Biomarker screening method and system based on transfer learning and difference network

Through the method based on transfer learning and differential network, the auxiliary data set and the target data set are integrated to generate differential networks, which solves the problem of insufficient efficiency and accuracy of gene regulation relationship analysis in gene expression data, and achieves more efficient and accurate biological Marker screening.

CN118609659BActive Publication Date: 2024-11-08SHANDONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411080505.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-08
Estimated Expiration
2044-08-08

AI Technical Summary

Technical Problem

Existing methods for screening biomarker for gene expression data are difficult to effectively analyze the regulatory relationship between genes under different physiological states, resulting in insufficient efficiency and accuracy of biomarker screening.

Method used

Using a method based on transfer learning and differential networks, differential networks are generated by integrating auxiliary data sets and target data sets with certain similarities, and using transfer learning to integrate data to improve the estimation accuracy of gene regulatory networks and relax the sparse assumption. Optimize the estimation process of differential networks.

Benefits of technology

It improves the efficiency and accuracy of differential network estimation, reduces errors, and enhances the accuracy of edge selection. It has better performance than the single dataset method, can quickly converge and provide theoretical guarantees.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118609659B_ABST
    Figure CN118609659B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of biomarker screening, and discloses a biomarker screening method and system based on transfer learning and difference network, based on sample estimation of the first and second covariance matrices, a difference network corresponding to an auxiliary data set is obtained; based on a second disease group data subset that has suffered from disease A and a second control group data subset that has not suffered from disease A, a third and fourth covariance matrix is ​​generated; based on sample estimation of the third and fourth covariance matrices and the difference network corresponding to the auxiliary data set, the difference between the difference network corresponding to the target data set and the auxiliary data set is obtained; the difference between the difference networks is added to the difference network corresponding to the auxiliary data set to obtain the difference network corresponding to the target data set, and the biomarker screened is determined based on the difference network corresponding to the target data set. The present invention improves the estimation and inference accuracy of the gene regulatory network in the target source.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of biomarker screening, and in particular to a biomarker screening method and system based on transfer learning and difference network. Background Art

[0002] Using gene expression data to screen biomarkers is one of the main ways to study biomarkers in recent years. This type of method is generally based on gene expression data to find differentially expressed genes, and then perform subsequent analysis on the differentially expressed genes. However, the inventors found that the gradual study of genomics in recent years has shown that there is a complex relationship between genes and genes and related expression products, and the regulation of gene expression is not isolated. They regulate and relate to each other to form an interactive network. Gene-Gene Interaction (GGI, Gene-Gene Interaction) includes the most common gene regulation, protein-DNA, protein-RNA, protein-protein interaction and RNA-RNA interaction. The internal or external interactions between genes and biological molecules such as proteins, RNA, and metabolites constitute a huge complex network. Among them, biological molecules such as genes serve as network nodes, and the physical and functional connections between molecules serve as network edges to connect them. Due to the involvement of complex biochemical processes, complex diseases are considered to be symptoms caused by systemic disorders of these complex networks.

[0003] In biological cells, genes are expressed through transcription and translation, and their expression products can activate or inhibit the expression levels of other genes. This is gene regulation. In different physiological states such as health, disease, and postoperative intervention, the regulatory relationship between genes is different. Therefore, how to analyze the regulatory relationship between genes under different physiological states and discover the differences between them will help to find genes that cause differences in physiological states, that is, biomarkers, which is of great significance for drug research and development. Summary of the invention

[0004] In order to address the deficiencies of the prior art, the present invention provides a biomarker screening method and system based on transfer learning and difference networks; by integrating data from other sources that have certain similarities with the target source, the accuracy of estimation and inference of the gene regulatory network in the target source is improved.

[0005] On the one hand, a biomarker screening method based on transfer learning and difference network is provided, and the method is not applied to the diagnosis and treatment of diseases, and the method comprises: obtaining an auxiliary data set, the auxiliary data set comprising: a first disease group data subset of patients suffering from disease A and a first control group data subset of patients not suffering from disease A; generating a first covariance matrix based on the first disease group data subset of patients suffering from disease A; generating a second covariance matrix based on the first control group data subset of patients not suffering from disease A; and obtaining a difference network corresponding to the auxiliary data set based on sample estimation of the first covariance matrix and the second covariance matrix.

[0006] Acquire a target data set, the target data set comprising: a second disease group data subset of patients suffering from disease A and a second control group data subset of patients not suffering from disease A; generate a third covariance matrix based on the second disease group data subset of patients suffering from disease A; generate a fourth covariance matrix based on the second control group data subset of patients not suffering from disease A; obtain the difference between the difference networks corresponding to the target data set and the auxiliary data set based on sample estimates of the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set.

[0007] The differences between the difference networks are added to the difference network corresponding to the auxiliary data set to obtain the difference network corresponding to the target data set, and the screened biomarkers are determined based on the difference network corresponding to the target data set.

[0008] On the other hand, a biomarker screening system based on transfer learning and difference network is provided, and the system is not used for the diagnosis and treatment of diseases, and the system includes: an auxiliary data set acquisition module, which is configured to: acquire an auxiliary data set, and the auxiliary data set includes: a first disease group data subset of patients suffering from disease A and a first control group data subset of patients not suffering from disease A; generate a first covariance matrix based on the first disease group data subset of patients suffering from disease A; generate a second covariance matrix based on the first control group data subset of patients not suffering from disease A; and obtain a difference network corresponding to the auxiliary data set based on sample estimation of the first covariance matrix and the second covariance matrix.

[0009] The target data set acquisition module is configured to: acquire a target data set, the target data set comprising: a second disease group data subset of patients suffering from disease A and a second control group data subset of patients not suffering from disease A; generate a third covariance matrix based on the second disease group data subset of patients suffering from disease A; generate a fourth covariance matrix based on the second control group data subset of patients not suffering from disease A.

[0010] The difference determination module is configured to obtain the difference between the target data set and the difference network corresponding to the auxiliary data set based on the sample estimation of the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set.

[0011] The screening module is configured to: add the difference between the difference networks with the difference network corresponding to the auxiliary data set to obtain the difference network corresponding to the target data set, and determine the screened biomarker based on the difference network corresponding to the target data set.

[0012] The above technical solution has the following advantages or beneficial effects: The patent of the present invention aims to utilize the idea of ​​directly estimating the difference network, relax the assumption of sparsity in the two covariance matrices or their precision matrices, and instead only assume the sparsity of the difference network between the two networks, and use transfer learning to integrate the data in the auxiliary data source, explore the transfer learning method in the difference network, and provide corresponding theoretical guarantees. The work of this patent has expanded the research and theory of difference networks in the field of transfer learning, and verified that the estimated convergence speed obtained by transfer learning is faster than that of a single sample. Therefore, when there are a sufficient number of auxiliary samples similar to the target data set, the present invention can use the method proposed in the patent of the present invention to estimate the difference network in the target data set. This method not only improves the efficiency of estimation, but also outperforms the existing single data set difference network estimation method in terms of performance.

[0013] Compared with using a single target dataset to estimate its difference network, the method proposed in this patent draws on the difference network information in the auxiliary dataset that has a certain similarity with it to help estimate the difference network in the target dataset. Through certain numerical simulations, the results show that the difference network estimated by the method proposed in this patent has a smaller error and more accurate selection of difference edges. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0015] Figure 1 This is a framework diagram of biomarker difference network analysis under the transfer learning framework of Example 1;

[0016] Figure 2 This is a flow chart of biomarker difference network analysis under the transfer learning framework of Example 1. DETAILED DESCRIPTION

[0017] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0018] In this embodiment, all data is obtained in compliance with laws and regulations and based on the user's consent, and is used legally.

[0019] The focus of the patent research of this invention is not limited to a single network, but includes the differences between different networks in healthy and diseased states. In the biomedical field, differential network analysis has been widely used in various types of networks, such as gene regulatory networks, protein-protein interaction networks, metabolic networks or brain networks. Differential network analysis is particularly important when the research problem is not only focused on a specific network, but also on the changes in the network between different disease states. In fact, this analysis method has become an important supplement to differential expression analysis. The present invention can help scientific researchers screen out biomarkers that affect the development of the current disease for patients who have already suffered from a certain disease.

[0020] Biomarkers refer to biological indicators that exist in organisms and can be measured and evaluated as normal biological processes, pathological processes, or responses to therapeutic interventions. These markers can be molecular, biochemical, physical, or physiological characteristics that reflect health or disease states through their correlation or changes. Biomarkers are widely used in disease diagnosis, disease monitoring, prognosis assessment, treatment effect evaluation, and new drug research and development. Common types of biomarkers include: gene markers; protein markers; imaging markers, etc. In a biomarker network, these markers provide valuable information about the function and state of biological systems through their position and mutual relationship in the network. In this network, biomarkers not only provide information individually, but also help researchers understand more complex biological processes and disease mechanisms through their interconnected relationships in the network, thereby promoting the development of precision medicine and personalized treatment.

[0021] Example 1: This example provides a biomarker screening method based on transfer learning and difference network; Figure 1 and Figure 2 As shown, a biomarker screening method based on transfer learning and difference network includes: S101: obtaining an auxiliary data set, the auxiliary data set including: a first disease group data subset of patients suffering from disease A and a first control group data subset of patients not suffering from disease A; generating a first covariance matrix based on the first disease group data subset of patients suffering from disease A; generating a second covariance matrix based on the first control group data subset of patients not suffering from disease A; and obtaining a difference network corresponding to the auxiliary data set based on sample estimation of the first covariance matrix and the second covariance matrix.

[0022] S102: Obtain a target data set, the target data set comprising: a second disease group data subset of patients suffering from disease A and a second control group data subset of patients not suffering from disease A; generate a third covariance matrix based on the second disease group data subset of patients suffering from disease A; generate a fourth covariance matrix based on the second control group data subset of patients not suffering from disease A; obtain the difference between the difference networks corresponding to the target data set and the auxiliary data set based on sample estimates of the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set.

[0023] S103: Add the differences between the difference networks with the difference network corresponding to the auxiliary data set to obtain the difference network corresponding to the target data set, and determine the screened biomarker based on the difference network corresponding to the target data set.

[0024] Furthermore, the first disease group data subset of patients suffering from disease A includes: a number of samples of patients suffering from disease A, each sample of patients suffering from disease A is provided with a number of biomarkers; the first control group data subset of patients not suffering from disease A includes: a number of samples of patients not suffering from disease A, each sample of patients not suffering from disease A is provided with a number of biomarkers.

[0025] Furthermore, the second disease group data subset of patients who have already suffered from disease A includes: a number of samples of patients who have already suffered from disease A, each of which has a number of biomarkers; the second control group data subset of patients who have not suffered from disease A includes: a number of samples of patients who have not suffered from disease A, each of which has a number of biomarkers. Note: The biomarkers contained in the first disease group, the first control group, the second disease group, and the second control group must be consistent.

[0026] Further, the S101: obtaining an auxiliary data set, the auxiliary data set comprising: a first disease group data subset of patients suffering from disease A and a first control group data subset of patients not suffering from disease A; generating a first covariance matrix based on the first disease group data subset of patients suffering from disease A; generating a second covariance matrix based on the first control group data subset of patients not suffering from disease A; obtaining a difference network corresponding to the auxiliary data set based on sample estimation of the first covariance matrix and the second covariance matrix, specifically comprising: defining two groups of sample data from an auxiliary data set as control groups respectively: and disease groups: ;in ,and and They follow normal distribution and ;in, Indicates the first control group data in the auxiliary data set that does not suffer from disease A. The sample The measured value of a biomarker, Indicates the first disease group data of patients with disease A in the auxiliary data set. The sample The measured value of a biomarker, represents the sample size of the first control group data in the auxiliary data set that does not suffer from disease A, represents the sample size of the first disease group data in the auxiliary data set that has disease A. represents the true mean vector of the biomarkers in the first control group data without disease A in the auxiliary data set ( dimensional vector), represents the true covariance matrix of the biomarkers in the first control group data without disease A in the auxiliary data set ( dimensional matrix), Represents the true mean vector of biomarkers in the first disease group data of patients with disease A in the auxiliary data set ( dimensional vector), Represents the true covariance matrix of the biomarkers in the first disease group data of patients with disease A in the auxiliary data set ( dimensional matrix).

[0027] Based on the real difference network matrix Definition: ;in, and are the inverse matrices of the true covariance matrices of the biomarkers in the first control group and the first disease group data in the auxiliary data set that do not suffer from disease A, that is, the precision matrix, and the following equation exists: .

[0028] Therefore, to estimate the difference network matrix in the auxiliary dataset , optimize the following function: ; ;in, is the adjustment parameter, and are the sample covariance matrices of the two groups of samples in the auxiliary dataset.

[0029] . .

[0030] in, .

[0031] in, represents the sample mean vector of the first control group data without disease A in the auxiliary data set, that is, the vector composed of the sample mean of each biomarker therein, represents the sample mean vector of the first disease group data of patients with disease A in the auxiliary data set, that is, the vector composed of the sample mean of each biomarker therein, Indicates the first control group data in the auxiliary data set that does not suffer from disease A. Sample A vector of biomarkers, Indicates the first disease group data of patients with disease A in the auxiliary data set. Sample A vector of biomarkers.

[0032] Finally, the auxiliary dataset difference network matrix estimated by the optimization problem , Symmetricalization is performed, assuming that the symmetric position elements are and ,and , then the element The value of remains unchanged, and the element The value of the element is changed to The final symmetric difference network matrix ,in , ;,in is an indicator function that takes the value of 1 when the condition in the brackets is met, otherwise it is 0. In this step, only the auxiliary dataset was used, and an initial estimate of the difference network matrix in the auxiliary dataset was made.

[0033] Further, the S102: obtaining a target data set, the target data set comprising: a second disease group data subset of patients with disease A and a second control group data subset of patients without disease A; generating a third covariance matrix based on the second disease group data subset of patients with disease A; generating a fourth covariance matrix based on the second control group data subset of patients without disease A, comprising: defining two groups of sample data from the target data set as control groups: and disease groups: ,in ,and and They follow normal distribution and .

[0034] ; ;

[0035] in, , Indicates the first The sample The measured value of a biomarker, Indicates the first The sample The measured value of a biomarker, represents the sample size of the second control group data in the target data set that does not suffer from disease A, represents the sample size of the second disease group data in the target data set that has disease A. Represents the true mean vector of the biomarkers in the second control group data without disease A in the target data set ( dimensional vector), Represents the true covariance matrix of the biomarkers in the second control group data without disease A in the target data set ( dimensional matrix), Represents the true mean vector of biomarkers in the second disease group data of patients with disease A in the target data set ( dimensional vector), Represents the true covariance matrix of the biomarkers in the second disease group data of the target data set that has disease A ( dimensional matrix), Indicates the second control group data in the target data set that does not have disease A The sample covariance matrix of the biomarkers, Indicates that the second disease group data in the target data set already has disease A The sample covariance matrix of the biomarkers, represents the sample mean vector of the second control group data in the target data set that does not suffer from disease A, that is, the vector composed of the sample mean of each biomarker therein, represents the sample mean vector of the second disease group data of patients with disease A in the target data set, that is, the vector composed of the sample mean of each biomarker therein, Indicates the first Sample A vector of biomarkers, Indicates the first Sample A vector of biomarkers.

[0036] Further, the sample estimation based on the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set obtains the difference between the difference network corresponding to the target data set and the auxiliary data set, including: based on the sample covariance matrix of the control group and the disease group samples in the target data set, and the difference network corresponding to the auxiliary data set, determining the difference between the two difference networks, including: defining the true difference network in the target data set as: ;in, and The inverse matrix of the true covariance matrix of the biomarkers in the second control group and the second disease group data without disease A in the target data set, i.e., the precision matrix; Assuming that the true difference network in the auxiliary data set The real difference network in the target dataset Have similarity, define the similarity measure as , and by constraining To ensure similarity. Based on the initial estimate of the difference network in the auxiliary dataset and the difference matrix between the two datasets in the true difference network , using the data of the target dataset, obtain the difference matrix The estimates are as follows: ;in, is the adjustment parameter, and are the sample covariance matrices of the two groups of samples in the target dataset.

[0037] Finally, the estimated difference matrix It is an asymmetric matrix, so it needs to be symmetric to obtain the difference matrix between the difference networks of the final two data sets (target data set and auxiliary data set) , as follows: For the difference matrix The symmetrical position elements in the and ,and , then the element The value of remains unchanged, and the element The value of the element is changed to The value of , the difference matrix after symmetry , ,in , ,in is an indicator function, which takes the value 1 when the condition in the brackets is met, otherwise it takes the value 0.

[0038] In this step, the difference matrix between the corresponding difference networks in the two datasets is estimated using the target dataset in combination with the corresponding difference network matrix estimated in the previous step S101 using only the auxiliary dataset.

[0039] Further, the step S103: adding the difference between the difference networks to the difference network corresponding to the auxiliary data set to obtain the difference network corresponding to the target data set specifically includes: according to the relationship between the two data sets in the real difference network: , and the above estimated and Compute an estimate of the differential network matrix in the target dataset : ;in, represents the estimation of the difference matrix between the difference networks of the two data sets in the above S102, represents the estimation of the difference network matrix corresponding to the auxiliary data set in S101 above, Represents the difference network matrix corresponding to the real target dataset, represents the difference network matrix corresponding to the real auxiliary dataset, Represents the true difference matrix between the true difference networks corresponding to the two datasets.

[0040] In this step, the difference network matrix corresponding to the auxiliary data set estimated in S101 and the difference matrix between the difference networks corresponding to the two data sets estimated in S102 are used to obtain the final difference network matrix estimate corresponding to the target data set of interest.

[0041] Furthermore, the S103 determines the biomarkers to be screened based on the difference network corresponding to the target data set, specifically including: the larger the value of the element in the difference network matrix corresponding to the optimized target data set, the more the row element, column element and the relationship between the row element and the column element corresponding to the element are likely to be considered as disease-related elements, and thus screened as markers; the row elements and column elements whose element values ​​are greater than a set threshold are screened out, and then the biomarkers corresponding to the row elements, the biomarkers corresponding to the column elements, and the relationship between the two biomarkers are screened out.

[0042] It should be understood that there are many scenarios in which the present invention can be applied, for example: 1. Diabetes research field: Although existing studies have shown that metabolomics biomarkers are associated with the onset of diabetes through traditional univariate statistical methods, with the advancement of analytical technology, metabolomics can simultaneously identify and quantify multiple biomarkers through high-throughput detection, which makes traditional statistical methods no longer applicable, and high-dimensional statistical methods need to be considered to process massive data. In addition, the data of the population of interest is often insufficient, which limits the breadth and depth of the research. In this case, public data can be used for analysis. However, there may be a certain heterogeneity between the population included in the public data and the population of interest in the actual study, which increases the complexity of the analysis. In order to solve these problems, the method of the present invention can be used to use the information in the public data to assist in identifying the differences in the metabolomics biomarker network between the disease group and the normal group in the population of interest. Through this method, metabolomics biomarkers closely related to type 2 diabetes and their interrelationships can be screened out more efficiently and accurately. This not only helps to deeply understand the metabolic mechanism of diabetes, but also provides new clues for early diagnosis and personalized treatment.

[0043] 2. Research field of brain dysfunction: Brain functional network is one of the most commonly used methods to study brain dysfunction. With the advancement of brain functional imaging technology, it is possible to capture the activities of multiple brain regions with high resolution, allowing researchers to use these technologies to characterize brain functional networks in detail. Current studies have shown that abnormal brain functional connectivity is significantly associated with brain dysfunction diseases (such as Alzheimer's disease, Parkinson's disease, epilepsy, etc.). However, brain imaging data is expensive to obtain, and the amount of data collected in a single study is limited, which limits the accuracy of research conclusions. In this case, public brain imaging databases can provide valuable supplementary resources. However, there may be heterogeneity between the subjects in the public data and the population of interest in the actual study, which increases the complexity of the analysis. In order to solve these problems, the method of the present invention can play an important role. It uses information in public data to assist in identifying differences in brain functional connection networks between patients with brain dysfunction and healthy controls in the population of interest. Through this method, brain regions and their functional connections (i.e., functional connections between brain regions) that are closely related to brain dysfunction diseases can be efficiently screened out. This not only helps to deeply understand the neural mechanisms of brain dysfunction diseases, but also provides new insights and clues for early diagnosis and personalized treatment.

[0044] Statistical methods are used to analyze these complex biomarker networks and solve practical problems. In particular, the present invention focuses on how the biomarker network shows structural changes in various disease states.

[0045] Embodiment 2: This embodiment provides a biomarker screening system based on transfer learning and difference network, including: an auxiliary data set acquisition module, which is configured to: acquire an auxiliary data set, the auxiliary data set including: a first disease group data subset of patients with disease A and a first control group data subset of patients without disease A; generate a first covariance matrix based on the first disease group data subset of patients with disease A; generate a second covariance matrix based on the first control group data subset of patients without disease A; obtain a difference network corresponding to the auxiliary data set based on sample estimation of the first covariance matrix and the second covariance matrix; a target data set acquisition module, which is configured to: acquire a target data set, the target data set including: A second disease group data subset suffering from disease A and a second control group data subset not suffering from disease A; generating a third covariance matrix based on the second disease group data subset suffering from disease A; generating a fourth covariance matrix based on the second control group data subset not suffering from disease A; obtaining the difference between the difference network corresponding to the target data set and the auxiliary data set based on sample estimates of the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set; a screening module, which is configured to: add the difference between the difference networks with the difference network corresponding to the auxiliary data set to obtain the difference network corresponding to the target data set, and determine the screened biomarker based on the difference network corresponding to the target data set.

[0046] Further, an auxiliary data set is obtained, the auxiliary data set comprising: a first disease group data subset of patients with disease A and a first control group data subset of patients without disease A; a first covariance matrix is ​​generated based on the first disease group data subset of patients with disease A; a second covariance matrix is ​​generated based on the first control group data subset of patients without disease A; based on sample estimation of the first covariance matrix and the second covariance matrix, a difference network corresponding to the auxiliary data set is obtained, specifically comprising: based on the true difference network matrix Definition: ;in, and are the inverse matrices of the true covariance matrices of the biomarkers in the first control group and the first disease group data that do not suffer from disease A in the auxiliary data set, respectively, and the following equation exists: ; Therefore, to estimate the difference network matrix in the auxiliary dataset , optimize the following function: ; ;in, is the adjustment parameter, and They are the sample covariance matrices of the two groups of samples in the auxiliary data set: ;

[0047] ;

[0048] in, ;in, represents the sample mean vector of the first control group data without disease A in the auxiliary data set, that is, the vector composed of the sample mean of each biomarker therein, represents the sample mean vector of the first disease group data of patients with disease A in the auxiliary data set, that is, the vector composed of the sample mean of each biomarker therein, Indicates the first control group data in the auxiliary data set that does not suffer from disease A. Sample A vector of biomarkers, Indicates the first disease group data of patients with disease A in the auxiliary data set. Sample biomarkers; finally, the auxiliary dataset difference network matrix estimated by the optimization problem , Symmetricalization is performed, assuming that the symmetric position elements are and ,and , then the element The value of remains unchanged, and the element The value of the element is changed to The final symmetric difference network matrix ;

[0049] in, , ;

[0050] in, is an indicator function, which takes the value 1 when the condition in the brackets is met, otherwise it takes the value 0.

[0051] Further, a target data set is obtained, the target data set comprising: a second disease group data subset of patients with disease A and a second control group data subset of patients without disease A; a third covariance matrix is ​​generated based on the second disease group data subset of patients with disease A; a fourth covariance matrix is ​​generated based on the second control group data subset of patients without disease A, including: defining two groups of sample data from the target data set as control groups: and disease groups: ,in ,and and They follow normal distribution and ; ; ;in, , Indicates the first The sample The measured value of a biomarker, Indicates the first The sample The measured value of a biomarker, represents the sample size of the second control group data in the target data set that does not suffer from disease A, represents the sample size of the second disease group data in the target data set that has disease A. represents the true mean vector of the biomarkers in the second control group data without disease A in the target data set, represents the true covariance matrix of the biomarkers in the second control group data without disease A in the target dataset, represents the true mean vector of biomarkers in the second disease group data of patients with disease A in the target dataset, represents the true covariance matrix of the biomarkers in the second disease group data of patients with disease A in the target dataset, Indicates the second control group data in the target data set that does not have disease A The sample covariance matrix of the biomarkers, Indicates that the second disease group data in the target data set already has disease A The sample covariance matrix of the biomarkers, represents the sample mean vector of the second control group data in the target data set that does not suffer from disease A, that is, the vector composed of the sample mean of each biomarker therein, represents the sample mean vector of the second disease group data of patients with disease A in the target data set, that is, the vector composed of the sample mean of each biomarker therein, Indicates the first Sample A vector of biomarkers, Indicates the first Sample A vector of biomarkers.

[0052] Further, the sample estimation based on the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set obtains the difference between the difference network corresponding to the target data set and the auxiliary data set, including: based on the sample covariance matrix of the control group and the disease group samples in the target data set, and the difference network corresponding to the auxiliary data set, determining the difference between the two difference networks, including: defining the true difference network in the target data set as: ;in, and The inverse matrix of the true covariance matrix of the biomarkers in the second control group and the second disease group data without disease A in the target data set, i.e., the precision matrix; Assuming that the true difference network in the auxiliary data set The real difference network in the target dataset Have similarity, define the similarity measure as , and by constraining To ensure similarity; based on the initial estimate of the difference network in the auxiliary dataset and the difference matrix between the two datasets in the true difference network , using the data of the target dataset, obtain the difference matrix The estimates are as follows: ;in, is the adjustment parameter, and are the sample covariance matrices of the two groups of samples in the target data set; finally, the estimated difference matrix It is an asymmetric matrix, so it needs to be symmetric to obtain the difference matrix between the difference networks of the final target dataset and the auxiliary dataset , as follows: For the difference matrix The symmetrical position elements in the and ,and , then the element The value of remains unchanged, and the element The value of the element is changed to The value of , the difference network matrix after symmetry ; , ,in is an indicator function, which takes a value of 1 when the condition in the brackets is met, otherwise it takes a value of 0. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A biomarker screening method based on transfer learning and difference network, which is not used for the diagnosis and treatment of diseases, characterized in that: The method comprises: Acquire an auxiliary data set, the auxiliary data set comprising: a first disease group data subset of patients suffering from disease A and a first control group data subset of patients not suffering from disease A; generate a first covariance matrix based on the first disease group data subset of patients suffering from disease A; generate a second covariance matrix based on the first control group data subset of patients not suffering from disease A; obtain a difference network corresponding to the auxiliary data set based on sample estimation of the first covariance matrix and the second covariance matrix; The sample estimation based on the first covariance matrix and the second covariance matrix obtains a difference network corresponding to the auxiliary data set, specifically including: Based on the real difference network matrix Definition: ; in, and are the inverse matrices of the true covariance matrices of the biomarkers in the first control group without disease A and the first disease group with disease A in the auxiliary data set, respectively, and the following equation exists: ; Therefore, to estimate the difference network matrix in the auxiliary dataset , optimize the following function: ; ; in, is the adjustment parameter, and They are the sample covariance matrices of the two groups of samples in the auxiliary data set: ; ; in, ; represents the sample mean vector of the first control group data without disease A in the auxiliary data set, that is, the vector composed of the sample mean of each biomarker therein, represents the sample mean vector of the first disease group data of patients with disease A in the auxiliary data set, that is, the vector composed of the sample mean of each biomarker therein, Indicates the first control group data in the auxiliary data set that does not suffer from disease A. Sample A vector of biomarkers, Indicates the first disease group data of patients with disease A in the auxiliary data set. Sample A vector of biomarkers, represents the sample size of the first control group data in the auxiliary data set that does not suffer from disease A, Represents the sample size of the first disease group data with disease A in the auxiliary data set; Finally, the auxiliary dataset difference network matrix estimated by the optimization problem is , Symmetricalization is performed, assuming that the symmetric position elements are and ,and , then the element The value of remains unchanged, and the element The value of the element is changed to The value of , the final symmetric difference network matrix: , , in, , ; is an indicator function, which takes the value 1 when the condition in the brackets is met, otherwise it takes the value 0; Acquire a target data set, the target data set comprising: a second disease group data subset of patients suffering from disease A and a second control group data subset of patients not suffering from disease A; generate a third covariance matrix based on the second disease group data subset of patients suffering from disease A; generate a fourth covariance matrix based on the second control group data subset of patients not suffering from disease A; obtain a difference between the difference networks corresponding to the target data set and the auxiliary data set based on sample estimates of the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set; The sample estimation based on the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set obtains the difference between the difference network corresponding to the target data set and the auxiliary data set, including: Based on the sample covariance matrix of the control group and the disease group in the target data set, and the difference network corresponding to the auxiliary data set, the difference between the two difference networks is determined, including: The true difference network in the target dataset is defined as: ;in, and The inverse matrices of the true covariance matrices of the biomarkers in the second control group and the second disease group data without disease A in the target data set, i.e., the precision matrix; Assuming the true difference network in the auxiliary dataset The real difference network in the target dataset Have similarity, define the similarity measure as , and by constraining To ensure similarity; Initial estimation of the difference network based on auxiliary dataset and the difference matrix between the two datasets in the true difference network , using the data of the target dataset, obtain the difference matrix The estimates are as follows: ; in, is the adjustment parameter, and are the sample covariance matrices of the two groups of samples in the target data set; finally, the estimated difference matrix It is an asymmetric matrix, so it needs to be symmetric to obtain the difference matrix between the difference networks of the final target dataset and the auxiliary dataset , as follows: For the difference matrix The symmetrical position elements in the and ,and , then the element The value of remains unchanged, and the element The value of the element is changed to The value of , the difference matrix after symmetry , ; , ,in is an indicator function, which takes the value 1 when the condition in the brackets is met, otherwise it takes the value 0; The differences between the difference networks are added to the difference network corresponding to the auxiliary data set to obtain the difference network corresponding to the target data set, and the screened biomarkers are determined based on the difference network corresponding to the target data set.

2. The biomarker screening method based on transfer learning and difference network according to claim 1, characterized in that: Obtain a target data set, the target data set comprising: a second disease group data subset of patients who have suffered from disease A and a second control group data subset of patients who do not suffer from disease A; generate a third covariance matrix based on the second disease group data subset of patients who have suffered from disease A; generate a fourth covariance matrix based on the second control group data subset of patients who do not suffer from disease A, comprising: Define the two groups of sample data from the target data set as control groups: and disease groups: ,in ,and and They follow normal distribution and ; ; ; in, , Indicates the first The sample The measured value of a biomarker, Indicates the first The sample The measured value of a biomarker, represents the sample size of the second control group data in the target data set that does not suffer from disease A, represents the sample size of the second disease group data in the target data set that has disease A. represents the true mean vector of the biomarkers in the second control group data without disease A in the target data set, represents the true covariance matrix of the biomarkers in the second control group data without disease A in the target dataset, represents the true mean vector of biomarkers in the second disease group data of patients with disease A in the target dataset, represents the true covariance matrix of the biomarkers in the second disease group data of patients with disease A in the target dataset, Indicates the second control group data in the target data set that does not have disease A The sample covariance matrix of the biomarkers, Indicates that the second disease group data in the target data set already has disease A The sample covariance matrix of the biomarkers, represents the sample mean vector of the second control group data in the target data set that does not suffer from disease A, that is, the vector composed of the sample mean of each biomarker therein, represents the sample mean vector of the second disease group data of patients with disease A in the target data set, that is, the vector composed of the sample mean of each biomarker therein, Indicates the first Sample A vector of biomarkers, Indicates the first Sample A vector of biomarkers.

3. The biomarker screening method based on transfer learning and difference network according to claim 1, characterized in that: The difference between the difference networks is added to the difference network corresponding to the auxiliary data set to obtain the difference network corresponding to the target data set, specifically including: According to the relationship between the two data sets in the real difference network: , and the above estimated and Compute an estimate of the differential network matrix in the target dataset : ; in, represents the estimation of the difference matrix between the difference networks of two datasets, represents the estimation of the difference network matrix corresponding to the auxiliary dataset, Represents the difference network matrix corresponding to the real target dataset, represents the difference network matrix corresponding to the real auxiliary dataset, Represents the true difference matrix between the true difference networks corresponding to the two datasets.

4. The biomarker screening method based on transfer learning and difference network according to claim 1, characterized in that: Based on the difference network corresponding to the target data set, the biomarkers to be screened are determined, specifically including: screening out row elements and column elements whose element values ​​are greater than a set threshold, and then screening out the biomarkers corresponding to the row elements, the biomarkers corresponding to the column elements, and the relationship between the two biomarkers.

5. A biomarker screening system based on transfer learning and difference networks, the system being not applicable to the diagnosis and treatment of diseases, characterized in that, The system comprises: The auxiliary data set acquisition module is configured to: acquire an auxiliary data set, the auxiliary data set comprising: a first disease group data subset of patients suffering from disease A and a first control group data subset of patients not suffering from disease A; generate a first covariance matrix based on the first disease group data subset of patients suffering from disease A; generate a second covariance matrix based on the first control group data subset of patients not suffering from disease A; obtain a difference network corresponding to the auxiliary data set based on sample estimation of the first covariance matrix and the second covariance matrix; The sample estimation based on the first covariance matrix and the second covariance matrix obtains a difference network corresponding to the auxiliary data set, specifically including: Based on the real difference network matrix Definition: ;in, and are the inverse matrices of the true covariance matrices of the biomarkers in the first control group without disease A and the first disease group with disease A in the auxiliary data set, respectively, and the following equation exists: ; Therefore, to estimate the difference network matrix in the auxiliary dataset , optimize the following function: ; ;in, is the adjustment parameter, and They are the sample covariance matrices of the two groups of samples in the auxiliary data set: ; ; in, ;in, represents the sample mean vector of the first control group data without disease A in the auxiliary data set, that is, the vector composed of the sample mean of each biomarker therein, represents the sample mean vector of the first disease group data of patients with disease A in the auxiliary data set, that is, the vector composed of the sample mean of each biomarker therein, Indicates the first control group data in the auxiliary data set that does not suffer from disease A. Sample A vector of biomarkers, Indicates the first disease group data of patients with disease A in the auxiliary data set. Sample A vector of biomarkers, represents the sample size of the first control group data in the auxiliary data set that does not suffer from disease A, Represents the sample size of the first disease group data with disease A in the auxiliary data set; Finally, the auxiliary dataset difference network matrix estimated by the optimization problem is , Symmetricalization is performed, assuming that the symmetric position elements are and ,and , then the element The value of remains unchanged, and the element The value of the element is changed to The final symmetric difference network matrix , ,in, , ;in is an indicator function, which takes the value 1 when the condition in the brackets is met, otherwise it takes the value 0; The target data set acquisition module is configured to: acquire a target data set, the target data set comprising: a second disease group data subset of patients who have suffered from disease A and a second control group data subset of patients who have not suffered from disease A; generate a third covariance matrix based on the second disease group data subset of patients who have suffered from disease A; generate a fourth covariance matrix based on the second control group data subset of patients who have not suffered from disease A; A difference determination module is configured to: obtain a difference between the target data set and the difference network corresponding to the auxiliary data set based on sample estimates of the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set; The sample estimation based on the third covariance matrix and the fourth covariance matrix and the difference network corresponding to the auxiliary data set obtains the difference between the difference network corresponding to the target data set and the auxiliary data set, including: Based on the sample covariance matrix of the control group and the disease group in the target data set, and the difference network corresponding to the auxiliary data set, the difference between the two difference networks is determined, including: The true difference network in the target dataset is defined as: ; in, and The inverse matrices of the true covariance matrices of the biomarkers in the second control group and the second disease group data without disease A in the target data set, i.e., the precision matrix; Assuming the true difference network in the auxiliary dataset The real difference network in the target dataset Have similarity, define the similarity measure as , and by constraining To ensure similarity; Initial estimation of the difference network based on auxiliary dataset and the difference matrix between the two datasets in the true difference network , using the data of the target dataset, obtain the difference matrix The estimates are as follows: ; in, is the adjustment parameter, and are the sample covariance matrices of the two groups of samples in the target data set; finally, the estimated difference matrix It is an asymmetric matrix, so it needs to be symmetric to obtain the difference matrix between the difference networks of the final target dataset and the auxiliary dataset , as follows: For the difference matrix The symmetrical position elements in the and ,and , then the element The value of remains unchanged, and the element The value of the element is changed to The value of , the difference matrix after symmetry , ; , ,in is an indicator function, which takes the value 1 when the condition in the brackets is met, otherwise it takes the value 0; The screening module is configured to: add the difference between the difference networks with the difference network corresponding to the auxiliary data set to obtain the difference network corresponding to the target data set, and determine the screened biomarker based on the difference network corresponding to the target data set.

6. The biomarker screening system based on transfer learning and difference network as claimed in claim 5, characterized in that: Obtain a target data set, the target data set comprising: a second disease group data subset of patients who have suffered from disease A and a second control group data subset of patients who do not suffer from disease A; generate a third covariance matrix based on the second disease group data subset of patients who have suffered from disease A; generate a fourth covariance matrix based on the second control group data subset of patients who do not suffer from disease A, comprising: Define the two groups of sample data from the target data set as control groups: and disease groups: ; in ,and and They follow normal distribution and ; ; ; in, , Indicates the first The sample The measured value of a biomarker, Indicates the first The sample The measured value of a biomarker, represents the sample size of the second control group data in the target data set that does not suffer from disease A, represents the sample size of the second disease group data in the target data set that has disease A. represents the true mean vector of the biomarkers in the second control group data without disease A in the target data set, represents the true covariance matrix of the biomarkers in the second control group data without disease A in the target dataset, represents the true mean vector of biomarkers in the second disease group data of patients with disease A in the target dataset, represents the true covariance matrix of the biomarkers in the second disease group data of patients with disease A in the target dataset, Indicates the second control group data in the target data set that does not have disease A The sample covariance matrix of the biomarkers, Indicates that the second disease group data in the target data set already has disease A The sample covariance matrix of the biomarkers, represents the sample mean vector of the second control group data in the target data set that does not suffer from disease A, that is, the vector composed of the sample mean of each biomarker therein, represents the sample mean vector of the second disease group data of patients with disease A in the target data set, that is, the vector composed of the sample mean of each biomarker therein, Indicates the first Sample A vector of biomarkers, Indicates the first Sample A vector of biomarkers.