KR20230129933A - Method and apparatus for risk prediction of diabetic nephropathy based on integration of polygenic and clinical information - Google Patents

Abstract

The present disclosure relates to a method for predicting the risk of developing diabetic nephropathy based on the fusion of multiple genes and clinical information, and an electronic device for performing the same. According to one embodiment, a method for predicting the risk of developing diabetic nephropathy using an electronic device includes acquiring genomic data of a clinical sample from an external device connected to the electronic device; Based on reference omics data including a plurality of preset reference SNP (Single Nucleotide Polymorphism) data and a plurality of miRNA data related to the onset of diabetic nephropathy, a plurality of target errors identical to the reference omics data are identified from the genomic data. Selecting mix data; and calculating a risk score using a genotype of each of the plurality of target omics data and a weight for each of the plurality of target omics data. It includes, and the reference omics data includes a plurality of reference SNP data related to the development of diabetic nephropathy and a plurality of mi RNA data related to the development of diabetic nephropathy, and the omics data related to the development of diabetic nephropathy The corrected effect size can be reflected using the linkage imbalance and global scaling parameter (shrinkage factor).

Description

Method and device for predicting risk of developing diabetic nephropathy based on multigene and clinical information fusion {METHOD AND APPARATUS FOR RISK PREDICTION OF DIABETIC NEPHROPATHY BASED ON INTEGRATION OF POLYGENIC AND CLINICAL INFORMATION}

The present disclosure relates to a method for intelligently predicting the risk of developing diabetic nephropathy and an electronic device for performing the same. More specifically, it relates to a method for intelligently predicting the risk of developing diabetic nephropathy based on a deep learning network and an electronic device that performs the same.

According to the 2020 announcement by the Korean Diabetes Association, the prevalence of diabetes among the adult population over 30 years of age in Korea was 13.8% in 2018, and when the estimated population was applied, it was calculated to be 4.94 million. In terms of comorbidities, 53.2% of those with diabetes were obese, 61.3% had hypertension, and 72% had hypercholesterolemia. Among those with diabetes, the prevalence of both hypertension and hypercholesterolemia was 43.7. It was %.

As can be seen from the etymology of diabetes, which means ‘sugar comes from noodles,’ it has a close relationship with the kidneys. Diabetes is the fifth most common cause of disease and death worldwide, and this can be said to be due to complications caused by diabetes rather than diabetes itself. Diabetic nephropathy, along with diabetic retinopathy and diabetic neuropathy, are major microvascular complications of diabetes, while coronary artery disease, cerebral infarction, and peripheral vascular disease are macrovascular complications of diabetes.

Diabetic nephropathy is a disease in which kidney function deteriorates due to continuous damage to the glomeruli inside the kidney due to high blood sugar levels. Diabetes causes kidney disease to progress slowly over 10 to 15 years, so if diabetes is not properly treated in the early stages, kidney function will be ruined before you know it. In severe cases, it can lead to end-stage renal failure requiring dialysis and transplantation.

Although intensive research has been conducted on the cause, diagnosis, and treatment of kidney disease over the past several years, a definitive treatment that can effectively treat diabetic nephropathy has not yet been developed.

To date, albuminuria is the most widely used biomarker for early diabetic nephropathy. However, albuminuria is not a specific marker only for diabetic nephropathy but is also observed in other kidney diseases, is filtered by the glomeruli but is secreted again in the tubules, not all diabetic patients with microalbuminuria progress to overt proteinuria, and even without albuminuria. It has limitations as a biomarker in that diabetic nephropathy may occur. For this reason, there is a need to develop technology to intelligently read whether a person to be diagnosed has diabetic nephropathy using a deep learning network so that diabetic nephropathy can be diagnosed early and accurately.

Korean Patent No. 10-18176650000

According to one embodiment, a method for predicting the risk of developing diabetic nephropathy based on the fusion of multiple genes and clinical information using an artificial intelligence model and an electronic device for performing the same may be provided.

In addition, according to one embodiment, a method for analyzing the results of an artificial intelligence model for predicting the risk of developing diabetic nephropathy based on the fusion of multigene and clinical information and an electronic device for performing the same may be provided.

According to an embodiment of the present disclosure for achieving the above-mentioned technical problem, a method of predicting the risk of developing diabetic nephropathy by an electronic device includes obtaining genomic data of a clinical sample from an external device connected to the electronic device; Based on reference omics data including a plurality of preset reference SNP (Single Nucleotide Polymorphism) data and a plurality of miRNA data related to the onset of diabetic nephropathy, a plurality of target errors identical to the reference omics data are identified from the genomic data. Selecting mix data; and calculating a risk score using a genotype of each of the plurality of target omics data and a weight for each of the plurality of target omics data. It includes, and the reference omics data includes a plurality of reference SNP data related to the development of diabetic nephropathy and a plurality of mi RNA data related to the development of diabetic nephropathy, and the omics data related to the development of diabetic nephropathy The corrected effect size can be reflected using the linkage imbalance and global scaling parameter (shrinkage factor).

According to one embodiment, the method further includes obtaining user identification information of a user corresponding to the genomic data of the clinical sample along with the genomic data of the clinical sample; Obtaining personal characteristic information of the user from the external device based on the user identification information; and identifying a user type of the user corresponding to the clinical sample based on the user's personal characteristic information. may include.

According to one embodiment, the method includes acquiring omics learning data related to the risk of developing diabetic nephropathy for learning a prediction model from an external device connected to the electronic device; subgrouping the obtained omics learning data based on preset individual user characteristics; And generating a plurality of prediction models for each user's individual characteristics by performing K-fold cross-validation on the subgrouped omics data; It may further include.

According to one embodiment, the method includes subgrouping genomic data of a clinical sample obtained from the external device based on the user's personal characteristics; Further comprising: calculating the risk score by inputting a plurality of target omics data selected from the subgrouped genomic data into a plurality of prediction models for each characteristic of the user's individual, Obtaining output values from each of the prediction models; and calculating the risk score by applying different weights applied to the output values of the plurality of prediction models to the output values of each of the plurality of prediction models, according to the identified user type. may include.

According to one embodiment, the method is performed by inputting a plurality of target omics data selected from the subgrouped genomic data and output values of each of the plurality of prediction models into a pre-learned Cox proportional hazard survival analysis model. , calculating the risk score for the clinical sample for each user type and predetermined period from the Cox proportional hazards survival analysis model; It may further include.

In addition, according to another embodiment for achieving the above-described technical problem, an electronic device for predicting the risk of developing diabetic nephropathy includes a network interface; A memory that stores one or more instructions; and at least one processor executing the one or more instructions; Includes, by executing the one or more instructions, to obtain genomic data of a clinical sample from an external device connected to the electronic device, a plurality of preset reference SNP (Single Nucleotide Polymorphism) data and related to the onset of diabetic nephropathy A plurality of target omics data identical to the reference omics data are selected from the genomic data based on reference omics data including a plurality of miRNA data, and the genotype and the plurality of each of the plurality of target omics data are selected. A risk score is calculated using a weight for each of the target omics data, and the reference omics data includes a plurality of reference SNP data related to the development of diabetic nephropathy and a plurality of miRNA data related to the development of diabetic nephropathy. It may reflect the effect size corrected using the linkage imbalance and global scaling parameter (Shrinkage factor) of the reference omics data related to the development of diabetic nephropathy.

In addition, according to another embodiment for solving the above technical problem, a method for predicting the risk of developing diabetic nephropathy using an electronic device includes: acquiring genomic data of a clinical sample from an external device connected to the electronic device; Based on reference omics data including a plurality of preset reference SNP (Single Nucleotide Polymorphism) data and a plurality of miRNA data related to the onset of diabetic nephropathy, a plurality of target errors identical to the reference omics data are identified from the genomic data. Selecting mix data; and calculating a risk score using a genotype of each of the plurality of target omics data and a weight for each of the plurality of target omics data. It includes, and the reference omics data includes a plurality of reference SNP data related to the development of diabetic nephropathy and a plurality of mi RNA data related to the development of diabetic nephropathy, and the omics data related to the development of diabetic nephropathy A computer-readable recording medium recording a program for executing a method on a computer, characterized in that it reflects the effect size corrected using the linkage imbalance and the global scaling parameter (shrinkage factor). can be provided.

Figure 1 is a diagram schematically showing the operation process of a system for evaluating the risk of diabetic nephropathy.
Figure 2 is a flowchart of a method for predicting the risk of developing diabetic nephropathy by an electronic device according to an embodiment.
FIG. 3 is a diagram illustrating a specific process of how an electronic device calculates a risk score according to an embodiment.
FIG. 4 is a diagram illustrating a process in which an electronic device generates a prediction model based on SNP data and miRNA data obtained from GWAS and TWAS databases.
Figure 5 shows the process by which an electronic device generates a prediction model based on genomic data subgrouped according to individual characteristics, and the Cox proportional hazard survival analysis model that generates the probability of disease onset by period based on the generation result of the prediction model. This is a drawing showing the process of use.
Figure 6 is a diagram showing the results of an electronic device selecting reference omics data including reference SNP data.
Figure 7 is a diagram showing an example of the ROC curve that verifies the diabetic nephropathy risk analysis model used by the electronic device.
FIG. 8 is a diagram illustrating the structure of an electronic device for analyzing the risk of diabetic nephropathy and a system including the same according to an embodiment.

Terms used in this specification will be briefly described, and the present disclosure will be described in detail.

The terms used in the present disclosure have selected general terms that are currently widely used as much as possible while considering the functions in the present disclosure, but this may vary depending on the intention or precedents of those skilled in the art, the emergence of new technologies, etc. In addition, in certain cases, there are terms arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the description of the relevant invention. Therefore, the terms used in this disclosure should be defined based on the meaning of the term and the overall content of this disclosure, rather than simply the name of the term.

When it is said that a part "includes" a certain element throughout the specification, this means that, unless specifically stated to the contrary, it does not exclude other elements but may further include other elements. In addition, terms such as "... unit" and "module" used in the specification refer to a unit that processes at least one function or operation, which may be implemented as hardware or software, or as a combination of hardware and software. .

Below, with reference to the attached drawings, embodiments of the present disclosure will be described in detail so that those skilled in the art can easily practice them. However, the present disclosure may be implemented in many different forms and is not limited to the embodiments described herein. In order to clearly explain the present disclosure in the drawings, parts that are not related to the description are omitted, and similar parts are given similar reference numerals throughout the specification.

Figure 1 is a diagram schematically showing the operation process of a system for evaluating the risk of diabetic nephropathy.

The system 10 for evaluating the risk of diabetic nephropathy according to an embodiment may include a reference database 120 and an electronic device 1000. The system 10 for evaluating the risk of diabetic nephropathy according to one embodiment uses the 10-fold LOGO (10-fold LOGO ( Leave One Group Out) meta-analysis can be performed to perform correlation analysis and integrated meta-analysis for groups 1 to 9, and a genetic risk prediction model can be built using the summary statistics of the meta-analysis results.

The genetic risk of group 10 can be estimated using the constructed prediction model model, and the genetic risk of all 10 groups can be calculated using the correlation analysis results of the independent group. In addition, the system for evaluating the risk of diabetic nephropathy according to the present disclosure (10) predicts the risk of developing diabetic nephropathy at a specific age of the subject by applying the Cox proportional hazards survival model by adding epidemiological and clinical data of clinical samples as covariates. A deep learning-based disease risk prediction model with improved accuracy can be completed.

According to one embodiment, the system 10 for evaluating the risk of diabetic nephropathy stores reference omics data including a plurality of reference SNP data related to the development of diabetic nephropathy and a plurality of miRNA data related to the development of diabetic nephropathy. A reference database 120 that predicts the risk of diabetic nephropathy, a first server 150 that predicts the risk of diabetic nephropathy, a second server 110 that performs Meta-GWAS analysis, acquires genomic data of target clinical samples, and obtains genomic data. It may include an electronic device 1000 that transmits to servers.

The first server 150 selects a plurality of target omics data or a plurality of target SNP data from the genomic data of the clinical sample based on the Meta-GWAS analysis results of the second server 110, and selects the plurality of selected target SNP data. A diabetic nephropathy risk score can be determined by inputting low-dimensional embeddings determined from target omics data or multiple target SNP data into a preset prediction model.

According to one embodiment, the first server 150 may be an analysis device or computing device that acquires sample genomic data for a specific user and determines the risk of developing diabetic nephropathy for the obtained genomic data. For example, the first server 150 may receive genomic data of a specific sample from the second server (or genomic information generating device) 110, the electronic device 1000, or a separate database. According to one embodiment, the genomic data of a specific sample may include identification information for the individual.

The first server 150 generates a diabetes risk of the sample based on the sample's genomic data. The first server 150 may generate a diabetes risk by performing GWAS on the genomic data of clinical samples. The first server 150 may provide analysis results for the sample to the service user (A). For example, the first server 150 may transmit the analysis result to the personal terminal 50.

Meanwhile, the first server 150 may receive reference SNP information for identifying SNPs related to diabetes from the reference database 120. According to another embodiment, the first server 150 provides reference omics data including a plurality of reference SNP (Single Nucleotide Polymorphism) data and a plurality of miRNA data related to the development of diabetic nephropathy from the reference database 120. You can also obtain . The reference database 120 is built in advance and stores reference SNP data and a plurality of miRNA data.

The first server 150 may select a plurality of target omics data from the genomic data of the sample using reference SNP data or reference omics data. According to another embodiment, the first server 150 may select a plurality of target SNP data from the genomic data of the sample using reference SNP data. The first server 150 may determine the risk of developing diabetic nephropathy for the genomic data of the sample using a plurality of target omics data or a plurality of target SNP data.

According to one embodiment, the plurality of reference SNP data used by the first server 150 are rs12531478-A, rs17373728-C, rs5750250-G, rs11107616-C, rs136161-G, rs4879670-G, rs13259109-G, rs1298908 - G, rs304029-G, rs9510795-A, rs10952362-C, rs4667466-T, rs10778560-C, rs7975752-G, rs731565-T, rs4849965-C, rs6910061-A, rs1424609-G, rs2596230-G, rs1677894-G, It may include one or more SNP data from the SNP data group including rs5750250-G and rs136161-G.

The second server 110 performs correlation analysis and integrated meta-analysis of a specific Korean population by performing the Meta-GWAS technique, which predicts an individual's risk or susceptibility to a specific disease by analyzing the diversity of DNA base sequences in the genome. The first server 150 can generate a genetic risk prediction model model using summary statistics from the results of the meta-analysis.

For example, the second server 110 may be a computing device or server for generating genomic information that generates genomic data for a sample. According to one embodiment, although not shown in FIG. 1, the second server 110 may store the generated genomic information in a separate database.

The operation process of the system 10 or the first server 150 and the second server 110 for evaluating the risk of developing diabetic nephropathy described above may also be performed by the electronic device 1000. For example, the electronic device 1000 may acquire reference SNP data or reference omics data including the reference SNP data and a plurality of miRNA data from a reference database connected to the electronic device 1000. Additionally, the electronic device 1000 may acquire genomic data of a target user or a specific clinical sample from an external device connected to the electronic device 1000. The electronic device 1000 may identify a plurality of target omics data identical to the acquired reference omics data from the genomic data of the clinical sample. According to another embodiment, the electronic device 1000 may identify a plurality of target SNP data that are the same as the reference SNP data from the genomic data of the clinical sample, based on the acquired reference SNP data.

The electronic device 1000 may calculate a risk score using the genotype of each of the identified plurality of target omics data and the weight for each of the plurality of target omics data. According to one embodiment, the reference omics data that the electronic device 1000 acquires from the reference database may include a plurality of reference SNP data related to the development of diabetic nephropathy and a plurality of miRNA data related to the development of diabetic nephropathy. . In addition, according to one embodiment, the plurality of reference SNP data includes SNP data related to the onset of diabetic nephropathy, and the effect size corrected using the linkage imbalance and global scaling parameter (Shrinkage factor) of the related SNP data. size) can be reflected.

Figure 2 is a flowchart of a method for predicting the risk of developing diabetic nephropathy by an electronic device according to an embodiment.

In S210, the electronic device 1000 may acquire genomic data of a clinical sample from an external device connected to the electronic device. For example, the electronic device 1000 may obtain genomic data of clinical samples from a server or database that stores genomic data of clinical samples.

In S220, the electronic device 1000 determines the reference error in the genomic data based on reference omics data including a plurality of preset reference SNP (Single Nucleotide Polymorphism) data and a plurality of miRNA data related to the development of diabetic nephropathy. Multiple target omics data identical to the mix data can be selected.

In S230, the electronic device 1000 may calculate a risk score using the genotype of each of the plurality of target omics data and the weight of each of the plurality of target omics data. A specific method by which the electronic device 1000 calculates the risk score will be described in detail with reference to FIG. 3, which will be described later.

In addition, although not shown in FIG. 2, in S210, the electronic device 1000 may further obtain user identification information of the user corresponding to the genomic data of the clinical sample along with the genomic data of the clinical sample. The electronic device 1000 may obtain the user's personal characteristic information from an external device connected to the electronic device based on the user identification information. The electronic device 1000 may identify the user type of the user corresponding to the clinical sample based on the user's personal characteristic information. According to one embodiment, the user's user type may be information reflecting personal characteristics that may vary depending on the user's physical information or the presence or absence of a disease. The electronic device 1000 can be used to subgroup by classifying the personal characteristics of the identified user or the genomic data of clinical samples reflecting the user's personal characteristics.

Additionally, although not shown in FIG. 2, the electronic device 1000 may use at least one prediction model to calculate a risk score. Additionally, the electronic device 1000 may acquire OBIX learning data including at least one of SNP data or miRNA data related to the onset of diabetic nephropathy from an external device to learn a prediction model. The electronic device 1000 subgroups the acquired omics learning data based on preset individual user characteristics, and performs K-fold cross-validation on the subgrouped omics learning data to obtain a plurality of data for each user's individual characteristics. Predictive models can also be created.

According to one embodiment, the electronic device 1000 may calculate a risk score using a plurality of prediction models generated by the method described above.

For example, the electronic device 1000 subgroups the genomic data of a clinical sample acquired from an external device based on the user's individual characteristics, and selects from the subgrouped genomic data the same as preset reference omics data, Multiple target omics data can be selected. The electronic device 1000 inputs the plurality of target omics data selected from the subgrouped genomic data into a plurality of prediction models for each user's individual characteristics, thereby generating output values from each of the plurality of prediction models. It can be obtained.

The electronic device 1000 is applied to the output values of a plurality of prediction models according to the type of the identified user (e.g., type representing different medical health characteristics depending on smoking status, disease status, height, weight, BMI index range). A risk score can be calculated by determining different weights and applying the determined different weights to the output values of each of the plurality of prediction models.

In addition, although not shown in FIG. 2, the electronic device 1000 may calculate a risk score for developing diabetic nephropathy for each user type and specific period to which the individual may belong, in addition to the process of calculating the risk score for the individual. there is. To this end, the electronic device 1000 according to an embodiment applies a plurality of subgrouped target omics data selected from genomic data and output values of each of the plurality of prediction models to a pre-learned Cox proportional hazard survival analysis model. By inputting, a risk score for developing diabetic nephropathy may be calculated from the Cox proportional hazards survival analysis model for the clinical sample for each user type and predetermined period. The electronic device 1000 according to the present disclosure can effectively provide risk information on the user's risk of developing diabetic nephropathy by scoring the risk of developing diabetic nephropathy for each user's user type and specific period for a specific clinical sample.

FIG. 3 is a diagram illustrating a specific process of how an electronic device calculates a risk score according to an embodiment.

In S310, the electronic device 1000 displays an effect size derived from an association test between cases and controls for SNP data from a Genome Wide Association Study (GWAS) database that stores genome-wide association analysis data connected to the electronic device. Size) and colocalization association test to determine whether gene expression results are affected by the same SNP from the TWAS (Transcriptome-based Genome-Wide Association Study) database, which stores transcriptome-wide association analysis data. Extract the effect size derived from .

In S320, the electronic device 1000 displays the main effect weighted by the effect size of each risk allele and the mutual effect weighted by the effect size of the co-regulatory miRNA data for m SNP data associated with the risk of developing diabetic nephropathy. The effect of action can be determined.

In S330, the electronic device 1000 inputs the determined interaction effect into a variational auto encoder using L_1 normalization, and restores the randomly sampled value from the distribution output from the variational auto encoder to the decoder. Nonlinear low-dimensional embeddings can be created. According to one embodiment, the variational autoencoder and decoder used by the electronic device 1000 may be a network model that generates nonlinear low-dimensional embedding based on a neural network.

In S340, the electronic device 1000 inputs the low-dimensional embedding into a preset prediction model model to obtain a Polygenic Risk Score (PRS) defined as the expected value of the multi-condition posterior probability distribution of the estimated beta_j from the prediction model model. can be calculated with the above risk score.

The process by which the electronic device 1000 shown in FIG. 3 calculates a risk score based on a plurality of target omics data or the genotype and weight of each of a plurality of target SNPs in the plurality of target omics data is as follows. It can be performed based on Equation 1.

In Equation 1, PRS _i represents the genetic risk score of individual i, i is an identification number that distinguishes the individual's genomic data, and j is the target omics data or SNP data of the target omics data. The identification number for, G _ij is The prior probability distribution framework represents the effect size for the allele type SNP _j , is defined as a nonlinear low-dimensional space. probability distribution of, Is Expected value of multi-conditional posterior probability distribution including prior probability and epidemiological information, Is is the prior probability estimator, D is the epidemiological information, N is the Gaussian distribution, is a global scaling parameter shared by the genetic aspect, indicating and controlling the sparsity level of the model. May represent an L_1 normalized reduction estimation parameter for the target omics data or SNP data of the target omics data.

More specifically, the modeling framework according to the present disclosure determines the effect size of SNP data through the above-described variable G _ij is designed to introduce a prior distribution of the LOCAL Shrinkage parameter for the allele type SNP _j Can be used to control the sparsity level of the model by adaptively maintaining large signals while imposing strong shrinkage on noise that is relatively close to 0.

FIG. 4 is a diagram illustrating a process in which an electronic device generates a prediction model based on SNP data and miRNA data obtained from GWAS and TWAS databases.

According to one embodiment, the electronic device 1000 may obtain reference data 424 or various omics data from an external device connected to the electronic device. For example, the electronic device 1000 may acquire a plurality of reference SNP data stored in a GWAS database from an external device (eg, a server) connected to the electronic device, and set the acquired SNP data as the discovery set 410. According to one embodiment, the discovery set 410 acquired by the electronic device 1000 may include SNP data and some miRNA data for 356,000 people according to a cohort study from UK Biobank. According to one embodiment, the discovery set 410 may include SNP data related to the risk of developing diabetic nephropathy according to the results of a cohort survey.

According to one embodiment, the electronic device 1000 may acquire miRNA data from an external device connected to the electronic device. For example, the electronic device 1000 may acquire a plurality of miRNA data stored in the TWAS database from an external device connected to the electronic device, and set the acquired miRNA data as the training set 420. According to one embodiment, the training set 420 acquired by the electronic device 1000 includes a plurality of miRNA data obtained from databases (e.g., GTEx EUR, MESA EUR, MESA AA+ Hispanics) that store various miRNA data. It can be included. Additionally, according to one embodiment, the training set 420 may include a plurality of miRNA data related to the onset of diabetic nephropathy according to the results of the cohort survey.

The electronic device 1000 sets some of the SNP data and miRNA data related to the onset of diabetic nephropathy obtained from an external device as a learning dataset (e.g., training dataset), and sets the remaining portion as a test data set (e.g., verification). can be set to three). According to one embodiment, the ratio of the learning data set and the test data set used by the electronic device 1000 may be 7:3, but is not limited to this, and is based on the verification result of the prediction model finally generated by the electronic device. It can be reset.

The electronic device 1000 determines a target data set 430 including target omics data identified as being the same as preset reference omics data from omics data including SNP data and miRNA data acquired from an external device. You can. According to one embodiment, target omics data may include target SNP data and target miRNA data.

The electronic device 1000 may set some of the SNP data acquired from an external device as a learning data set and generate prediction models 462 and 464 based on the set learning data set. For example, in S440, the electronic device 1000 may generate a prediction model 462 that outputs a PRS score related to the risk of developing diabetic nephropathy based on a learning data set including SNP data. In addition, in S450, the electronic device 1000 sets some of the miRNA data acquired from an external device as a learning data set, and based on the set learning data set, a prediction model that outputs a PRS score related to the risk of developing diabetic nephropathy. (464) can be generated.

In Figure 4, the electronic device 1000 is a prediction model 462 and a prediction model based on each of the learning data set including SNP data related to the onset of diabetic nephropathy and the learning data including miRNA data related to the onset of diabetic nephropathy. (464) is shown to generate each, but according to another embodiment, the electronic device 1000 generates a prediction model that outputs a PRS score based on omics data including SNP data and miRNA data. Of course, it can be created.

In addition, according to one embodiment, the electronic device 1000 provides a result of verifying the prediction model 462 based on a test data set (e.g., verification data set) provided as part of the SNP data and a result of verifying the prediction model 462 as part of the miRNA data. Based on the results of verifying the prediction model 464 based on a test data set (e.g., verification data set), higher weight is applied to the output value of the prediction model showing higher performance, and the verification result shows relatively lower performance. In a weighted sum method that applies a relatively low weight to the output value of the prediction model representing, the output values of the prediction model 462 and 464 are weighted and the PRS score according to the weighted sum result is identified as the risk score You may.

Figure 5 shows the process by which an electronic device generates a prediction model based on genomic data subgrouped according to individual characteristics, and the Cox proportional hazard survival analysis model that generates the probability of disease onset by period based on the generation result of the prediction model. This is a drawing showing the process of use.

The electronic device 1000 according to the present disclosure can subgroup omics data according to user characteristics of individuals subject to clinical sample acquisition. According to one embodiment, the user characteristic information may include information about the user's height, weight, BMI index, presence of disease according to urine test results, gender, presence of smoking, and presence of other diseases. User characteristic information can be a standard for determining user type. The electronic device 1000 subgroups genomic data according to the user characteristic information of the individual subject to clinical sample acquisition, and inputs the subgrouped genomic data into a prediction model generated for each subgroup, thereby making predictions learned for each user characteristic information. PRS scores can be obtained from each model.

More specifically, the electronic device 1000 may obtain the corresponding user identification information along with the genomic data of the clinical sample. Based on the user identification information, the electronic device 1000 provides personal characteristic information (e.g., smoking status, gender information, height information, weight information, BMI index, etc.) can be identified. According to one embodiment, the electronic device 1000 may distinguish user types according to user characteristic information based on user identification information.

When user characteristic information is identified, the electronic device 1000 may subgroup the genomic data of the clinical sample according to the user characteristic information. According to another embodiment, when user characteristic information is identified, the electronic device 1000 may classify (eg, subgroup) a plurality of target omics data identified from the genomic data of clinical samples according to user characteristics.

The electronic device 1000 inputs subgrouped genomic data (e.g., subgrouped target omics data) into a prediction model generated for each subgroup, thereby calculating the PRS score obtained from each prediction model learned for each user characteristic information. Comprehensively, based on the comprehensive results, the user's risk of developing diabetic nephropathy can be accurately predicted.

The electronic device 1000 applies different weights to the result values of the prediction model learned specifically for various personal characteristics of the user, based on the user type determined based on the genomic data of the currently acquired clinical sample and the corresponding user identification information. can be applied, and the final risk of developing diabetic nephropathy for the user can be determined by weighting and summing the results of the prediction models according to different weights. That is, the electronic device 1000 according to the present disclosure can effectively diagnose the risk of developing diabetic nephropathy more accurately by using prediction models specifically learned for various personal characteristics of the user.

Below, a process in which the electronic device 1000 learns or generates each prediction model in order to generate a prediction model specialized for each user's individual characteristics described above will be described. For example, the electronic device 1000 may acquire omics data (eg, omics learning data) including SNP data and miRNA data related to the risk of developing diabetic nephropathy from an external device. In S510, the electronic device 1000 may subgroup the acquired omics data based on the user's individual characteristics.

In S520, the electronic device 1000 may generate a plurality of prediction models by performing meta-analysis on the acquired omics data. According to one embodiment, the electronic device 1000 may generate a plurality of prediction models by performing K-fold cross-validation on the acquired omics data. For example, in S522, the electronic device 1000 sets K = 1 and predicts the risk of developing diabetic nephropathy based on omics data according to the first user characteristic item among the subgrouped omics data. 1 A prediction model can be created.

In the same way, in S524, the electronic device 1000 may generate a second prediction model that predicts the risk of developing diabetic nephropathy based on omics data according to the second user characteristic item among the subgrouped omics data. . By repeating the above-described method, the electronic device 1000 is trained specifically for the user's individual characteristics, thereby generating a plurality of prediction models that output a risk score for developing diabetic nephropathy. The electronic device 1000 determines the risk of developing diabetic nephropathy of the individual to be diagnosed by inputting target omics data from genomic data of clinical samples of the individual to be diagnosed into the plurality of generated prediction models.

Even if the electronic device 1000 calculates the risk of developing diabetic nephropathy by reflecting the user's individual characteristics according to the above-described process, the risk score only represents information about the risk probability of developing diabetic nephropathy according to the individual's characteristics, and There is a limitation in not being able to indicate the risk of developing diabetic nephropathy according to period or period.

Therefore, the electronic device 1000 according to the present disclosure inputs subgrouped genomic data (e.g., subgrouped omics data) and result values of prediction models specifically learned for each individual's characteristics into the Cox proportional hazard survival analysis model. , it is also possible to determine the risk of developing diabetic nephropathy by user type 532 and specific period 534 according to the user's personal characteristics.

For example, the electronic device 1000 according to the present disclosure subgroups omics data including SNP data and miRNA data related to the risk of developing diabetic nephropathy from an external device by user personal characteristics, thereby creating a prediction model for each subgroup. You can create learning data sets and verification data sets for creation. The electronic device 1000 receives subgrouped genomic data based on the training data set and verification data set for each subgroup and the result of a prediction model specifically learned for each individual's characteristics. When subgrouped genomic data is input, the user type 532 and A Cox proportional hazards survival analysis model that outputs the risk of developing diabetic nephropathy for a specific period (534) can be trained.

When genomic data of a clinical sample is acquired, the electronic device 1000 subgroups the genomic data of the obtained clinical sample, inputs the subgrouped genomic data into prediction models specifically learned for each individual characteristic, and inputs the learned prediction model. By inputting the output value of and the subgrouped genomic data into the learned Cox proportional hazard survival analysis model, a risk score for developing diabetic nephropathy can be obtained for each user type (532) and specific period (534). Therefore, the electronic device 1000 according to the present disclosure can not only identify the user type 532 according to the user's personal characteristics with respect to the genomic data of the target clinical sample, but also identify the user type 532 according to the user's personal characteristics and in a specific period. The risk of developing diabetic nephropathy can be effectively diagnosed.

Figure 6 is a diagram showing the results of an electronic device selecting reference omics data including reference SNP data.

Referring to figure 620 of FIG. 6 , reference data selected by the electronic device 1000 in relation to the onset of diabetic nephropathy is shown. For example, the electronic device 1000 may acquire SNP data through a GWAS database and acquire miRNA data through a TWAS database. The electronic device 1000 may generate a model that calculates a risk score for SNP data, miRNA data, or omics data including SNP data and miRNA data in various ways.

According to one embodiment, the accuracy of the genetic risk score is greatly affected by the size of the sample population used to predict the degree of effect of the SNP. In order to use GWAS results, the electronic device can obtain summary statistics data from public databases and check whether the required information below is included. The essential information may include chromosome, position, allele information (effect, other allele), effect (beta), standard error, sample size, and p-value.

According to one embodiment, when the electronic device does not have GWAS results conducted by an independent research group or its application to the Korean population is limited considering the specificity of the research group, a 10-fold LOGO (Leave One Group Out) meta-analysis is performed. can be performed. At this time, the effect of utilizing GWAS results from a virtual independent research group can be applied by dividing electronic devices into several groups and analyzing them.

Below, we will explain the specific marker selection process for building a genetic risk model using SNPs selected by the electronic device based on the researcher's input to the electronic device.

The GWAS catalog (https://www.ebi.ac.uk/gwas/) database was used as a public database, and SNPs discovered in papers with a study group size of more than 5,000 people were primarily selected as source SNP information. Alternatively, a different value may be used as the cut-off value of the study group size for selecting source SNP information. SNPs are filtered based on the effect size corrected using Linkage Disequilibrium (LD) value and shrinkage factor (e.g. parameter) from the source SNP information first discovered using a computer device. The filtering process can be performed by Equation 1 above.

To specifically explain the item shown in figure 620 of FIG. 6, SNP (single nucleotide polymorphism) 602 is a genetic gene that shows a difference in one base sequence (A, T, G, C) in the DNA base sequence. It refers to a mutation item, and the Risk Allele (603) refers to the DNA sequence of an allele that increases the risk of developing disease in a given population, and the Non-risk Allele (604) refers to the given It refers to the DNA sequence of an allele that does not increase the risk of disease in the population, the Effect size item (605) indicates the contribution of the SNP to the genetic variation of the trait, and the P-value item (606) represents the statistical significance probability of the association test when the major allele and minor allele are risk alleles, and the Major Allele item (607) refers to the DNA sequence of the most common allele in a given population. , Minor Allele item 608 refers to the DNA sequence of the second most common allele in a given population, and Minor Allele Frequency item 609 refers to the DNA sequence of the second most common allele in a given population. It refers to the frequency at which the most common allele occurs, and the Mapped Gene item (610) may refer to a gene identified in the relative position of the chromosome.

Figure 7 is a diagram showing an example of the ROC curve that verifies the diabetic nephropathy risk analysis model used by the electronic device.

As shown in FIG. 7, the electronic device can use the ROC curve to verify the performance of the prediction model used by the electronic device 1000 to analyze the risk of developing diabetic nephropathy. According to one embodiment, as a result of verifying the performance of the Bayesian neural network model used by the electronic device 1000, an AUC value of 0.797 for sensitivity and specificity was achieved, and the regression used by the electronic device 1000 was achieved. For the model, an AUC value of 0.656 was achieved.

The electronic device 1000 according to the present disclosure may verify the performance of a model used to predict the risk of developing diabetic nephropathy and determine whether to use data output from the model based on the verification result. Additionally, according to one embodiment, the electronic device 1000 may re-learn the diabetic nephropathy risk analysis model by modifying and updating it based on the verification results of the diabetic nephropathy risk analysis model learned by the electronic device.

FIG. 8 is a diagram illustrating the structure of an electronic device for analyzing the risk of diabetic nephropathy and a system including the same according to an embodiment.

According to one embodiment, the system 10 for evaluating the risk of developing diabetic nephropathy includes a control device 200, input devices 210 and 230, memory 220, computing device 240, and network interface 250. It can be included. However, it is not limited to the above-described example, and the system may include more components than those shown in FIG. 8, or may be prepared with fewer components. Additionally, according to one embodiment, the configuration of the system 10 for evaluating the risk of developing diabetic nephropathy shown in FIG. 8 may correspond to the configuration of the electronic device 1000 according to the present disclosure.

According to one embodiment, the system 10 and the electronic device 1000 for evaluating the risk of developing diabetic nephropathy include, in addition to the configuration shown in FIG. 8, a system used to analyze the risk of diabetic nephropathy stored in the memory 220. It may further include at least one processor executing one or more instructions. According to one embodiment, a processor may be included in the configuration of the control device 200 shown in FIG. 8.

The input devices 210 and 230 may receive genomic data of clinical samples. According to one embodiment, the input devices 210 and 230 may correspond to a user interface or an input interface. The system 10 for evaluating the risk of developing diabetic nephropathy according to an embodiment acquires genomic data of a specific clinical sample using input devices 210 and 230.

More specifically, the input devices 210 and 230 may refer to means through which a user inputs a sequence to control a system or electronic device. For example, the user input interface (not shown) includes a key pad, a dome switch, and a touch pad (contact capacitive type, pressure-type resistive type, infrared detection type, surface ultrasonic conduction type, Integral tension measurement method, piezo effect method, etc.), jog wheel, jog switch, etc., but are not limited to these.

Additionally, according to one embodiment, a user input interface (not shown) may receive a user's input sequence for a screen output by an electronic device or system on a display. Additionally, a user input interface (not shown) may receive a touch input from a user touching the display or a key input through a graphical user interface on the display.

Memory 220 may store one or more instructions. For example, one or more instructions stored in the memory 220 may be used to perform a method of predicting the risk of developing diabetic nephropathy by being performed by the control device 200. Additionally, the memory 220 may store data input to or output from a system or electronic device, in addition to programs for processing and control of the processor.

Additionally, according to one embodiment, the memory 220 may store reference omics data, reference SNP (Single Nucleotide Polymorphism) data, genomic data of clinical samples, plural target omics data, or plural target SNP data. Additionally, the memory 220 may store information about an artificial intelligence model used by an electronic device or system. According to one embodiment, the memory 220 may further store training data information used by the electronic device to train an artificial intelligence model, and may further store parameter information for the artificial intelligence model.

For example, when models based on not only the learned neural network but also the already created neural network are modified, the memory 220 may further store information about the layers of the modified models and the weights between the layers.

The memory 220 is a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (for example, SD or XD memory, etc.), and RAM. (RAM, Random Access Memory) SRAM (Static Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, magnetic disk , and may include at least one type of storage medium among optical disks.

The computing device 240 selects a plurality of target omics data or a plurality of target SNP data based on the reference SNP data or reference omics data from the genomic data of the clinical sample and executes one or more instructions stored in the memory. , a risk score for developing diabetic nephropathy for the clinical sample can be calculated using the genotype and effect size for each of the plurality of target omics data or the plurality of target SNP data. Additionally, the computing device 240 may perform a linkage analysis process with covariate data such as patient epidemiological information or other diagnostic information.

According to one embodiment, the memory 220 may further store information about artificial intelligence models, machine learning models, neural network models, and prediction models used by the electronic device 1000. According to one embodiment, the artificial intelligence model used by the electronic device or system may be a model that can be learned based on an artificial intelligence learning algorithm. According to one embodiment, the artificial intelligence model may include a neural network model. For example, a neural network model is an artificial neural network, which may refer to a computing system inspired by biological neural networks. Unlike classic algorithms that perform tasks according to predefined conditions, artificial neural network models can learn to perform tasks by considering multiple samples.

According to one embodiment, an artificial neural network model may have a structure in which artificial neurons are connected, and connections between neurons may be referred to as synapses. Neurons can process received signals and transmit the processed signals to other neurons through synapses. The output of a neuron may be referred to as an activation, and the neuron and/or synapse may have a weight that may vary, and depending on the weight, the influence of the signal processed by the neuron may increase or decrease. .

For example, a neural network model may include a plurality of blocks defined by layers and weights related to the connection strengths of the layers. More specifically, the neural network model is a neural network model, where each of the plurality of neural network layers within the neural network model has a plurality of weights (weight values), and a neural network is formed through calculations between the calculation results of the previous layer and the plurality of weights. Perform calculations. The plurality of weights of the plurality of neural network layers can be optimized based on the learning results of the artificial neural network.

For example, during the learning process, a plurality of weights may be modified and updated so that loss or cost values obtained from an artificial intelligence model (eg, neural network model) are reduced or minimized. The artificial intelligence model used by the electronic device according to the present disclosure may include a deep neural network (DNN), for example, a Convolutional Neural Network (CNN), a Deep Neural Network (DNN), and a Recurrent Neural Network (RNN). ), RBM (Restricted Boltzmann Machine), DBN (Deep Belief Network), BRDNN (Bidirectional Recurrent Deep Neural Network), LSTM (Long Short-Term Memory) model, or Deep Q-Networks, etc. It is not limited to one example.

According to one embodiment, the electronic device according to the present disclosure may be a smartphone, PC, mobile phone, laptop, media player, server, or other mobile or non-mobile computing device equipped with an AI program and capable of analyzing medical data. It is not limited to this.

Additionally, according to one embodiment, the system or electronic device according to the present disclosure may be connected for communication to an external device such as a server. For example, a server may include other computing devices that are connected to the electronic device through a network and can transmit and receive data with the electronic device.

According to one embodiment, the control device 200 may include a memory that stores one or more instructions and at least one processor that performs the one or more instructions. The control device 200 may perform at least part of a method of predicting the risk of developing diabetic nephropathy by linking with the computing device 240.

For example, a processor (not shown) of the control device 200 typically controls the overall operation of the electronic device or system. For example, a processor (not shown) can generally control a user input interface (not shown), a network interface, an input device, a computing device, etc. by executing programs stored in the memory 220.

According to one embodiment, a processor (not shown) acquires genomic data of a clinical sample from an external device connected to the electronic device, and collects a plurality of preset reference SNP data or a plurality of miRNA data related to the onset of diabetic nephropathy, and the Selecting a plurality of target omics data or a plurality of target SNP data from the genomic data based on a plurality of reference omics data including a plurality of reference SNP data, and selecting the selected target omics data or target SNP data A risk score can be calculated using each genotype and the weight for each.

According to one embodiment, the network interface 250 may include one or more components that allow an electronic device or system to communicate with other devices (not shown) and servers. According to one embodiment, the server with which the electronic device 1000 or the system 10 communicates is a local area network (LAN), a wide area network (WAN), or a value added network (VAN). , and may include a combination of at least one of a mobile radio communication network.

The other device (not shown) may be a computing device such as the electronic device 1000 or a sensing device, but is not limited thereto. According to one embodiment, the network interface 250 may include a short-range communication unit (not shown) or a long-distance communication unit (not shown).

The short-range wireless communication unit (not shown) includes a Bluetooth communication unit, a Bluetooth Low Energy (BLE) communication unit, a Near Field Communication unit, a WLAN (Wi-Fi) communication unit, a Zigbee communication unit, and an infrared communication unit. (IrDA, infrared Data Association) communication unit, WFD (Wi-Fi Direct) communication unit, UWB (ultra wideband) communication unit, Ant+ communication unit, etc., but is not limited thereto.

The remote communication unit may include a mobile communication unit or a broadcast reception unit. For example, the mobile communication unit transmits and receives wireless signals to at least one of a base station, an external terminal, and a server on a mobile communication network. Here, the wireless signal may include various types of data according to voice signals, video call signals, or text/multimedia message transmission and reception. The broadcast receiver receives broadcast signals and/or broadcast-related information from the outside through a broadcast channel. Broadcast channels may include satellite channels and terrestrial channels. Of course, depending on the implementation example, the electronic device may not include a broadcast receiver (not shown).

According to one embodiment, the electronic device or system may further include an audio/video (A/V) input unit (not shown). The A/V input unit is for inputting audio signals or video signals, and may include cameras and microphones. The camera can acquire related medical image or video image frame data through an image sensor in video call mode or shooting mode, and the image captured through the image sensor can be processed through a processor or a separate image processing unit (not shown). You can.

The microphone can receive acoustic signals from an external device or a user. The microphone can receive the user's voice input. Microphones can use various noise removal algorithms to remove noise generated in the process of receiving external acoustic signals.

The method according to one embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and configured for the present disclosure or may be known and usable by those skilled in the art of computer software.

Additionally, a computer program device including a recording medium storing a program for performing a method different from the above embodiment may be provided. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc.

Although the embodiments of the present disclosure have been described in detail above, the scope of the rights of the present disclosure is not limited thereto, and various modifications and improvements made by those skilled in the art using the basic concept of the present disclosure defined in the following claims are also possible. falls within the scope of rights.

Claims (6)

In a method for an electronic device to predict the risk of developing diabetic nephropathy,
Obtaining genomic data of a clinical sample from an external device connected to the electronic device;
Based on reference omics data including a plurality of preset reference SNP (Single Nucleotide Polymorphism) data and a plurality of miRNA data related to the onset of diabetic nephropathy, a plurality of target errors identical to the reference omics data are identified from the genomic data. Selecting mix data; and
Calculating a risk score using a genotype of each of the plurality of target omics data and a weight for each of the plurality of target omics data; Including,
The reference omics data includes a plurality of reference SNP data related to the development of diabetic nephropathy and a plurality of mi RNA data related to the development of diabetic nephropathy,
A method characterized in that it reflects the effect size corrected using the linkage imbalance and global scaling parameter (Shrinkage factor) of the omics data related to the development of diabetic nephropathy. The method of claim 1, wherein the step of calculating the risk score is
The effect size derived from the association test between cases and controls for SNP data is extracted from the GWAS database that stores genome-wide association analysis data connected to the electronic device, and the transcriptome-wide association analysis data is stored. Extracting an effect size derived from a colocalization association test for whether gene expression results are affected by the same SNP from the TWAS database;
Determining a main effect weighted by the effect size of each risk allele and an interaction effect weighted by the effect size of co-regulatory miRNA data for the m SNPs associated with the risk of developing diabetic nephropathy;
Generating non-linear low-dimensional embedding by inputting the determined interaction effect into a variational auto encoder using L_1 normalization and restoring randomly sampled values from the distribution output from the variational auto encoder to a decoder; and
Inputting the low-dimensional embedding into a preset prediction model model to calculate a Polygenic Risk Score (PRS), defined as the expected value of the multi-condition posterior probability distribution of the estimated beta_j, as the risk score from the prediction model model. ; Method, including. The method of claim 1, wherein the step of calculating the risk score is
Calculating the risk score based on Equation 1 below; Including,
[Equation 1]

In Equation 1, PRS _i represents the genetic risk score of individual i, i is an identification number that distinguishes the individual's genomic data, and j is the target omics data or SNP data of the target omics data. The identification number for, G _ij is Prior probability distribution framework, is defined as a nonlinear low-dimensional space. probability distribution of, Is Expected value of multi-conditional posterior probability distribution including prior probability and epidemiological information, Is is the prior probability estimator, D is the epidemiological information, N is the Gaussian distribution, is a global scaling parameter (shrinkage parameter), is an L_1 normalized reduction estimation parameter for the target omics data or SNP data of the target omics data. The method of claim 1, wherein the plurality of reference SNP data of the reference omics data are:
rs12531478-A, rs17373728-C, rs5750250-G, rs11107616-C, rs136161-G, rs4879670-G, rs13259109-G, rs1298908-G, rs304029-G, rs9510795-A, rs10952362-C, rs4667466-T, rs10778560- C, RS7975752-G, RS731565-T, RS484965-C, RS6910061-A, RS1424609-G, RS2596230-G, RS1677894-G, RS5750250-G, RS5750250-G, RS136161-G Including data A method, characterized in that. In the electronic device for predicting the risk of developing diabetic nephropathy,
network interface;
A memory that stores one or more instructions; and
At least one processor executing the one or more instructions; Including,
By executing one or more of the instructions above,
Obtaining genomic data of clinical samples from an external device connected to the electronic device,
Based on reference omics data including a plurality of preset reference SNP (Single Nucleotide Polymorphism) data and a plurality of miRNA data related to the onset of diabetic nephropathy, a plurality of target errors identical to the reference omics data are identified from the genomic data. Select mix data,
Calculate a risk score using the genotype of each of the plurality of target omics data and a weight for each of the plurality of target omics data,
The reference omics data includes a plurality of reference SNP data related to the development of diabetic nephropathy and a plurality of miRNA data related to the development of diabetic nephropathy,
An electronic device, characterized in that it reflects the effect size corrected using the linkage imbalance and global scaling parameter (Shrinkage factor) of the reference omics data related to the onset of diabetic nephropathy. In a method for an electronic device to predict the risk of developing diabetic nephropathy,
Obtaining genomic data of a clinical sample from an external device connected to the electronic device;
Based on reference omics data including a plurality of preset reference SNP (Single Nucleotide Polymorphism) data and a plurality of miRNA data related to the onset of diabetic nephropathy, a plurality of target errors identical to the reference omics data are identified from the genomic data. Selecting mix data; and
Calculating a risk score using a genotype of each of the plurality of target omics data and a weight for each of the plurality of target omics data; Including,
The reference omics data includes a plurality of reference SNP data related to the development of diabetic nephropathy and a plurality of mi RNA data related to the development of diabetic nephropathy,
A program for executing the method on a computer, characterized in that it reflects the effect size corrected using the linkage imbalance of the omics data related to the development of diabetic nephropathy and a global scaling parameter (Shrinkage factor) A computer-readable recording medium that records information.

Images