A phenotype analysis method for disease prediction

Abstract

The present application relates to the technical field of disease prediction, and discloses a phenotype analysis method for disease prediction, which comprises the following steps: constructing a database, determining the phenotype similarity of rare diseases and common diseases, referring to the difference of gene data of rare diseases and common diseases, constructing a "rare disease-common disease" common database, processing patient data to find the best combination of phenotype characteristics, calculating the disease matching score, calculating the cross-entropy loss of the phenotype characteristic matching model of disease prediction, taking the weighted sum of the classification loss function in the "rare disease-common disease" common database and the cross-entropy loss of the phenotype characteristic matching model as the total loss function of the disease prediction model, outputting the prediction result, extracting effective phenotype characteristics based on a graph convolution network for data comparison of rare diseases and common diseases, and extracting disease matching difference from the commonality of rare diseases and common diseases to predict diseases, which can solve the problem of easy confusion between diseases and to a certain extent, avoid misdiagnosis and missed diagnosis of rare diseases.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of disease prediction, and specifically relates to a phenotype analysis method for disease prediction. BACKGROUND

[0002] At present, disease prediction mostly adopts a deep learning method, and computer technology is used to assist in predicting the diagnosis result, but in order to improve the disease prediction diagnosis accuracy, complex medical multi-modal data need to be fully utilized to extract effective information hidden therein, and these data include medical imaging data and corresponding non-imaging phenotype characteristics such as the age, height and body function of a patient. These data are difficult to process by using a traditional deep learning method, and not every phenotype characteristic contributes to disease prediction. For disease prediction, an effective method is used to screen out phenotype characteristics that have a negative effect on the disease prediction result, which can effectively improve the accuracy of the disease prediction model, and has important diagnostic significance in the prediction of rare diseases. Since there is great commonality between rare diseases and common diseases, it is easy to misdiagnose a rare disease as a common disease in general diagnosis, which has a great adverse effect on the subsequent diagnosis and treatment of the patient. Research has found that an adaptive multi-layer aggregation graph convolutional network can be used for disease prediction, mainly by designing an encoder to automatically select the best combination of phenotype characteristics, introducing a multi-layer aggregation graph convolutional network with a multi-aggregation mode to select advantageous structural information for each node, and designing a group graph structure according to the spatial distribution and text similarity of the phenotype characteristics, allowing each effective phenotype characteristic to have a positive effect on the disease prediction result, so that the disease prediction model can automatically find the best phenotype characteristic information from each layer, thereby improving the disease diagnosis accuracy to a certain extent. However, for rare diseases, no classification detection is performed between diseases in the disease classification process, especially in the case where rare diseases and common diseases have great similarity in phenotype characteristics, it is difficult to distinguish the difference between rare diseases and corresponding common diseases, which easily leads to misdiagnosis and missed diagnosis of rare diseases. Therefore, how to extract effective phenotype information for rare disease diagnosis is of great significance to medical disease prediction. SUMMARY

[0003] The present application aims to provide a phenotype analysis method for disease prediction, which extracts effective phenotype characteristics based on a graph convolutional network for comparison of rare disease data and common disease data, and fuses the rare disease data and the common disease data to construct a "rare disease-common disease" common database, matches the phenotype characteristics of a patient with the database, and extracts disease matching differences from the commonality of rare diseases and common diseases, thereby effectively solving the problems in the background art.

[0004] To achieve the above object, the present application provides the following technical scheme:

[0005] A phenotype analysis method for disease prediction, comprising the following analysis steps:

[0006] Step one: constructing a database, including constructing a rare disease database, constructing a common disease database, and constructing a patient information database;

[0007] Step two: database comparison and fusion processing, by determining the phenotype similarity of rare disease phenotype characteristics and common disease phenotype characteristics, and referring to the difference between rare disease gene data and common disease gene data, a "rare disease-common disease" common database is constructed;

[0008] Step three: patient information data processing, defining a set of patient phenotype characteristics H as K={K h}, including patient personal basic information, patient family genetic history, and patient physical characteristics, using an adjacency matrix

[0009] Finding suitable phenotype characteristics and calculating the response phenotype characteristic selection score, and then calculating the edge weight to obtain the best combination of phenotype characteristics;

[0010] Step four: input the best combination of patient phenotype characteristics into the rare disease database, the common disease database, and the "rare disease-common disease" common database, respectively calculate the disease matching score of the best combination of patient phenotype characteristics in multiple databases, and output Z={Z i}∈R n×p , representing the matching score of all data, Z i represents the matching score of the i node in the database;

[0011] Step five: calculate the cross-entropy loss of the phenotype characteristic matching model for disease prediction Y ij represents the label information of the data;

[0012] Step six: taking the weighted sum of the classification loss function L H-C in the "rare disease-common disease" common database and the cross-entropy loss L w of the phenotype characteristic matching model as the total loss function of the disease prediction model, the smaller the total loss function value, the higher the prediction accuracy;

[0013] Step seven: output the prediction result.

[0014] As a further scheme of the present application: the rare disease database in step one includes rare disease categories, rare disease genetic information, rare disease phenotype characteristics, and rare disease gene data. In combination with existing rare disease knowledge bases and corresponding rare disease cases, different phenotypes and gene sequences are correspondingly positioned in the rare disease entries. The common disease database includes common disease categories, common disease genetic information, common disease phenotype characteristics, and common disease gene data. In combination with common disease pathogenic genes and clinical medical cases, different phenotypes and gene sequences are correspondingly supplemented in the common disease entries. The patient information database includes patient disease symptoms and medical examination data.

[0015] As a further scheme of the present application: in step two, the data classification in the rare disease database is based on the clustering algorithm to perform upsampling processing on the samples, increase the accuracy of data classification, and reduce the negative impact of the cross and overlap phenomenon between each disease category in the rare disease database on data classification.

[0016] As a further scheme of the present application: in step two, the phenotype similarity of the rare disease phenotype characteristics and the common disease phenotype characteristics is defined as sim. For the rare disease x and the common disease y, there is a phenotype similarity sim(x, y). The phenotype similarity sim(x, y) of the rare disease x and the common disease y is used as prior information for detecting gene difference. The difference var(x, y) of the rare disease gene data and the common disease gene data n = α * W p * var(x, y) n-1 + (1-α) * sim(x, y), α represents a weight value, W p represents a gene interaction network.

[0017] As a further scheme of the present application: the classification loss function in the “rare disease-common disease” co-class database is N represents the number of samples participating in classification in the database, is used to verify the correlation loss of the disease x and the disease y, is used for the matching degree of the disease x and the disease y, γ represents an influence factor in the optimization process, d is the dimension of the feature vector of the disease x and the disease y, C x , C y represent the covariance matrix of the feature vector of the disease x and the disease y in the d dimension, and φ is an input initial constant.

[0018] As a further scheme of the present application: in step three, the adjacency matrix α h is the phenotype selection score of the phenotype K h , γ is the distance measurement of the value of the phenotype feature K h of two nodes, v, w ∈ H, wherein is the phenotype characteristic K h the number of qualified samples.

[0019] As a further scheme of the present application: when K h is a non-quantitative phenotype characteristic, define as a function of threshold value θ, refers to the number of qualified samples in the p and u categories of the phenotype characteristic K h , and γ = 1, when Kh is a quantitative phenotype characteristic, define as a function of threshold value δ, refers to the number of qualified samples in the p category of the phenotype characteristic K h , define the closed interval D [α, β] ∈ {K h}, refers to the number of qualified samples in the p category of the phenotype characteristic K h that do not belong to the closed interval D, and

[0020]

[0021] Compared with the prior art, the present application has the beneficial effects that:

[0022] Based on the graph convolution network, effective phenotype characteristics are extracted for data comparison of rare diseases and common diseases, and rare disease data and common disease data are fused to construct a "rare disease-common disease" common database. The phenotype characteristics of patients are matched with the database, and the disease matching difference is extracted from the commonality of rare diseases and common diseases to predict the disease, which can solve the confusion problem between diseases and avoid misdiagnosis and missed diagnosis of rare diseases to a certain extent. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0024] Figure 1 is a structural schematic diagram of a phenotype analysis method for disease prediction. DETAILED DESCRIPTION

[0025] In order to make the technical problems, technical solutions and beneficial effects of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application and not to limit the present application.

[0026] Embodiment 1:

[0027] Please refer to Figure 1 In the embodiments of the present application, a phenotype analysis method for disease prediction comprises the following steps:

[0028] Step one: constructing a database, including constructing a rare disease database, constructing a common disease database, and constructing a patient information database;

[0029] Step two: database comparison and fusion processing, by determining the phenotype similarity of rare disease phenotype characteristics and common disease phenotype characteristics, and referring to the difference between rare disease gene data and common disease gene data, a "rare disease-common disease" common database is constructed, the phenotype similarity of rare disease phenotype characteristics and common disease phenotype characteristics is defined as sim, and there is a phenotype similarity sim(x, y) between rare disease x and common disease y The phenotype similarity sim(x, y) between rare disease x and common disease y is used as prior information for detecting gene difference, and the difference var(x, y) between rare disease gene data and common disease gene data n = α * W p * var(x, y) n-1 + (1-α) * sim(x, y), α represents a weight value, W p represents a gene interaction network, and the classification loss function in the "rare disease-common disease" common database is N represents the number of samples participating in classification in the database is used to verify the correlation loss of disease x and disease y, is used to match the degree of disease x and disease y, γ represents an influence factor in the optimization process, d is the dimension of the feature vector of disease x and disease y, C x , C y represents the covariance matrix of the feature vector of disease x and disease y in d dimensions, and φ is an input initial constant;

[0030] Step three: patient information data processing, defining a set of patient phenotype characteristics H as a set K = {K h}, including patient personal basic information, patient family genetic history, and patient physical representation, using an adjacency matrix to find suitable phenotype characteristics and calculate the response phenotype characteristic selection score, and then calculating the edge weight to obtain the best combination of phenotype characteristics, wherein α h is the phenotype K hphenotype selection score, γ is the phenotype feature K h ranging of the value of the phenotype feature Kh, v, w ∈ H, wherein is the number of samples meeting the requirements of the phenotype feature Kh, when K h is a non-quantitative phenotype feature, define as a function of the threshold value θ, refers to the number of samples meeting the requirements in the p and u categories of the phenotype feature K h , and γ = 1, when Khis a quantitative phenotype feature, define as a function of the threshold value δ, refers to the number of samples meeting the requirements in the p category of the phenotype feature K h , define the closed interval D [α, β] ∈ {K h}, refers to the number of samples meeting the requirements in the p category of the phenotype feature K h that do not belong to the closed interval D, and

[0031] Step four: input the best combination of patient phenotype features into the rare disease database, common disease database and “rare disease-common disease” co-class database, respectively calculate the disease matching score of the best combination of patient phenotype features in multiple databases, output Z = {Z i} ∈ R n×p represents the matching score of all data, Z i represents the matching score of the i node in the database;

[0032] Step five: calculate the cross-entropy loss of the phenotype feature matching model of disease prediction Y ij represents the label information of the data;

[0033] Step six: take the weighted sum of the classification loss function L H-C in the “rare disease-common disease” co-class database and the cross-entropy loss L w of the phenotype feature matching model as the total loss function of the disease prediction model, the smaller the total loss function value, the higher the prediction accuracy;

[0034] Step seven: output the prediction result

[0035] By adopting the technical scheme, the effective phenotype features are extracted based on the graph convolution network for comparison of rare disease data and common disease data, the rare disease data and the common disease data are fused to construct a "rare disease-common disease" common database, the phenotype features of the patient are matched with the database, and the disease matching difference is extracted from the commonness of the rare disease and the common disease to predict the disease, which can solve the confusion problem between diseases and avoid misdiagnosis and missed diagnosis of the rare disease to a certain extent.

[0036] Embodiment 2:

[0037] Please refer to Figure 1 In the embodiment of the application, a phenotype analysis method for disease prediction comprises the following steps:

[0038] Step 1: Constructing a database, including constructing a rare disease database, constructing a common disease database, and constructing a patient information database, the rare disease database includes a rare disease category, rare disease genetic information, rare disease phenotype features, and rare disease gene data, different phenotypes and gene sequences are positioned in the rare disease item in combination with existing rare disease knowledge base and corresponding rare disease cases, the common disease database includes a common disease category, common disease genetic information, common disease phenotype features, and common disease gene data, different phenotypes and gene sequences are supplemented in the common disease item in combination with common disease pathogenic genes and clinical medical cases, and the patient information database includes patient disease symptoms and medical examination data.

[0039] Step 2: Database comparison and fusion processing, by determining the phenotype similarity of the rare disease phenotype features and the common disease phenotype features, and referring to the difference between the rare disease gene data and the common disease gene data, a "rare disease-common disease" common database is constructed, the phenotype similarity of the rare disease phenotype features and the common disease phenotype features is defined as sim, and the phenotype similarity of the rare disease x and the common disease y is The phenotype similarity sim(x, y) of the rare disease x and the common disease y is used as prior information for detecting gene difference, the difference var(x, y) between the rare disease gene data and the common disease gene data n = alpha * W p * var(x, y) n-1 + (1-alpha) * sim(x, y), alpha represents a weight value, W p represents a gene interaction network, and the classification loss function in the "rare disease-common disease" common database is N represents the number of samples participating in classification in the database, used for verifying the correlation loss of the disease x and the disease y, The matching degree of disease x and disease y, γ represents the influence factor in the optimization process, d is the dimension of the feature vector of disease x and disease y, C x , C y represents the covariance matrix of the feature vector of disease x and disease y in d dimensions, and φ is the input initial constant;

[0040] Step three: patient information data processing, define a set of patient phenotype characteristics H as K h} including patient personal basic information, patient family history of genetic diseases, patient physical characteristics, using the adjacency matrix Find the appropriate phenotype characteristics and calculate the response phenotype feature selection score, and then calculate the edge weight to obtain the best combination of phenotype characteristics, where α h is the phenotype selection score of phenotype K h , γ is the distance of the value of the phenotype characteristics K h of the two nodes, v, w ∈ H, where is the number of samples that meet the requirements of phenotype characteristics Kh, when Kh is a non-quantitative phenotype characteristic, define as a function of threshold θ, refers to the number of samples that meet the requirements in the p and u categories of phenotype characteristics K h , and γ = 1, when Kh is a quantitative phenotype characteristic, define as a function of threshold δ, refers to the number of samples that meet the requirements in the p category of phenotype characteristics K h , define the closed interval D [α, β] ∈ {K h}, refers to the number of samples that meet the requirements in the p category of phenotype characteristics K h that do not belong to the closed interval D, and

[0041]

[0042] Step four: input the best combination of patient phenotype characteristics into the rare disease database, common disease database and "rare disease-common disease" database, respectively calculate the disease matching score of the best combination of patient phenotype characteristics in multiple databases, output Z = {Z i} ∈ R n×p represent the matching score of all data, Z i represents the matching score of node i in the database;

[0043] Step five: calculate the cross-entropy loss of the phenotype feature matching model of disease prediction Y ij represents the label information of the data;

[0044] Step six: the classification loss function L in the "rare disease-common disease" co-class database H-C and the cross-entropy loss L of the phenotype feature matching model w The weighted sum of the above is the total loss function of the disease prediction model. The smaller the total loss function value is, the higher the prediction accuracy is.

[0045] Step seven: output the prediction result, evaluate the prediction result, and verify the accuracy of the prediction result by verifying the phenotype features of the patient according to the prediction result. Since there are few samples in the rare disease database but many disease categories involved, and there is overlap between each disease category, as the disease data in the database increases, the error rate or accuracy of the disease classification result will change to some extent. Therefore, during disease classification, it is necessary to ensure that the number of rare disease samples and the number of common disease samples are balanced as much as possible to prevent the instability of disease classification performance. The method of upsampling the samples based on the clustering algorithm in the rare disease database data classification, and downsampling the samples based on the Bayesian algorithm in the common disease database data classification is adopted to balance the data and increase the accuracy of data classification. At the same time, the sample set in the rare disease database is defined as {H i}, the sample set in the common disease database is defined as {C i}, and the matching degree of the verification sample H i and C i is verified d represents the dimension of the sample H i and C i . At this time, the improved classification loss function is N represents the total number of samples, and ε is the adjustment coefficient. P i represents the prediction probability of the sample features.

[0046] By using the above technical scheme: by inversely verifying the prediction result, extracting the phenotype features from the database, adding data balancing processing in the data classification process, and optimizing the verification model by improving the loss function of the prediction result verification model, the accuracy of the prediction result can be ensured to guarantee the balance of data classification, avoid the negative impact of large sample data difference on data classification, and improve the accuracy of the prediction result to a certain extent. The problem of easy confusion between rare diseases and common diseases is solved.

[0047] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and inventive concept of the present application, makes equivalent replacement or change within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method of phenotype analysis for disease prediction, characterized by, The method comprises the following analysis steps: Step one: constructing a database, including constructing a rare disease database, constructing a common disease database, and constructing a patient information database; Step two: database comparison fusion processing, by measuring the phenotype similarity of rare disease phenotype characteristics and common disease phenotype characteristics, and referring to the difference between rare disease gene data and common disease gene data, a "rare disease-common disease" database is constructed; in the rare disease database, the data classification is based on the clustering algorithm to do up-sampling processing, increase the accuracy of data classification, reduce the negative impact of the cross and overlap between each disease category in the rare disease database on data classification; define the phenotype similarity of rare disease phenotype characteristics and common disease phenotype characteristics as sim, for rare disease x and common disease y , the phenotype similarity sim(x, y) of rare disease x and common disease y is used as prior information for detecting gene difference, the difference between rare disease gene data and common disease gene data , α represents the weight value, W p represents the gene interaction network Step three: patient information data processing, define a set of patient phenotype characteristics H as a set , including patient personal basic information, patient family genetic history, patient physical characteristics, using adjacency matrix Find the appropriate phenotype characteristics and calculate the response phenotype characteristic selection score, and then calculate the edge weight to obtain the best combination of phenotype characteristics; Step four: input the best combination of patient phenotype characteristics into the rare disease database, common disease database and "rare disease-common disease" co-class database, respectively calculate the disease matching scores of the best combination of patient phenotype characteristics in multiple databases, and output Z represents the matching score of all data i Z represents the matching score of i node in the database Step five: calculate the cross-entropy loss of the disease prediction phenotype feature matching model , Y ij label information representing data; Step six: the classification loss function L in the "rare disease-common disease" co-classification database H-C and the cross-entropy loss L of the phenotype feature matching model W The weighted sum of the two is the total loss function of the disease prediction model. The smaller the total loss function value is, the higher the prediction accuracy is. Step seven: outputting a prediction result.

2. The method of claim 1, wherein the method is for disease prediction. The rare disease database in the step one comprises a rare disease category, rare disease genetic information, rare disease phenotype characteristics, and rare disease gene data. Different phenotypes and gene sequences are correspondingly positioned in the rare disease entries by combining existing rare disease knowledge bases and corresponding rare disease cases. The common disease database comprises a common disease category, common disease genetic information, common disease phenotype characteristics, and common disease gene data. Different phenotypes and gene sequences are correspondingly supplemented in the common disease entries by combining common disease pathogenic genes and clinical medical cases. The patient information database comprises patient disease symptoms and medical examination data.

3. The method of claim 1, wherein the method is a disease prediction-oriented phenotyping method. The classification loss function in the "rare disease-common disease" co-classification database is N represents the number of samples participating in classification in the database, Loss for verifying the correlation between disease x and disease y, The matching degree of disease x and disease y, Represent the influence factor in the optimization process, d is the dimension of the feature vector of disease x and disease y, C x , C y Represent the covariance matrix of the feature vectors of disease x and disease y in d dimensions, Is the input initial constant.

4. The method of claim 1, wherein the method is a disease prediction-oriented phenotyping method. The adjacency matrix in the step three , wherein is the phenotype selection score of the phenotype K h , is the distance measurement of the value of the phenotype feature K h h of the two nodes, ∈H, , wherein is the number of samples that meet the requirements of the phenotype feature Kh.

5. The method of claim 3, wherein the method is a disease prediction-oriented phenotyping method. When Kh is a non-quantitative phenotypic feature, define as a function of the threshold value θ, , refers to the number of samples in the p category of the phenotypic feature K h with the requirement, while Y = 1, when Kh h is a quantitative phenotypic feature, define as a function of the threshold value δ, , refers to the number of samples in the p category of the phenotypic feature K h with the requirement, define the closed interval , refers to the number of samples in the p category of the phenotypic feature K h with the requirement that does not belong to the closed interval D, while .

Images