Alzheimer's disease image genetics data analysis method

By introducing a sparse typical correlation analysis algorithm with autocorrelation constraints in the analysis of Alzheimer's imaging genetic data, the problem of autocorrelation not being considered in the existing technology is solved, and the accuracy of auxiliary diagnosis of Alzheimer's disease is improved , reveals the correlation between imaging and genetics.

CN119067908APending Publication Date: 2024-12-03SHANGHAI INST OF BIOLOGICAL SCI CHINESE ACAD OF SCI +2
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202310628820.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-30
Publication Date
2024-12-03

Smart Images

  • Figure CN119067908A_ABST
    Figure CN119067908A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of biological information, in particular to an Alzheimer's disease image genetics data analysis method. The method comprises the following steps: acquiring structural magnetic resonance imaging and gene expression data of samples of the Alzheimer's disease and a control group; extracting the gray matter volume of the brain region as brain region imaging feature data; dividing into a training set and a test set; performing differential expression analysis processing on the gene expression data; based on a sparse canonical correlation analysis algorithm guided by prior information, performing correlation analysis on the brain region imaging feature data and the gene expression data to obtain a corresponding weight; sorting the brain region imaging feature data and the gene expression data according to corresponding weights to obtain biomarkers; and constructing a diagnostic model of the Alzheimer's disease for auxiliary diagnostic analysis. According to the method, the biomarkers which are related to the Alzheimer's disease and have diagnostic significance in the image genetics data can be deeply mined, so that the development of Alzheimer's disease treatment targets is assisted.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and more specifically, to an Alzheimer's disease imaging genetics data analysis method guided by prior information. Background Art

[0002] Alzheimer's disease is a neurodegenerative disease characterized by severe memory loss, brain atrophy, personality and behavioral changes, etc. In the late stage of Alzheimer's disease, patients will lose the ability to take care of themselves, which will eventually endanger their lives. The incidence of Alzheimer's disease is increasing year by year, and there is currently no effective treatment.

[0003] Imaging genetics data analysis refers to the correlation analysis of imaging genomics data and genetic data, explaining the macroscopic changes in the brain from a genetic perspective and building a bridge between genotype and phenotype.

[0004] In recent years, many scholars have integrated imaging genetics data through canonical correlation analysis. The canonical correlation analysis algorithm can find the linear combination that maximizes the correlation between two data sets, and the linear combination coefficients of the two data sets are called weights u and v, respectively.

[0005] Due to the high dimensionality of imaging genetics data characteristics, some scholars have added sparse constraints on weights u and v based on canonical correlation analysis to help the algorithm perform feature selection in high-dimensional data.

[0006] Hao Xiaoke et al. expanded the sparse canonical correlation analysis algorithm to three dimensions, embedded the clinical information of the samples in another dimension, and proposed a three-way sparse canonical correlation analysis method, which can solve the problem that the existing two-variable method cannot embed prior information.

[0007] Taking into account the connectivity information between brain regions and genes, Kim et al. proposed a sparse canonical correlation analysis method based on joint connectivity, and applied it to the association analysis of multimodal brain imaging and single nucleotide polymorphisms, achieving good results.

[0008] However, the above-mentioned canonical correlation analysis method does not take into account the autocorrelation of each modality data, nor does it incorporate this correlation as a priori information into the analysis of Alzheimer's disease imaging genetics data, which may affect the association performance of the algorithm and subsequently affect the accuracy of auxiliary diagnosis analysis of Alzheimer's disease. Summary of the invention

[0009] The purpose of the present invention is to provide a method for analyzing Alzheimer's disease imaging genetics data to solve the problem in the prior art that it is difficult to perform accurate auxiliary diagnosis and analysis of Alzheimer's disease using Alzheimer's disease imaging genetics data.

[0010] In order to achieve the above object, the present invention provides a method for analyzing Alzheimer's disease imaging genetics data, comprising the following steps:

[0011] Step S1, obtaining structural magnetic resonance imaging and gene expression data of Alzheimer's disease and control group samples;

[0012] Step S2, preprocessing the structural magnetic resonance imaging data, and extracting the gray matter volume of the brain region as brain region imaging feature data;

[0013] Step S3: Divide all samples into training set and test set according to a certain ratio;

[0014] Step S4, performing differential expression analysis on the gene expression data in the training set to obtain processed gene expression data;

[0015] Step S5: Based on the sparse canonical correlation analysis algorithm guided by prior information, correlation analysis is performed on the processed brain region imaging feature data and gene expression data of the training set to obtain corresponding weights;

[0016] Step S6, sorting the brain region imaging feature data and the gene expression data according to the corresponding weights, and obtaining the brain region imaging feature data and the gene expression data within a preset weight value range as biomarkers;

[0017] Step S7: constructing a diagnostic model for Alzheimer's disease based on the biomarkers, wherein the diagnostic model is used for auxiliary diagnosis and analysis of Alzheimer's disease.

[0018] In one embodiment, the step S1 further comprises:

[0019] The structural magnetic resonance imaging data of the Alzheimer's disease and control group samples are in nii format;

[0020] The gene expression data of the Alzheimer's disease and control group samples are tabular data without missing values.

[0021] In one embodiment, the step S2 of preprocessing the structural magnetic resonance imaging data further comprises:

[0022] Affine registration of raw structural MRI data to the MNI standard space;

[0023] The structural MRI data of the training set were registered to the AAL template and feature extraction was performed.

[0024] In one embodiment, the step S3 further comprises:

[0025] 80% of all samples are used as training sets and 20% as test sets;

[0026] The training set is used for association analysis and screening of biomarkers, and the test set is used to test the performance of the diagnostic model.

[0027] In one embodiment, the step S4 further comprises:

[0028] The Limma algorithm was used to perform differential expression analysis on gene expression data;

[0029] The expression levels of differentially expressed genes whose algorithm parameter p was less than the preset threshold were retained as processed gene expression data for subsequent association analysis.

[0030] In one embodiment, the step S5 further comprises:

[0031] The autocorrelation-based penalty function is introduced into the sparse canonical correlation analysis model to study the correlation between brain regions and genes. The corresponding objective function is:

[0032]

[0033]

[0034] Among them, u represents the norm vector of X, and v represents the norm vector of Y;

[0035] λ1,λ2,β1,β2 are regularization parameters;

[0036] Given a dataset X∈R n×p , Y∈R n×q They correspond to structural magnetic resonance imaging data and gene expression data, respectively, and n represents the number of samples.

[0037] In one embodiment, the step S5 further comprises:

[0038] The alternating convex search method is used to solve the objective function and obtain the optimal values ​​of u and v;

[0039] The alternating convex search method further comprises:

[0040] Step S51, initializing u and v;

[0041] Step S52: fix u and solve v. The expression corresponding to the objective function of v is as follows:

[0042]

[0043]

[0044] Step S53: fix v and solve for u. The expression corresponding to the objective function of u is as follows:

[0045]

[0046]

[0047] Step S54, repeat step S52 and step S53 until the convergence condition is met, and obtain the corresponding weight vectors u and v.

[0048] In one embodiment, the step S6 further comprises:

[0049] The brain imaging feature data and gene expression data are sorted according to the corresponding weight absolute values, and the first several features with the largest weight absolute values ​​of the brain imaging feature data and gene expression data are obtained as biomarkers for subsequent auxiliary diagnosis analysis.

[0050] In one embodiment, the step S7 further comprises:

[0051] The first several features with the largest absolute value ranking based on brain imaging feature data and gene expression data were used as biomarkers, and the constructed diagnostic model was verified using the logistic regression algorithm.

[0052] The Alzheimer's disease imaging genetics data analysis method provided by the present invention can perform effective association analysis and feature selection in high-dimensional imaging genetics data on the basis of integrating prior information, explore potential biomarkers of Alzheimer's disease, and further analyze the relationship between genetic variation and internal brain morphology, so as to assist in the diagnosis of Alzheimer's disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The above and other features, properties and advantages of the present invention will become more apparent through the following description in conjunction with the accompanying drawings and embodiments, in which the same reference numerals always represent the same features, wherein:

[0054] Figure 1 A flow chart of a method for analyzing Alzheimer's disease imaging genetics data based on prior information guidance according to an embodiment of the present invention is disclosed;

[0055] Figure 2 A line graph of a sparse canonical correlation analysis algorithm in a parameter selection process according to an embodiment of the present invention is disclosed;

[0056] Figure 3 A visualization heat map of weight u and weight v according to an embodiment of the present invention is disclosed;

[0057] Figure 4 A visualized heat map of the first 10 brain regions according to an embodiment of the present invention is disclosed;

[0058] Figure 5aA diagnostic performance validation curve diagram of markers of the first 10 brain regions according to an embodiment of the present invention is disclosed;

[0059] Figure 5b A diagnostic performance validation curve diagram of the top 10 gene markers according to an embodiment of the present invention is disclosed. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is 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 used to explain the invention and are not used to limit the invention.

[0061] The protection scope of the Alzheimer's disease imaging genetics data analysis method based on prior information guidance described in the present invention is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing, or replacing steps in the prior art based on the principles of the present invention are included in the protection scope of the present invention.

[0062] Figure 1 A flow chart of a method for analyzing Alzheimer's disease imaging genetics data based on prior information guidance according to an embodiment of the present invention is disclosed, such as Figure 1 As shown, a method for analyzing Alzheimer's disease imaging genetics data proposed by the present invention comprises the following steps:

[0063] Step S1, obtaining structural magnetic resonance imaging and gene expression data of Alzheimer's disease and control group samples;

[0064] Step S2, preprocessing the structural magnetic resonance imaging data, and extracting the gray matter volume of the brain region as brain region imaging feature data;

[0065] Step S3: Divide all samples into training set and test set according to a certain ratio;

[0066] Step S4, performing differential expression analysis on the gene expression data in the training set to obtain processed gene expression data;

[0067] Step S5: Based on the sparse canonical correlation analysis algorithm guided by prior information, correlation analysis is performed on the processed brain region imaging feature data and gene expression data of the training set to obtain corresponding weights;

[0068] Step S6, sorting the brain region imaging feature data and the gene expression data according to the corresponding weights, and obtaining the brain region imaging feature data and the gene expression data within a preset weight value range as biomarkers;

[0069] Step S7: constructing a diagnostic model for Alzheimer's disease based on biomarkers for auxiliary diagnosis and analysis of Alzheimer's disease.

[0070] The present invention proposes an Alzheimer's disease imaging genetics data analysis method, which is used to discover the possible biological associations and mechanisms between Alzheimer's disease gene expression and brain regions. The autocorrelation matrix of imaging genetics data is embedded in a canonical correlation analysis algorithm through autocorrelation constraints. In order to facilitate feature selection in high-dimensional features, sparse constraints corresponding to weight vectors of imaging genomics and genetic features are added. In addition, in order to verify the feature selection ability of the algorithm, a diagnostic model for Alzheimer's disease is constructed using the top-ranked brain regions and gene data selected, and the structural magnetic resonance imaging data and gene expression data of Alzheimer's disease patients and a control group in the Alzheimer's disease NI database are used as examples to verify the effectiveness of the method.

[0071] The following is a detailed description of these steps of the Alzheimer's disease imaging genetics data analysis method proposed in the present invention. It should be understood that within the scope of the present invention, the above-mentioned technical features of the present invention and the technical features specifically described below (such as the embodiments) can be combined with each other and interrelated to form a preferred technical solution.

[0072] Step S1, obtaining structural magnetic resonance imaging and gene expression data of Alzheimer's disease and control group samples.

[0073] In this embodiment, the Alzheimer's disease includes mild cognitive impairment.

[0074] In this embodiment, the structural magnetic resonance imaging data of the Alzheimer's disease samples and the control samples are in nii format.

[0075] The nii format (an extension of the NIFTI format) was invented for multidimensional neuroimaging and can truly reflect metadata, including orientation information.

[0076] In this embodiment, the gene expression data is tabular data without missing values.

[0077] Each row of the table data represents a patient sample, and each column represents a feature of the patient sample, namely, the expression level of a gene.

[0078] Step S2: preprocess the structural magnetic resonance imaging data and extract the gray matter volume of the brain region as brain region imaging feature data.

[0079] Preprocessing the structural magnetic resonance imaging data further includes:

[0080] Affine registration of the raw data of structural MRI was performed to achieve head motion correction and register it to the MNI standard space;

[0081] The structural MRI data of the training set were registered to the AAL template and feature extraction was performed.

[0082] The full name of AAL is Anatomical Automatic Labeling. The AAL partition is provided by the Montreal Neurological Institute (MNI). The AAL template has a total of 116 areas.

[0083] Furthermore, the structural MRI raw image data was corrected for head motion using the DiffusionKit (diffusion MRI analysis software) toolkit through affine registration method, and then registered to the MNI standard space.

[0084] Furthermore, the CAT toolkit based on MATLAB software segmented the structural MRI data by registering the images to the AAL template, thereby achieving feature extraction of the structural MRI data of the training set.

[0085] In this embodiment, by registering the image to the AAL template, the gray matter volumes of the brain regions of 90 regions of interest were obtained as brain region imaging feature data, and the volumes were organized into tabular data.

[0086] Furthermore, the CAT12 toolkit of Matlab (https: / / neuro-jena.github.io / cat / ) was used to extract features from the structural MRI data in nii format. The CAT12 toolkit provides voxel-based morphometry (VBM) function.

[0087] Step S3: Divide all samples into training set and test set according to a certain ratio.

[0088] The samples are divided into training set and test set in a ratio of 8:2.

[0089] In this embodiment, the samples are divided into a training set and a test set in a ratio of 8:2, further comprising:

[0090] According to the sample labels, 80% of the imaging genetics data corresponding to all samples were used as the training set, and 20% of the imaging genetics data corresponding to all samples were used as the test set;

[0091] Among them, the training set is used for association analysis and screening of biomarkers, and the test set is used to test the performance of the diagnostic model.

[0092] In this example, a total of 210 samples were collected, including 148 diseased samples and 62 control samples.

[0093] The samples were divided into 168 training set samples (118 diseased samples and 50 control samples) and 42 test set samples (30 diseased samples and 12 control samples) in a ratio of 8:2.

[0094] Step S4: performing differential expression analysis on the gene expression data in the training set to obtain processed gene expression data.

[0095] In this embodiment, differential expression analysis is performed on the gene expression data of the training set, including:

[0096] The Limma (Linear Models for Microarray and RNA-Seq Data) algorithm was used to perform differential expression analysis on the Alzheimer's disease group samples and the control group samples in the training set to obtain the expression profiles corresponding to the differentially expressed genes;

[0097] The expression levels of differentially expressed genes whose algorithm parameter p was less than the preset threshold were retained as the processed gene expression data and organized into tabular data for subsequent association analysis.

[0098] In this embodiment, the algorithm parameter p is less than a preset threshold, further including: the algorithm parameter p<0.05.

[0099] In this example, 413 genes with significant differences between the diseased group and the control group with p<0.05 were finally retained.

[0100] Step S5: Based on the sparse canonical correlation analysis algorithm guided by prior information, correlation analysis is performed on the processed brain region imaging feature data and gene expression data of the training set to obtain corresponding weights.

[0101] Association analysis was performed on the two processed table data (one sample per row, one brain region / gene per column).

[0102] In this embodiment, a sparse canonical correlation analysis algorithm guided by prior information performs correlation analysis on two types of data in a training set, and obtains autocorrelated information from structural magnetic resonance imaging data and gene expression data.

[0103] The sparse canonical correlation analysis algorithm guided by prior information is an innovative algorithm based on canonical correlation analysis. The sparse canonical correlation analysis algorithm is introduced in the association analysis. The basic idea of ​​the sparse canonical correlation analysis algorithm is to obtain the linear combination that makes the two sets of data reach the maximum correlation. The output of the sparse canonical correlation analysis algorithm is the weight vectors u and v of the characteristics of the two data.

[0104] Canonical correlation analysis is a method for determining the maximum correlation between two data sets.

[0105] Given a dataset X∈R n×p , Y∈R n×q .

[0106] Among them, X represents the p feature of the imaging data, Y represents the q feature of the genetic data, and there are n samples in total.

[0107] The canonical correlation analysis (CCA) algorithm can be used to reduce the dimensionality of high-dimensional data X, Y to find a linear combination of the feature weights of X and Y to maximize the correlation between the two data.

[0108] The objective function of the canonical correlation analysis algorithm is expressed as follows

[0109]

[0110] Stu T X T Xu=v T Y T Yv=1 (1)

[0111] It is assumed that X and Y are z-score standardized with zero mean and unit variance.

[0112] u and v are the corresponding norm vectors.

[0113] For the canonical correlation analysis algorithm, an alternating iterative algorithm is used to update u and v separately.

[0114] Fix v first, then update u.

[0115] Specifically, to minimize equation (1), the derivative of this objective with respect to u is taken and set to zero, which yields equation (2).

[0116]

[0117] Then fix u and update v. The corresponding expression is as follows:

[0118]

[0119] The traditional canonical correlation analysis algorithm cannot meet the needs of association analysis between high-dimensional data. Therefore, we can impose l1 norm constraints on the weight vectors u and v to help the algorithm perform association analysis in high-dimensional features. This method is called sparse canonical correlation analysis algorithm.

[0120] Furthermore, considering that prior knowledge can effectively improve the correlation between data, the sparse canonical correlation analysis algorithm based on autocorrelation constraints proposed in the present invention adds the autocorrelation matrices of the two data as prior information to the canonical correlation analysis algorithm, which can improve the anti-noise performance of the algorithm and obtain biologically meaningful results; orthogonal constraints on the weight vectors u and v are also added to prevent redundant features from affecting the results; finally, the corresponding weight vectors u and v of the two modal data features can be output.

[0121] In this embodiment, the sparse canonical correlation analysis algorithm based on autocorrelation constraints proposed by the present invention introduces sparse constraints and autocorrelation constraints on u and v on the basis of the canonical correlation analysis algorithm.

[0122] Among them, the sparse constraint is used to control the sparsity of the weight vector and facilitate feature selection in high-dimensional features.

[0123] The autocorrelation constraint takes the correlation between the two different modal features as prior information, which can make the results of the association analysis more accurate.

[0124] Furthermore, an autocorrelation-based penalty function is introduced into the sparse canonical correlation analysis model to study the correlation between brain regions and genes. The corresponding objective function after the introduction is:

[0125]

[0126]

[0127] Among them, u represents the norm vector of X, and v represents the norm vector of Y;

[0128] λ1,λ2,β1,β2 are regularization parameters;

[0129] Given a dataset X∈R n×p , Y∈R n×q They correspond to structural magnetic resonance imaging data and gene expression data, respectively, and n represents the number of samples.

[0130] In order to keep the model sparse, it is important to use the l1 penalty controlled by β1 and β2;

[0131] The proposed autocorrelation matrix is ​​controlled by λ1 and λ2, and I is the identity matrix.

[0132] The autocorrelation matrix is ​​used to reflect the correlation strength between different features of the same modality data, inducing the algorithm to select features with stronger correlation for association.

[0133] In this embodiment, an alternating convex search method is used to solve the iterative rule for minimizing the objective function (4) to obtain the optimal values ​​of u and v.

[0134] The alternating convex search method first initializes u and v, then fixes u and solves v, and vice versa, and repeats the above process until convergence.

[0135] In this embodiment, the alternating convex search method further includes:

[0136] Step S51, initializing u and v;

[0137] Step S52: fix u and solve v. When u is fixed and v is updated, the objective function of v can be rewritten. The expression corresponding to the objective function of v is as follows:

[0138]

[0139]

[0140] When u is constant, v is minimized.

[0141] Since the autocorrelation matrix is ​​a positive definite matrix, the connectivity-based penalty is convex, that is, equation (5) is convex and a soft threshold-based optimization can be used. Therefore, the coordinate solution of (5) is defined as:

[0142]

[0143] By deleting the i-th column from the matrix C v Get a C v(≠i,:) .

[0144] C v(i,i) It is C v Finally, using Normalize v.

[0145] Step S53: fix v and solve for u. After obtaining v, fix v to solve for u.

[0146] The objective function of u can be rewritten as (7).

[0147]

[0148]

[0149] The value of u is obtained by formula (7), which is the same as the method of (6), and then (8) is obtained:

[0150]

[0151] If x>0, x + Defined as x; if x≤0, x + Defined as 0.

[0152] By deleting the i-th column from the matrix C u Get C from u(≠i,:) , which is obtained by removing the i-th coefficient from the previous iteration.

[0153] is the current iteration u i The update coefficient of C u(i,i) It is C u The i-th diagonal element of .

[0154] Finally, using Normalize u.

[0155] Step S54, repeat step S52 and step S53 until the convergence condition is met, and obtain the corresponding weight vectors u and v.

[0156] In this embodiment, the alternating convex search method needs to select 6 parameters in total, namely, λ1, λ2, β1, β2, γ1 and γ2.

[0157] Figure 2 The line graph of the sparse canonical correlation analysis algorithm in the parameter selection process according to an embodiment of the present invention is disclosed. Six parameters are selected from the parameter set [0.001 0.01 0.1 1] using the grid search method. There are 4096 parameter combinations in total. The canonical correlation coefficients corresponding to the 4096 parameter combinations are shown in the figure. Figure 2 As shown, the 198th parameter combination can achieve the maximum typical correlation coefficient.

[0158] Step S6: sort the brain region imaging feature data and the gene expression data according to the corresponding weights, and obtain the brain region imaging feature data and the gene expression data within a preset weight value range as biomarkers.

[0159] Through the sparse canonical correlation analysis algorithm based on autocorrelation constraints, the weight vectors u and v of the characteristics of the two modal (brain region / gene) data can be obtained, and the absolute values ​​of the weights are taken and sorted according to the weight size.

[0160] Take the absolute values ​​of the two vectors and sort them from large to small. The larger the corresponding feature weight is, the more important it is in the association process, and the top-ranked brain regions and genes can be obtained.

[0161] Furthermore, the absolute values ​​of the weight vectors u and v were taken, and then the weights were sorted from large to small, and the top 10 features with the largest weights of the two types of data (brain regions / genes) were obtained for subsequent analysis.

[0162] Figure 3 A visualization heat map of weight u and weight v according to an embodiment of the present invention is disclosed, such as Figure 3The figure shows how to obtain the corresponding weights of the two data features. The absolute values ​​of the feature vectors corresponding to the two data features are taken, and the larger the absolute value of the weight, the more important the corresponding feature is. The features are sorted from large to small according to the weight.

[0163] Tables 1 and 2 give the weight information of the top 10 brain regions and genes respectively. Figure 4 The visualized heat map of the first 10 brain regions according to an embodiment of the present invention is disclosed, such as Figure 4 Visualization results for the first 10 brain regions are shown.

[0164] Table 1 Names and weights of the top 10 brain regions

[0165] Chinese name Abbreviations Weight information Left frontal mid-orbital lMidFroOrb 0.4083 Right medial frontal orbit rMedFroOrb 0.3686 Right upper frontal orbit rSupFroOrb 0.3106 Right outer cover rRolOpe 0.2962 Left rectus muscle 1L 0.2463 Left medial frontal orbit MedFroOrb 0.2437 Left outer cover lRolOpe 0.2295 Right hippocampus rHip 0.2170 Left upper occipital bone SupOcc 0.1858 Left hippocampus Hip 0.1512

[0166] Table 2 Names and weights of the top 10 genes

[0167]

[0168]

[0169] The above genes have been proven to play an important role in the occurrence and development of Alzheimer's disease.

[0170] For example, the results of the study by Allan I Levey et al. showed that atomoxetine treatment could significantly change CDCP1, which is associated with brain disease and inflammation, in the cerebrospinal fluid of patients with mild cognitive impairment.

[0171] For example, numerous studies have demonstrated the potential role of aberrant Wnt signaling in cognitive impairment.

[0172] The analysis results in this embodiment not only found the left and right hippocampus, two known brain regions at risk for Alzheimer's disease, but also other brain regions that have been confirmed by a large number of studies.

[0173] For example, the results of the study by Jesper Riise et al. showed that the Wnt pathway components Wnt7b and TCFTL1 showed overlapping gene expression changes in two medial temporal lobe structures, while β-catenin was negatively expressed between brain regions.

[0174] For example, the study by Chenfei Ye et al. confirmed that the left orbital part of the superior and middle frontal gyri and the left hippocampus changed significantly during the progression of Alzheimer's disease.

[0175] Therefore, the top 10 features with the largest weights of the two types of data obtained in this embodiment can indeed be used as biomarkers for subsequent analysis of Alzheimer's disease.

[0176] Step S7: constructing a diagnostic model for Alzheimer's disease based on biomarkers for auxiliary diagnosis and analysis of Alzheimer's disease.

[0177] The top 10 features with the largest weights of the two types of data (brain region / gene) were used to construct a diagnostic model for Alzheimer's disease. Furthermore, the top several features with the largest absolute weights of brain region imaging feature data and gene expression data were used as biomarkers, and the logistic regression algorithm was used to verify the constructed diagnostic model and explore the diagnostic significance of biomarkers.

[0178] The logistic regression algorithm can be used for classification problems. Its essence is to use the sigmoid function to transform the model on the linear regression algorithm so that the target value is between 0 and 1.

[0179] In this embodiment, the logistic regression algorithm performs a weighted summation of the input features and inputs the result into the logistic function to convert the result of the linear combination into a probability value. This probability value can be understood as the probability that the predicted sample belongs to a certain category. Based on the results of the algorithm, the ROC curve is drawn to verify the performance of the diagnostic model.

[0180] Logistic regression was used to explore the diagnostic significance of the top 10 brain regions and top 10 gene signatures.

[0181] Figure 5a The diagnostic performance verification curve diagram of the markers of the first 10 brain regions according to an embodiment of the present invention is disclosed, such as Figure 5a The ROC curves (receiver operating characteristic curve) of the first 10 brain regions in the training set and the test set are given.

[0182] Figure 5b The diagnostic performance verification curve diagram of the markers of the first 10 genes according to an embodiment of the present invention is disclosed, as shown in FIG. Figure 5b The ROC curves of the top 10 genes in the training set and the test set are given.

[0183] Detailed information on the ROC curves for brain regions and genes is given in Tables 3 and 4 , respectively.

[0184] like Figure 5a-Figure 5b As shown in Tables 3 and 4, the area under the curve (AUC) of all brain regions and genes in the test set was greater than 0.5 and within the confidence interval, which indicates that the Alzheimer's disease imaging genetics data analysis method based on prior information guidance proposed in the present invention can accurately mine the imaging and genetic markers of Alzheimer's disease related to diagnosis.

[0185] Table 3 Detailed information of ROC curves for the first 10 brain regions

[0186] Predictor variables Area under the curve (AUC) Confidence Interval (CI) lMidFroOrb 0.631 0.439-0.822 rMedFroOrb 0.617 0.404-0.830 rSupFroOrb 0.637 0.454-0.821 rRolOpe 0.614 0.411-0.817 1L 0.506 0.279-0.733 MedFroOrb 0.544 0.303-0.786 lRolOpe 0.531 0.321-0.740 rHip 0.517 0.308-0.726 SupOcc 0.547 0.317-0.777 Hip 0.539 0.332-0.746

[0187] Table 4 Detailed information of ROC curves of the top 10 genes

[0188] Predictor variables Area under the curve (AUC) Confidence Interval (CI) CDCP1 0.697 0.529-0.866 EHD4 0.739 0.585-0.893 IZUMO4 0.642 0.457-0.827 TUBGCP2 0.503 0.318-0.688 UQCC1 0.504 0.285-0.724 C19ORF59 0.581 0.398-0.763 CTNNBIP1 0.711 0.515-0.907 TCF7L1 0.576 0.382-0.771 ABHD8 0.725 0.561-0.889 GPR34 0.636 0.430-0.842

[0189] The experimental results show that the Alzheimer's disease imaging genetics data analysis method based on prior information guidance proposed in this invention obtains more biologically interpretable brain regions and genes that are significantly related to Alzheimer's disease, has good correlation, and can extract stronger correlations and classification accuracy between different modalities even in the case of large noise, providing a new basis for identifying association patterns and biomarkers of Alzheimer's disease. In addition, the orthogonal constraint on the weight vector can also prevent redundant features from affecting the results.

[0190] The Alzheimer's disease imaging genetics data analysis method based on prior information guidance proposed in the present invention can, to a certain extent, deeply mine the diagnostic biomarkers related to Alzheimer's disease in imaging genetics data, thereby assisting the development of Alzheimer's disease therapeutic targets. The present invention effectively overcomes various shortcomings in the prior art and has high industrial utilization value.

[0191] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.

[0192] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0193] The above embodiments are provided for persons familiar with the art to implement or use the present invention. Personnel familiar with the art can make various modifications or changes to the above embodiments without departing from the inventive concept of the present invention. Therefore, the protection scope of the present invention is not limited to the above embodiments, but should be the maximum scope of the innovative features mentioned in the claims.

Claims

1. A method for analyzing Alzheimer's disease imaging genetics data, characterized in that: The following steps are involved: Step S1, obtaining structural magnetic resonance imaging and gene expression data of Alzheimer's disease and control group samples; Step S2, preprocessing the structural magnetic resonance imaging data, and extracting the gray matter volume of the brain region as brain region imaging feature data; Step S3: Divide all samples into training set and test set according to a certain ratio; Step S4, performing differential expression analysis on the gene expression data in the training set to obtain processed gene expression data; Step S5: Based on the sparse canonical correlation analysis algorithm guided by prior information, correlation analysis is performed on the processed brain region imaging feature data and gene expression data of the training set to obtain corresponding weights; Step S6, sorting the brain region imaging feature data and the gene expression data according to the corresponding weights, and obtaining the brain region imaging feature data and the gene expression data within a preset weight value range as biomarkers; Step S7: constructing a diagnostic model for Alzheimer's disease based on the biomarkers, wherein the diagnostic model is used for auxiliary diagnosis and analysis of Alzheimer's disease.

2. The Alzheimer's disease imaging genetics data analysis method according to claim 1, characterized in that: The step S1 further comprises: The structural magnetic resonance imaging data of the Alzheimer's disease and control group samples are in nii format; The gene expression data of the Alzheimer's disease and control group samples are tabular data without missing values.

3. The Alzheimer's disease imaging genetics data analysis method according to claim 1, characterized in that: The step S2, preprocessing the structural magnetic resonance imaging data, further comprises: Affine registration of the raw data of structural MRI was performed to achieve head motion correction and register it to the MNI standard space; The structural MRI data of the training set were registered to the AAL template and feature extraction was performed.

4. The Alzheimer's disease imaging genetics data analysis method according to claim 1, characterized in that: The step S3 further comprises: 80% of all samples are used as training sets and 20% as test sets; The training set is used for association analysis and screening of biomarkers, and the test set is used to test the performance of the diagnostic model.

5. The Alzheimer's disease imaging genetics data analysis method according to claim 1, characterized in that: The step S4 further comprises: The Limma algorithm was used to perform differential expression analysis on gene expression data; The expression levels of differentially expressed genes whose algorithm parameter p was less than the preset threshold were retained as processed gene expression data for subsequent association analysis.

6. The Alzheimer's disease imaging genetics data analysis method according to claim 1, characterized in that: The step S5 further comprises: The penalty function based on autocorrelation is introduced into the sparse canonical correlation analysis model to study the correlation between brain regions and genes. The corresponding objective function after the introduction is: Among them, u represents the norm vector of X, and v represents the norm vector of Y; λ1,λ2,β1,β2 are regularization parameters; Given a dataset X∈R n×p , Y∈R n×q They correspond to structural magnetic resonance imaging data and gene expression data, respectively, and n represents the number of samples.

7. The method for analyzing Alzheimer's disease imaging genetics data according to claim 6, characterized in that: The step S5 further comprises: The alternating convex search method is used to solve the objective function to obtain the optimal values ​​of u and b; The alternating convex search method further comprises: Step S51, initializing u and v; Step S52: fix u and solve v. The expression corresponding to the objective function of v is as follows: Step S53: fix v and solve for u. The expression corresponding to the objective function of u is as follows: Step S54, repeat step S52 and step S53 until the convergence condition is met, and obtain the corresponding weight vectors u and v.

8. The method for analyzing Alzheimer's disease imaging genetics data according to claim 1, characterized in that: The step S6 further comprises: The brain imaging feature data and gene expression data are sorted according to the corresponding weight absolute values, and the first several features with the largest weight absolute values ​​of the brain imaging feature data and gene expression data are obtained as biomarkers for subsequent auxiliary diagnosis analysis.

9. The method for analyzing Alzheimer's disease imaging genetics data according to claim 1, characterized in that: The step S7 further comprises: The first several features with the largest absolute value ranking based on brain imaging feature data and gene expression data were used as biomarkers, and the constructed diagnostic model was verified using the logistic regression algorithm.

Citation Information

Patent Citations

  • Alzheimer's disease genetic biomarker determination method and system

    CN110232679A

  • Method for mining biomarkers based on multi-map neuroimaging data

    CN112233805A

  • Guided multi-modal image genetics data feature analysis method

    CN114202075A

  • Biomarker panels, systems, and methods for risk stratification of a subject for alzheimer's disease

    WO2021186478A1