy treatment drugs, angina treatment drugs, laxatives / constipation drugs, osteoporosis treatment drugs, rheumatoid treatment drugs, adrenal corticosteroids, antibiotics, cold medicines, antithrombotic drugs, and others), whether they were taking them at the time of stool collection.
[0047] (Model making stage)
[0048] In the model creation stage, first, the model creator selects the disease to be evaluated. The correlation model creation device 500 includes an input unit for receiving the input of the selected disease, an extraction unit for extracting the ID information of the subject related to the disease from the attributes of each man and woman in the questionnaire DB 400, and an extraction unit for extracting the intestinal flora data of the corresponding ID information from the intestinal flora DB 300, and a creation unit for creating a correlation model by a creation method described later. The model creation device 500 stores the created correlation model in the correlation model DB 600.
[0049] Next, the model generator extracts a correlation model from the correlation model DB 600 and generates a score estimation model by a generation method described later. The generated score estimation model is stored in the score estimation model DB 800 .
[0050] (Disease evaluation index calculation stage)
[0051] In the disease evaluation index calculation stage, the evaluation provider submits stool collected with a stool collection kit, and for those who wish to evaluate the risk of a specific disease (hereinafter referred to as the user), the user is asked to submit the stool collection kit described in the intestinal bacterial flora DB preparation stage. The user's stool collection kit is processed by the intestinal bacterial DNA extraction device 100 and the intestinal bacterial flora analysis device 200 in the same way as the stool collection kits of the majority of subjects mentioned above. The user's intestinal bacterial flora data is stored in the intestinal bacterial flora DB 300. Furthermore, the disease that the user wants to evaluate (specific disease) and the user's ID information (including attributes such as the user's gender) are input into the disease evaluation index calculation device 900.
[0052] The disease evaluation index calculation device 900 includes an input unit for receiving input information, an extraction unit for extracting information related to the input information, a calculation unit for calculating the evaluation index, and an output unit for outputting an evaluation report. The input unit receives input of a specific disease and the ID information of the user. The extraction unit extracts information related to the ID information of the user (the user's gender and intestinal flora data) from the external intestinal flora DB300, and extracts a score estimation model for a specific disease from the score estimation model DB800. Furthermore, the disease evaluation index calculation device 900 inputs the intestinal flora data of the user into the score estimation model for the specific disease, and calculates the evaluation index for the risk of the specific disease. Based on the calculated evaluation index, an evaluation report is output. Furthermore, although the explanation is given separately for the operators who perform each stage, the same operator can also perform all stages.
[0053] (Flowchart of disease evaluation index calculation process)
[0054] Figure 2 1 is a flowchart of the disease evaluation index calculation process involved in this embodiment. This process is divided into two processing groups. It is divided into a first processing group (S100 to S120) in which a score estimation model is pre-made for each disease, and a second processing group (S200 to S240) in which an evaluation index for the risk of the disease that the user wishes to evaluate is calculated.
[0055] The first processing group includes a step of extracting intestinal flora data of a plurality of subjects for each disease (S100), a step of creating a correlation model for each disease (S110), and a step of creating a score estimation model for each disease (S120). The second processing group includes a step of inputting a disease that a user wishes to evaluate (S200), a step of extracting intestinal flora data of the user (S210), a step of inputting the intestinal flora data of the user into a score estimation model for a specific disease, a step of estimating a score of a latent variable (S220), and a step of calculating an evaluation index of the user's risk for a specific disease (S230).
[0056] (Explanation of S100)
[0057] In the present embodiment, "atopic dermatitis" is used as an example of a disease. Hereinafter, it is simply referred to as "atopy". S100 extracts intestinal flora data of a group of healthy people and a group of people suffering from atopy from the intestinal flora DB 300. Figure 3 to Figure 6 Displays the conditions for extracting each group. These conditions also include conditions that apply only to women.
[0058] Figure 3 The table shows the conditions of the control group involved in this embodiment. The control group (Normal Control, hereinafter referred to as NC) is a group of healthy people who all meet Figure 3 Group of people with the conditions mentioned. Figure 4 The table shows the conditions of the atopic dermatitis single group according to the present embodiment. The atopic dermatitis single group (hereinafter referred to as AS) is a group that satisfies all Figure 4 A group of people with the conditions mentioned in the above. Figure 5 The table shows the conditions of the atopic dermatitis and other disease suffering group according to the present embodiment. The atopic dermatitis and other disease suffering group (Atopic dermatitis Multi, hereinafter referred to as AM) is a group that satisfies all Figure 5 A group of people with the conditions mentioned in the above. Figure 6 The table shows the conditions of the other disease groups involved in this embodiment. The other disease group (Other Disease, hereinafter referred to as OD) is a group that satisfies all Figure 6 In the present embodiment, each group is further divided into male and female. In the following, the case of using a female group is described.
[0059] (Explanation of S110)
[0060] S110 is a step of making a correlation model for atopic reactions. In this embodiment, in order to use a female group, a correlation model for atopic reactions in women is made. First, a centered log-ratio transformation (CLR transformation) is performed on the intestinal bacterial flora data extracted in S100, and the count values of each genus of the intestinal bacterial flora data are transformed into CLR values. By performing this transformation process, it becomes easy to perform statistical processing.
[0061] Next, in this embodiment, a step of selecting bacterial variables (S115) is performed. S115 is performed to search for bacterial genera that represent the difference in intestinal bacterial flora between the control group (NC) and the group suffering from atopy only (AS). In this embodiment, the effect size between NC and AS is used as an indicator to select bacterial genera. The calculation of the effect size is performed 500 times, and when the absolute value of the effect size is arranged in descending order, the calculation parameters of the effect size are set in such a way that the bacterial genera that are within the top 20 are selected by all trials.
[0062] There are three reasons for setting the groups to be selected as the bacterial variables to two groups, NC and AS. The first reason is that when there are subjects suffering from diseases other than atopic dermatitis in the atopic dermatitis group, since the genera affected by the disease are also selected, there is a possibility that the accuracy of model creation will be affected, so it is set to AS instead of AM. The second reason is that when the condition of the control group is set to not simply suffering from atopic dermatitis, the genera affected by other diseases are also selected, and there is a possibility that the genera that characteristically show the difference between the intestinal flora of the healthy state and the intestinal flora of the atopic dermatitis state cannot be selected, so NC is set as the group of healthy people. The third reason is that since there is a possibility that the use of drugs affects the intestinal flora, there is a possibility that the state of the intestinal flora characteristically shown by atopic dermatitis is disturbed in subjects using certain drugs, so OD is not used.
[0063] Figure 7 This is a diagram for explaining the selection of bacterial variables involved in this embodiment. The value of the effect size is the average value of the calculation results of 500 times. For AS, 5 bacterial genera with high positive effect sizes are selected, and for NC, 4 bacterial genera with high negative effect sizes are selected. In this way, bacterial genera with positive effect sizes are easy to migrate from the state of healthy intestinal flora to the state of atopic dermatitis, while bacterial genera with negative effect sizes are difficult to migrate.
[0064] Figure 8 This is a diagram for explaining the change in the number of candidate bacterial genera for selection according to the present embodiment. The horizontal axis represents the number of executions of the effect amount calculation. The vertical axis represents the number of bacterial genera whose executions continue to enter the top 20 when the absolute value of the effect amount is arranged in descending order.
[0065] Returning to the process of S110, the step of creating a correlation model using the bacterial genus selected in S115 will be described. In the present embodiment, a correlation model is created using the intestinal bacterial flora data obtained by CLR conversion of NC and AS and the atopy condition data.
[0066] Fig. 9It is a diagram showing the results of estimating the parameters of the association model involved in the present embodiment. Among them, as one of the association models, the structural equation model (SEM; Structural Equation Model) is used. The structural equation model represents the relationship between the observed variables and the latent variables and the relationship between the latent variables. Therefore, the association model is a statistical analysis model constructed by the observed variables confirmed to be more than a certain correlation with a specific disease and the latent variables related to the observed variables.
[0067] Association model 1 is the area surrounded by solid lines. Rectangles represent observed variables, "e" represents the residual dispersion of observed variables, ellipses represent latent variables, and the numerical values of arrows from latent variables to observed variables or other latent variables represent the standardized parameters of the association model. For example, the observed variable Atopy is an observed variable about disease atopy, and is defined as a 2-value categorical variable indicating the status of atopy. It is "1" when atopy is present, and "0" when it is not present.
[0068] When creating the correlation model, it is assumed that there are two latent variables (LatentVariable) that describe the atopy condition variable. That is, the latent variable Lv1 exerts a positive influence on atopy, and the latent variable Lv2 exerts a negative influence.
[0069] first, Figure 7 All bacterial genera displayed are assigned as observed variables of latent variable Lv1 or Lv2. This is the initial association model. Starting from this initial association model, the model is corrected. In the model correction, the observed variables are deleted until the distributed and co-dispersed columns calculated by the process of structural equation modeling have no negative components.
[0070] After that, the observed variables were deleted in such a way that the p-value (significant probability) of each parameter became less than 0.05, and the final correlation model was adopted from the constructed model with GFI (Goodness of Fit Index) and AGFI (Adjusted GFI) values close to 1, RMSEA (Root Mean Square Error of Approximation) value close to 0, and the absolute value of the path coefficient from the latent variable to the disease status variable becoming the largest. The result of model modification, Figure 7 Among the displayed bacterial genera, Erysipelatoclostridium, Oscillibacter, and Ruminococcaceae UCG-005 were excluded from the observed variables.
[0071] The data of the combined group of NC and AS (i.e., intestinal bacterial flora data and atopy status data) were given to the created correlation model, and each parameter in the correlation model was calculated. Thus, when focusing on healthy people and atopy patients, the bacterial genus set as the observation variable was studied through the assumed latent variables (Lv1, Lv2) to explain the variable representing the atopy status.
[0072] The standardized path coefficient from the latent variable Lv1 assumed to have a positive influence on atopy to the atopy status variable was 0.32 (p < 0.01), and the standardized path coefficient from the latent variable Lv2 assumed to have a negative influence on atopy to the atopy status variable was -0.41 (p < 0.01). The standardized factor loadings from the latent variable Lv1 to each observed variable of Lv1 were 0.54 for Alistipes, 0.53 for Butyricimonas, and 0.44 for Coprobacter. In addition, the standardized factor loadings from the latent variable Lv2 to each observed variable of Lv2 were 0.58 for Fusicatenibacter, 0.42 for Agathobacter, and 0.20 for Streptococcus. The standardized factor loadings from the latent variables to the observed variables were all significant at p < 0.05. The value of 0.07 of the arrow between the latent variables indicates the magnitude of the correlation between Lv1 and Lv2.
[0073] The residual variance of the atopy status variable was 0.75, and the two assumed latent variables (Lv1, Lv2) in this correlation model accounted for about 25% of the variance of the atopy status variable.
[0074] Made in Fig. 9 When the association model 1 is prepared, the groups to be the target of the selection of the bacteria variable (S115) are set to two groups, NC and AS. The latent variable scores in the prepared association model are compared between the other groups.
[0075] from Fig. 9 The latent variable scores of each subject were calculated from the parameters of the association model 1, and the scores were compared between the groups. No significant difference was found in the score of Lv1 between NC and OD (p=0.27), and AS was significantly higher than NC (p<0.01). The score of AM tended to be higher than NC, but no significant difference was found (p=0.12). No significant difference was found in the score of Lv2 between NC and OD (p=0.55), and both AS and AM were significantly lower than NC (p<0.01).
[0076] In this embodiment, S115 is an example of a method for selecting an observed variable of a correlation model, and a bacterial variable is selected using the effect size as an index. S115 is an optional configuration.
[0077] There are various methods for selecting observation variables of the association model. For example, some statistical tests (Wilcoxo's nth order sum test, etc.) can be performed between the control group and the disease-stricken group, and bacteria with significant differences in abundance between the two groups can be used as observation variables of the association model. Alternatively, bacteria that have been reported to have the potential to become biomarkers of the target disease can be used as observation variables of the association model.
[0078] (Explanation of S120)
[0079] S120 is a step of creating a model for estimating the score of a latent variable when the disease status is unknown. When creating the association model, the data of the combined group of NC and AS (i.e., intestinal bacterial flora data and atopy status data) is used. However, the user is a person who wants to evaluate the risk of a specific disease, and the disease status is unknown. Therefore, it is necessary to estimate the model of the score of the latent variable when the disease status is unknown.
[0080] When the atopy status of a female user is unknown, in order to estimate the scores of Lv1 and Lv2, the part of the determination equation model is extracted from the correlation model 1 ( Fig. 9 The dashed line ( ) is used to create the score estimation model 2. Therefore, the same values as the values of the parameters of the measurement equation model are set as the values of the parameters of the score estimation model 2. The measurement equation model is a model that shows the influence of the latent variable on the observed variable of the genus.
[0081] In this embodiment, the user's Lv1 and Lv2 scores are estimated from the score estimation model 2. The tendency that the estimated latent variable scores are significant differences in the inter-group comparison of the estimated latent variable scores is the same as that of the association model 1. In the following, the estimated values of the latent variables Lv1 and Lv2 are represented by Lv1est and Lv2est, respectively.
[0082] (Explanation of S200)
[0083] S200 is a step of inputting a disease that the user wishes to evaluate. In this embodiment, a score estimation model for atopy is prepared by gender. Therefore, the user also inputs gender. The user is a female.
[0084] (Explanation of S210)
[0085] S210 is a step of extracting the intestinal flora data of the user. The user takes out the stool collection kit, and the intestinal flora data of the user is stored in the intestinal flora DB 300. S210 extracts the intestinal flora data of the user from the intestinal flora DB 300 using the ID information of the user.
[0086] (Explanation of S220)
[0087] S220 is a step of estimating the score of the latent variable using the atopic score estimation model 2. Specifically, the intestinal flora data of the user is input to the observed variable of the atopic score estimation model 2 to estimate the user's score of the latent variable of the atopic score estimation model 2.
[0088] (Explanation of S230)
[0089] S230 is a step of calculating an evaluation index for the risk of atopic reaction. Specifically, Fig. 9 The path coefficient from the latent variables Lv1 and Lv2 to the observed variable Atopy and the estimated values (Lv1est and Lv2est) of the latent variables Lv1 and Lv2 are used to calculate the user's risk of atopic reaction. Fig. 9 The dashed line indicates the part related to the risk calculation model 3.
[0090] (Risk calculation example)
[0091] Fig.10 Calculation of the risk and the accuracy of the calculated risk according to the present embodiment will be described.
[0092] For the female subject group, the combined group of NC, OD, AS, and AM was stratified and randomly divided into 80% learning group and 20% verification group. The association model was learned using the information of the latent variable score (Lv1 or Lv2) of the learning group and the atopy status, and an atopy probability estimation model was constructed to estimate the probability of atopy from the latent variable score. The latent variable score estimated value (Lv1est or Lv2est) of the verification group was applied to the constructed atopy probability estimation model, and the accuracy of the atopy probability estimation model was studied by ROC (Receiver Operating Characteristic) analysis. The atopy probability is the risk of suffering from atopy, and the results of the ROC analysis can be used as a disease evaluation indicator.
[0093] In the method for constructing an atopy probability estimation model, a logistic regression model is used in which latent variables Lv1 and / or Lv2 are used as explanatory variables and a binary categorical variable representing the user's atopy status is used as the objective function. Fig. 9 The risk calculation model 3 is equivalent.
[0094] Fig.10 The figure shows the ROC curve for this logistic regression model. The vertical axis represents "Sensitivity". When the probability of judging a person who actually suffers from the disease increases, the value of the vertical axis becomes close to 1. The horizontal axis represents "1-Spefificity". When the probability of judging a person who does not suffer from atopy as not suffering from the disease increases, the value of specificity becomes close to 1. Since the value of the horizontal axis becomes close to 0 by subtracting specificity from 1, the probability of judging a person who does not suffer from atopy as not suffering from the disease increases. The dotted line represents the result of the model with only Lv1 as the explanatory variable, and the thick solid line represents the result of the model with only Lv2 as the explanatory variable. The dotted line represents the result of the model with Lv1 and Lv2 as the explanatory variables. AUC represents the value of Area Under the Curve, and 95% CI represents the 95% confidence interval of AUC.
[0095] exist Fig.10 In the figure, the diagonal line (thin solid line) represents a probability of 0.5. When only the Lv2 model (thick solid line) is used, it can be visually (or intuitively) judged that the area surrounded by the thick solid line and the thin solid line is wider than other models, and the accuracy of the risk estimation method is high. In addition, AUC is an indicator of the risk estimation method. When the AUC value is close to 1, it can be objectively judged as a good risk estimation method. The AUC of the model using only Lv2 shows the highest value of 0.66. In this embodiment, when a direct risk estimation method such as a blood test is excluded, it can be judged as an indirect risk estimation method with high accuracy.
[0096] (Effect)
[0097] According to this embodiment, the intestinal flora data and disease status data of a plurality of subjects are used to create a correlation model for each disease ( Fig. 9 The correlation model 1) is used to generate the score estimation model in advance from the determination equation part of the correlation model ( Fig. 9 For users who wish to evaluate the risk of a specific disease, the score estimation model for the specific disease and the intestinal flora data of the user are used to estimate the score of the latent variable. Fig. 9 The risk calculation model 3) can calculate the user's risk for a specific disease by inputting the scores of the presumed latent variables.
[0098] Furthermore, when the user collects and presents stool using the stool collection kit, the user can easily receive an evaluation report on the risk of the user suffering from a specific disease without answering a detailed questionnaire on the disease status, etc. In addition, the business operator who provides the evaluation report can design, study, and propose prebiotics to the user.
[0099] (Variation 1: A case where the values of the parameters of the score estimation model are learned)
[0100] In the above-mentioned embodiment, the measurement equation model showing the influence of the latent variable on the observed variable of the bacterial genus is extracted from the association model, and the same value as the value of each parameter of the measurement equation model is set as the value of each parameter of the score estimation model. That is, the value of each parameter of the measurement equation model is used directly in the value of each parameter of the score estimation model without learning. In contrast, in the modification 1, the value of each parameter of the score estimation model is learned using the intestinal bacterial flora data of NC and AS.
[0101] The learning is carried out in the following 6 steps. (1) For the female subject group, the combined group of NC and AS is stratified and randomly divided into 80% learning group and 20% verification group. (2) The score of the latent variable (Lv1 and Lv2) of the association model is calculated for each subject belonging to the learning group. (3) For the learning group, the calculated score of the latent variable, the bacterial flora data of the selected genus, and the data on the atopic condition are summed up to balance the total data. (4) A linear multiple regression model is prepared in which the latent variables (Lv1 and Lv2) in the balanced learning group data are set as the target variables and the bacterial flora data of the genus are set as the explanatory variables. (5) The score of the latent variable (Lv1 and Lv2) of the verification group is estimated using the linear multiple regression model. (6) The score of the latent variable (Lv1 and Lv2) of the verification group calculated from each parameter of the association model and the score of the estimated latent variable (Lv1 and Lv2) are compared.
[0102] (Variant 2: A case where a correlation model is created for each disease)
[0103] Fig.11 This is a diagram showing an overview of a plurality of relationship models according to Modification 2. In this diagram, two relationship models are used, and the numerical values of arrows and the like are omitted for the sake of illustration.
[0104] In the above-mentioned embodiment, the case where the user wishes to evaluate the risk of a specific disease is described. However, there are cases where the user wishes to evaluate the risks of multiple diseases. For example, disease a is atopy and disease b is wheezing. In such a case, an association model (1a, 1b) is made for each disease, a score estimation model (2a, 2b) is made for each disease, and a risk calculation model (3a, 3b) is also made. In this case, the disease evaluation index calculation system of this embodiment can calculate the risk of each disease, or the risk of multiple diseases.
[0105] For the risk of multiple diseases, for example, the maximum value among the risk values of each disease may be recorded in the evaluation report as the highest risk. In addition, the categorical variable representing the prevalence status of each disease is taken as 2 values and classified into 2 groups, but it is not limited to this, and classification into 3 or more groups may be performed. For example, when the user wishes to evaluate the risk of atopy and wheezing, the variables of the prevalence status of each disease of the combined association model (a, b) may be combined to classify the prevalence status of each disease as "0" when the patient does not suffer from atopy and wheezing, classify the prevalence status of the patient only suffering from atopy as "1", classify the prevalence status of the patient only suffering from wheezing as "2", and classify the prevalence status of the patient both suffering from atopy and wheezing as "3".
[0106] (Variant 3: Case of using latent variables for diseases)
[0107] Fig.12 This is a diagram showing an overview of the association model according to Modification 3. In this diagram, a latent variable using a plurality of disease afflicted condition variables as observed variables is used in the association model, and in the case of the drawing, the numerical values of the arrows and the like are omitted.
[0108] Among them, atopy and wheezing are diseases related to allergies, and heart disease and hypertension are classified as diseases related to the circulatory system. Furthermore, the latent variables related to the diseases (atomy and wheezing) are Lv related to allergies, and the latent variables related to the diseases (heart disease and hypertension) are Lv related to the circulatory system.
[0109] Fig.12 The association model 1' is to Fig. 9 The association model 1 newly adds the model of Lv for allergies and Lv for the circulatory system. Fig.12 When the association model 1' is used, it can also be Fig. 9 The score estimation model 2' and the risk calculation model 3' are prepared by the same method as described in .
[0110] When the user wishes to evaluate one or more risks of atopy, wheezing, heart disease, and hypertension, the user's intestinal flora data is input into the score estimation model 2' to estimate the scores of the latent variables (Lv1' and Lv2') regarding the genus. Next, the scores of the latent variables (Lv regarding allergies and Lv regarding the circulatory system) regarding the disease of the risk calculation model 3' are estimated. Finally, the risk of the disease that the user wishes to evaluate can be calculated using the path coefficient from the Lv regarding allergies to the observed variables (atopic reactions and wheezing), the path coefficient from the Lv regarding the circulatory system to the observed variables (heart disease and hypertension), and the scores regarding the estimated Lv regarding allergies and the Lv regarding the circulatory system.
[0111] The embodiments (including modified examples) of the present invention are described above, but two or more embodiments may be implemented in combination. Alternatively, one embodiment may be implemented partially. Furthermore, two or more embodiments may be implemented in combination.
[0112] In addition, the present invention is not limited to the description of the embodiments of the present invention in any way. Various modified embodiments that can be easily conceived by those skilled in the art are also included in the present invention without departing from the description of the patent claims.
[0113] For example, the number of latent variables is assumed to be 2, but it may be 1 or more than 3. In addition, in the present embodiment, the ID information of the subjects related to the disease is extracted from the attributes of each man and woman in the questionnaire DB 400 to create a correlation model, but the attributes of each man and woman may not be extracted according to the type of disease.
[0114]
Explanation of symbols
[0115] 100: Intestinal bacterial DNA extraction device
[0116] 200: Intestinal bacterial flora analysis device
[0117] 300: Intestinal Bacteria DB
[0118] 400: Questionnaire DB
[0119] 500: Related model making device
[0120] 600: Association Model DB
[0121] 700: Score estimation model making device
[0122] 800: Score estimation model DB
[0123] 900: Disease evaluation index calculation device
Claims
1. Disease evaluation index calculation system, which has: An input unit for inputting one or more diseases that a user wishes to study, an extracting unit for extracting first intestinal bacterial flora data of healthy persons and second intestinal bacterial flora data of persons suffering from the disease using a first database storing intestinal bacterial flora-related data as a result of analyzing stools of a plurality of subjects, a second database storing questionnaire result data regarding the disease given to the subjects, and a specified extraction condition, a first preparation unit which inputs the first intestinal bacterial flora data, the second intestinal bacterial flora data and the questionnaire result data, and prepares a first observed variable regarding one or more bacteria indicating a difference between the first intestinal bacterial flora data and the second intestinal bacterial flora data, a second observed variable indicating the condition of the disease, one or more latent variables related to the first observed variable and the second observed variable, and a parameter from the one or more latent variables to the first and second observed variables, and / or a correlation model composed of parameters between the plurality of latent variables, a second creation unit that uses the first observed variable as an explanatory variable and creates a score estimation model using the score of the latent variable as a target variable, an estimating unit that inputs the intestinal bacterial flora data of the user into the explanatory variable of the score estimation model to estimate the score of the latent variable, and A calculation unit calculates the risk of the user for the disease using parameters from the one or more latent variables to the second observed variable and / or parameters between the multiple latent variables and the estimated score.
2. The disease evaluation index calculation system of claim 1, wherein the parameter from the one or more latent variables to the second observed variable is a path coefficient from the one or more latent variables to the second observed variable.
3. The disease evaluation index calculation system according to claim 1, wherein the first observed variable is bacteria having a significant difference in abundance between the first intestinal bacterial flora data and the second intestinal bacterial flora data.
4. The disease evaluation index calculation system according to claim 1, wherein when the second observed variable is a binary categorical variable, the calculation unit uses a logistic regression model.
5. The disease evaluation index calculation system according to claim 1, wherein the people suffering from the disease include people suffering from the disease currently and / or people suffering from the disease in the past.
6. The disease evaluation index calculation system according to claim 1, wherein the first creation unit further comprises a selection unit for selecting the one or more bacteria using the effect amount as an index.
7. The disease evaluation index calculation system according to claim 1, wherein the second creation unit extracts a determination equation model showing the influence of the latent variable on the first observed variable from the correlation model, The same values as the values of the parameters of the measurement equation model are set as the values of the parameters of the score estimation model.
8. The disease evaluation index calculation system according to claim 7, wherein the second creation unit further comprises a learning unit that learns the value of each parameter of the score estimation model using the first intestinal bacterial flora data and the second intestinal bacterial flora data.
9. The disease evaluation index calculation system according to claim 1, wherein when there are multiple diseases, The first generating unit generates a plurality of the correlation models for each of the diseases, The second generating unit generates a plurality of the score estimation models for each of the diseases, The calculation unit calculates the risk for each of the diseases or the risks for the plurality of diseases.
10. The disease evaluation index calculation system according to claim 1, wherein when there are multiple diseases, the second observation variable for the disease becomes multiple, The first creation unit adds other potential variables related to the plurality of second observed variables to the correlation model, The calculation unit estimates the score of the other latent variable from the estimated score, and calculates the risk of the user for the plurality of diseases using the parameter from the other latent variable to the second observed variable and the score of the other latent variable.
11. A method for calculating disease evaluation indicators using a computer, wherein the computer: Accepts input of one or more diseases that the user wishes to study, extracting first intestinal bacterial flora data of healthy persons and second intestinal bacterial flora data of persons suffering from the disease using a first database storing data on intestinal bacterial flora as a result of analyzing stools of a plurality of subjects, a second database storing data on questionnaire results regarding the disease given to the subjects, and a specified extraction condition, The first intestinal bacterial flora data, the second intestinal bacterial flora data and the questionnaire result data are input, and a first observed variable about one or more bacteria indicating the difference between the first intestinal bacterial flora data and the second intestinal bacterial flora data, a second observed variable indicating the condition of the disease, one or more latent variables related to the first observed variable and the second observed variable, and a parameter from the one or more latent variables to the first and second observed variables, and / or a correlation model composed of parameters between the multiple latent variables is prepared. Using the first observed variable as an explanatory variable and creating a score estimation model using the score of the latent variable as the target variable, The intestinal bacterial flora data of the user is input into the explanatory variable of the score estimation model to estimate the score of the latent variable, The risk of the user for the disease is calculated using parameters from the one or more latent variables to the second observed variable and / or parameters between the plurality of latent variables and the estimated score.
12. A disease evaluation index calculation program, which enables a computer to execute the following steps: An input step, which accepts input of one or more diseases that the user wishes to study, an extraction step of extracting first intestinal bacterial flora data of healthy persons and second intestinal bacterial flora data of persons suffering from the disease using a first database storing data on intestinal bacterial flora as a result of analyzing stools of a plurality of subjects, a second database storing data on questionnaire results regarding the disease given to the subjects, and a specified extraction condition, The first preparation step comprises inputting the first intestinal bacterial flora data, the second intestinal bacterial flora data and the questionnaire result data, and preparing a first observed variable regarding one or more bacteria indicating a difference between the first intestinal bacterial flora data and the second intestinal bacterial flora data, a second observed variable indicating the condition of the disease, one or more latent variables related to the first observed variable and the second observed variable, and a parameter from the one or more latent variables to the first and second observed variables, and / or a correlation model composed of parameters between the plurality of latent variables. The second preparation step is to prepare a score estimation model using the first observed variable as an explanatory variable and the score of the latent variable as a target variable, an estimating step of inputting the intestinal bacterial flora data of the user into the explanatory variable of the score estimation model to estimate the score of the latent variable, and A calculation step, which uses parameters from the one or more latent variables to the second observed variable and / or parameters between the multiple latent variables and the estimated score to calculate the user's risk for the disease.
