A method for identifying causal genes in brain cells based on a Mendelian randomization framework

Abstract

The present invention discloses a method for identifying causal genes of brain cells based on a Mendelian randomization framework, which screens genetic variations related to brain diseases and behavioral characteristics as initial instrumental variables, performs whole transcriptome association analysis, and then uses causal inference methods for inference, and finally performs testing and screening. The method involves the fields of computational biology, biomedical engineering and applied mathematics technology, and solves the technical problem of how to accurately identify causal genes that affect complex brain characteristics at the cell type level, and fully integrates the advantages of multiple inference methods, thereby providing a new approach for finding more effective therapeutic targets. The present invention infers the causal relationship between genes and complex brain characteristics in specific brain cell types from massive genetic data and complex brain characteristics, locates potential brain trait markers, and realizes the identification of genes that affect complex brain characteristics in specific cell types, thereby laying a solid foundation for the development of more accurate and efficient therapeutic targets.

Description

Technical Field

[0001] The present invention belongs to the fields of computational biology, biomedical engineering and applied mathematics technology, and specifically is a method for identifying causal genes of brain cells based on a Mendelian randomization framework. Background Art

[0002] GWAS (Genome-Wide Association Studies) have revealed thousands of genetic variants that are closely associated with complex traits of the human brain. Most of these variants are hidden in non-coding regions, revealing a huge unsolved mystery in the genome. In order to unveil the biological veil of these non-coding genetic signals, eQTLs (expression quantitative trait loci) have become a core tool for analyzing GWAS results as a bridge connecting genetic variation and gene expression levels. The high overlap between GWAS sites and eQTLs suggests a potential pathway by which certain genetic variations may be indirectly associated with the risk of brain diseases by regulating gene expression.

[0003] Therefore, the deep integration of eQTLs and GWAS findings aims to accurately identify causal genes, which not only helps to further elucidate the molecular mechanisms of complex traits, but also provides key clues for the development of drug targets. This process is highly dependent on Mendelian randomization (MR), a powerful causal inference technology. MR has shown irreplaceable importance in revealing the causal relationship between gene expression and complex neurological, psychological and behavioral traits.

[0004] At present, the analysis of identifying causal genes affecting complex brain-related traits is limited to the pan-tissue level and a single inference method. There is a lack of causal gene identification research at the cell type level. Due to the particularity of gene expression data, a single inference method is often too limited. With the development of GWAS, eQTLs and GWAS data are integrated to identify causal genes affecting complex brain traits. The existing methods have the following defects:

[0005] 1. It is limited to the pan-tissue level and lacks more sophisticated causal gene prediction research at the cell type level;

[0006] 2. Only a single inference method is used for prediction, without considering the specificity of gene expression data;

[0007] 3. Failure to reveal the common genetic basis of multiple complex brain phenotypes at the cellular level.

[0008] In response to the above problems, the present invention fully integrates the advantages of multiple inference methods to infer the causal relationship between potential gene markers in specific brain cell types and multiple complex brain traits, providing a more sophisticated approach to finding more effective therapeutic targets. Summary of the invention

[0009] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a method for identifying causal genes in brain cells based on a Mendelian randomization framework. The main technical problem solved is how to accurately identify causal genes that affect complex brain characteristics at the cell type level, and fully integrate the advantages of multiple inference methods, thereby providing a new approach to finding more effective therapeutic targets. The present invention solves the above-mentioned problem by inferring the causal relationship between genes in specific brain cell types and complex brain characteristics from massive genetic data and complex brain characteristics, and locating potential brain trait markers.

[0010] To achieve the above object, the first aspect of the present invention provides a method for identifying causal genes of brain cells based on a Mendelian randomization framework, comprising the following steps:

[0011] S1: Preliminarily collect, clean, and organize the eQTL data of the required brain cell types and the genome-wide association analysis data of brain diseases and behavioral traits, and preliminarily screen out genetic variants related to brain diseases and behavioral traits as initial instrumental variables;

[0012] S2: Download the required prediction weights and covariance files from the external data repository and perform whole transcriptome association analysis to obtain the association between gene expression in specific cell types and brain diseases;

[0013] S3: Based on the correlation between gene expression of specific cell types and brain diseases, multiple complementary two-stage causal inference methods are used to infer the causal relationship between gene expression of specific cell types and brain diseases; the multiple complementary two-stage inference methods include: SMR method, PMR-Egger method and GSMR method;

[0014] S4: Test the robustness and repeatability of the causal relationship, and screen the causal relationship based on the test results to obtain relevant causal relationships;

[0015] S5: Based on the relevant causes, potential causal pathways between cell type-specific genes, brain imaging features, and brain diseases / behaviors are constructed to reveal the regulatory role of causal genes in brain diseases and behavioral traits.

[0016] The above scheme can reveal the common genetic basis of various complex brain phenotypes at the cellular level, promoting a deeper understanding of brain structure, disease and behavioral effects.

[0017] Preferably, in S2, the whole transcriptome analysis uses the MetaXcan method to analyze the correlation between genes and complex brain phenotypes. g Conduct analysis;

[0018] Correlation Z g The expression is as follows:

[0019] Among them, w lg is the weight of SNP l in predicting the expression of gene g; is the GWAS regression coefficient of SNP l; yes The standard deviation of is the estimated variance of SNP1; is the estimated variance of the predicted expression of gene g.

[0020] Preferably, the model in the SMR method is expressed by the formula Obtain the causal effect of exposure factor x on outcome variable y in, is the estimated SNP effect of a trait GWAS, is the estimated SNP effect on the expression level of a gene from the eQTL data, The variance of can be expressed by the Delta method as Statistics For inspection of significance.

[0021] It should be noted that the SMR method detects whether gene expression levels are associated with complex traits under the premise that there is a causal relationship or pleiotropy between exposure and outcome variables.

[0022] Preferably, the model in the PMR-Egger method is expressed as:

[0023] Where x represents the exposure factor vector (i.e., gene expression) of n1 dimension in gene expression data. x represents an n1×p-dimensional genotype matrix, p is the number of instrumental variables, y is represented as an n2-dimensional outcome factor vector (i.e., brain-related complex phenotype), Z y It represents an n2×p genotype matrix, μ x and μ y are the intercepts on the x-axis and y-axis respectively, represents the n2-dimensional exposure factor vector in the GWAS data, β represents the p-dimensional vector of the effect size of the instrumental variable on the exposure factor, α represents the causal relationship scalar of the exposure variable on the outcome variable, γ represents the p-dimensional vector of the horizontal multi-effect of the instrumental variable on the outcome variable, and ε x , ε represents the residual vector.

[0024] It should be noted that the PMR-Egger method is based on the assumption of horizontal pleiotropy and estimates the causal effect of gene expression levels on traits by constructing a specific statistical model within the MR likelihood framework.

[0025] Preferably, the effects of exposure factors and outcome variables obtained by the GSMR model in the GSMR method can be expressed as:

[0026]

[0027] in, is the SNP effect estimate of the GWAS for a trait y, is the SNP effect estimate of the GWAS for a trait x, The variance of can be expressed as Statistics For inspection of significance.

[0028] It should be noted that the GSMR method is an extension of the SMR method and is used to estimate the causal effect of exposure factors on outcomes, while taking into account the influence of multiple variants and linkage disequilibrium between variants.

[0029] Preferably, in S3, the multiple complementary two-stage causal inference methods also include the Wald ratio method.

[0030] It should be noted that the Wald ratio method calculates the change in disease risk per standard deviation change in gene expression through the instrumental cis-eQTL of a specific gene. It is a general method that is applicable to any type of exposure and only requires one SNP as an instrumental variable.

[0031] It should be noted that the criteria for determining causality are as follows:

[0032] 1) Significant after false discovery rate (FDR) correction in at least one single instrumental variable method (corrected P < 0.05 was significant);

[0033] 2) significant after FDR correction in at least one of the multiple instrumental variable methods (corrected P < 0.05 was significant);

[0034] 3) show consistent direction of causal effects across multiple methods;

[0035] When any of the above conditions is met, there is a causal relationship between gene expression in a specific cell type and brain disease.

[0036] The reliability of the results is ensured by integrating multiple complementary causal inference methods.

[0037] Preferably, the testing of the robustness and repeatability of the causal relationship and screening the causal relationship according to the test results comprises the following steps:

[0038] Step 1: Use the HEIDI test to test whether there is heterogeneity in the results of the SMR method; if yes, eliminate the causal relationship; if not, jump to step 2;

[0039] Step 2: Use the HEIDI-outlier test to test whether there is heterogeneity in the results of the GSMR method; if yes, eliminate the causal relationship; if not, jump to step 3;

[0040] Step 3: Use the Cragg-Donald statistic to calculate the statistic F of each tool and determine whether the validity of each instrumental variable is reliable; if not, eliminate the causal relationship; if yes, jump to step 4.

[0041] Step 4: Verify the repeatability of genes of specific cell types corresponding to the causal relationship through an external database and a hypergeometric test to obtain verification results; wherein the verification results include qualified and unqualified;

[0042] Step 5: Eliminate the causal relationships with unqualified verification results.

[0043] It should be noted that in step 1, the HEIDI test is used to test whether there is heterogeneity in the results of the SMR method; the test statistic is:

[0044] Wherein, the subscript i represents SNP i, m represents the number of eQTLs significantly associated with the exposure factor, and z d(i) represents the difference between the results of SNP i in an eQTL region and the SNP with the highest association with the exposure factor in the region, d i =b xy(i) -b xy(top) It represents the difference between the SMR results of the SNP with the highest association with the exposure factor included in the analysis and SNP i in the same eQTL region;

[0045] Among them, cov(d i ,d j )=cov(b xy(i) ,b xy(j) )-cov(b xy(i) ,b xy(top) )-cov(b xy(j) ,b xy(top) )+var(b xy(top) ), subscript j represents SNP j.

[0046] Statistics T HEIDI The distribution of can be obtained by Saddlepoint method (Davies, RB Numerical inversion of a characteristic function. Biometrika 60, 415–417 (1973)). Under the null hypothesis that there is no heterogeneity, d = {d1, d2, …d m}=0, then T HEIDI = 0. According to the SMR results, T HEIDI value, and by T HEIDI The p-value of the significance level is obtained from the distribution. If the p-value is less than 0.05, it can be judged that T HEIDI It is significantly different from 0, indicating that the results of the SMR method are heterogeneous and need to be eliminated.

[0047] In step 2, the HEIDI-outlier test is used to test whether there is heterogeneity in the results of the GSMR method. The test statistic involved in this process is:

[0048] Among them, d i =b xy(i) -b xy(top) The content is the same as step 1. T obeys X 2 distributed.

[0049] Under the null hypothesis that there is no heterogeneity, d = {d1, d2, …d m}=0, then T=0. The T value is obtained according to the GSMR results, and the significance level p value is obtained from the distribution of T. If the p value is less than 0.05, it can be judged that T is significantly not 0, indicating that the results of the GSMR method are heterogeneous and need to be eliminated.

[0050] In step three, the Cragg-Donald statistic is used to calculate the statistic F of each instrumental variable to evaluate the effectiveness of a single SNP as an instrumental variable. The specific formula is:

[0051]

[0052] In the formula, represents the proportion of the exposure factor explained by the instrumental variable, N represents the sample size of the eQTL summary data, β is the effect size of the instrumental variable of the eQTL summary data, se is the standard error of β, and k is the number of instruments used in the MR estimation.

[0053] If the F value is greater than the empirical value of 10, it means that the SNP selected as the instrumental variable fails to meet the validity criteria, and the analysis results based on this SNP need to be eliminated.

[0054] The external databases in step 4 include but are not limited to BrainMeta, PsychENCODE, and NHGRI-EBIGWAS Catalog.

[0055] Preferably, in S5, the bidirectional causal relationship between brain imaging features and brain diseases / behaviors is jointly determined based on the inference methods of multiple instrumental variables: GSMR, PMR-Egger, IVW, MR-Egger, Weighted median, Simple mode, and Weighted mode.

[0056] Compared with the prior art, the present invention provides a causal inference framework that integrates the advantages of multiple algorithms to infer causal relationships in specific cell types, and is used to infer the causal relationship between genes and complex brain characteristics in specific brain cell types from massive genetic data and complex brain characteristics, locate potential brain trait markers, and achieve the identification of genes that affect complex brain characteristics in specific cell types, thereby laying a solid foundation for the development of more accurate and efficient therapeutic targets. It also integrates the advantages of multiple algorithms, provides robust and repeatable results, and the automatic and fast analysis framework can dig out a large number of potential causal relationships between variables in the database in a short time, greatly saving the time and energy of scientific researchers. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 The figure is a general framework diagram of the method for identifying causal genes of brain cells based on the Mendelian randomization framework of the present invention;

[0059] Figure 2 This is a diagram showing the overall results of the application of the brain cell causal gene identification method based on the Mendelian randomization framework in the present invention;

[0060] Figure 3 This is an example diagram of the predicted gene-IDP pair gene-DB in the present invention that can be replicated in an external database;

[0061] Figure 4 A schematic diagram of genes that have causal relationships among various phenotypes in the present invention;

[0062] Figure 5 A schematic diagram of the causal genes shared among brain-related complex traits and the corresponding FDR values ​​in the present invention;

[0063] Figure 6 It is a schematic diagram of the presentation pattern of shared causal genes between DB and IDP categories in brain cell types in the present invention;

[0064] Figure 7 Schematic diagram of the potential biological causal pathway of the cell type-specific gene-DB-IDP in the present invention. DETAILED DESCRIPTION

[0065] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] The present invention focuses on the situation where gene expression in specific types of brain cells is a risk factor. In previous studies, the exploration of causal genes for complex traits has mostly focused on integrating GWAS data with pan-tissue eQTLs. However, in recent years, the rise of research on cell type-specific eQTLs has revealed that they have a more precise range of action and a stronger effect than pan-tissue eQTLs. Studies such as OneK1K, Julien et al. and the publication of projects such as MetaBrain have further fine-tuned eQTLs to specific cell types and opened a new chapter in the cell type specificity of genetic regulation of gene expression. The mapping of single-cell eQTL maps not only reveals novel and previously obscured cell type-specific genetic regulatory patterns in brain diseases, but also suggests that Alzheimer's disease (AD) risk-associated variants may play a key role mainly through cis-eQTLs in microglia, while schizophrenia (SCZ) exhibits a wider range of cell type genetic complexity. These findings profoundly reveal that compared with macro-level regulation, fine regulation at the cellular level is more sensitive to analyzing the pathological mechanisms of complex diseases, but our understanding of how genetic variation contributes to the causal genes of brain traits at the cell type level is still insufficient.

[0067] In addition, the shared genetic and cellular basis between brain imaging-derived phenotypes (IDPs) and diseases and behaviors (DBs) has opened up new perspectives for the study of complex traits. Many studies have shown that there are significant genetic correlations between complex traits, diseases and brain imaging features. GWAS has precisely located multiple genetic loci and genes that play a key role in brain imaging phenotypes and various diseases, such as the MAPT gene (17q21) and the APOE gene (19q13). Of particular importance is that shared causal genes for psychiatric disorders, neurodegenerative diseases, and brain structural traits show unique expression patterns in specific brain cell types. For example, there is a significant association between the specific expression of AKT3 in astrocytes and cerebral cortical surface area, Parkinson's disease, and major depressive disorder, and there is a significant association between the specific expression of DIP2B in excitatory neurons and SCZ and brainstem volume. Cell type-specific genetic enrichment analysis of single nucleotide variants across the genome further confirmed the close association between oligodendrocytes and white matter microstructure and depression, and between excitatory neurons and brain surface area and SCZ. These findings suggest that shared genetic foundations at the cellular level profoundly influence the relationship between brain structure and complex traits, and revealing the underlying pathways may provide new opportunities for the development of therapeutic targets.

[0068] Based on this, the present invention has developed a method for identifying causal genes in brain cells based on a Mendelian randomization framework, which can infer the causal relationship between genes in specific cell types and complex brain characteristics and locate markers of potential brain-related traits. The advantages of the present invention are:

[0069] (1) It can enable the precise identification of cell type-specific genes that affect complex brain features, providing more targeted therapeutic targets.

[0070] (2) It can reveal the common genetic basis of various complex brain phenotypes at the cellular level and promote a deeper understanding of the impact on brain structure, disease and behavior.

[0071] (3) It integrates multiple complementary causal inference methods to ensure the reliability of the results. This invention has been successfully applied to eQTL data of eight different brain cell types and large-scale GWAS data of 123 imaging-derived phenotypes and 26 brain diseases / behaviors, locating multiple potential causal marker genes associated with complex brain phenotypes, providing clues and inspiration for the development of targeted brain health management strategies and potential therapeutic targets.

[0072] See also Figure 1 - Figure, the first aspect of the present invention provides a method for identifying causal genes of brain cells based on a Mendelian randomization framework, comprising the following steps:

[0073] S1: Preliminarily collect, clean, and organize the gene expression summary data of the required brain cell types and the genome-wide association analysis data of brain diseases and behavioral traits, and preliminarily screen out genetic variants associated with brain diseases / behaviors and behavioral brain imaging traits as initial instrumental variables;

[0074] S2: Download the required prediction weights and covariance files from the PredictDB data warehouse and the dorsolateral prefrontal cortex (DLPFC) data warehouse of the CMC database, and perform whole transcriptome association analysis to obtain the relationship between gene expression of specific cell types and brain diseases;

[0075] Among them, the PredictDB data warehouse (https: / / predictdb.org / ) covers 13 conventional brain tissues, such as the amygdala, anterior cingulate cortex (BA24), caudate nucleus, cerebellar hemispheres and cerebellum itself, widespread cortical areas, specific areas such as the frontal cortex (BA9), hippocampus, hypothalamus, nucleus accumbens, lentiform nucleus, cervical spine area, substantia nigra, covering 14 common brain tissues.

[0076] The dorsolateral prefrontal cortex (DLPFC) data repository of the CMC database (https: / / github.com / laurahuckins / CMC_DLPFC_prediXcan), which contains brain tissue DLPFC;

[0077] S3: Based on the relationship between gene expression in specific cell types and brain diseases, multiple complementary two-stage causal inference methods are used to infer the causal relationship between gene expression in specific cell types and brain diseases; the multiple complementary two-stage inference methods include: SMR method, PMR-Egger method and GSMR method;

[0078] S4: Test the robustness and repeatability of the causal relationship and screen the causal relationship based on the test results;

[0079] S5: Construct potential causal pathways between cell type-specific genes, brain imaging features, and brain diseases / behaviors to reveal the regulatory roles of these causal genes in brain diseases and behavioral traits.

[0080] In this example, the whole transcriptome analysis of S2 was performed using the MetaXcan method (Barbeira, AN, et al. Exploring the phenotypic consequences of tissue specific gene expression variation inferred from GWAS summary statistics. Nat Commu, 2018) to analyze the association between genes and complex brain phenotypes;

[0081] The associative expression is as follows:

[0082] Among them, w lg is the weight of SNP l in predicting the expression of gene g; is the GWAS regression coefficient of SNP l; yes The standard deviation of is the estimated variance of SNP1; is the estimated variance of the predicted expression of gene g.

[0083] Screening for genes significantly associated with complex brain traits: Gene-phenotype pairs with P < 0.05 in at least one brain tissue were retained for subsequent Mendelian randomization analysis.

[0084] The implementation method of this embodiment in S3 is as follows:

[0085] Secondary screening of initial instrumental variables for analysis: Screening of corresponding single / multiple, related / independent SNPs as instrumental variables for the input requirements of each algorithm used subsequently;

[0086] Causal inference was performed using four complementary two-sample univariate Mendelian randomization analysis methods.

[0087] In this embodiment, the model in the SMR method is expressed by the formula Obtain the effects of exposure factors and outcome variables; among them, is the estimated SNP effect of a trait GWAS, is the estimated SNP effect on the expression level of a gene from the eQTL summary data, The variance of can be expressed by the Delta method as Statistics For inspection of significance.

[0088] The SMR method (Zhu Z., et al. Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets. Nat Genet, 2016) detects whether gene expression levels are associated with complex traits under the premise of assuming that there is a causal relationship or pleiotropy between exposure and outcome variables. This method is designed specifically for gene expression as an exposure factor and only requires one SNP as an instrumental variable, thereby reducing the amount of calculation during analysis.

[0089] Similarly, the model in the PMR-Egger method is expressed as:

[0090] Where x represents the exposure factor vector (i.e., gene expression) of n1 dimensions. x represents an n1×p-dimensional genotype matrix, p is the number of instrumental variables, y is represented as an n2-dimensional outcome factor vector (i.e., brain-related complex phenotype), Z y It represents an n2×p genotype matrix.

[0091] The PMR-Egger method (Yuan Z., et al. Testing and controlling for horizontal pleiotropy with probabilistic Mendelian randomization in transcriptome-wide association studies. Nat Commun, 2020) is based on the assumption of horizontal pleiotropy. Within the MR likelihood framework, a specific statistical model is constructed to estimate the causal effect of gene expression levels on traits. This method is specifically designed to use gene expression as an exposure factor. It can effectively handle multiple related variants as instrumental variables and consider linkage disequilibrium (LD) between variants.

[0092] Similarly, the effects of exposure factors and outcome variables obtained by the GSMR model in the GSMR method can be expressed as:

[0093] in, is the SNP effect estimate of the GWAS for a trait y, is the SNP effect estimate of the GWAS for a trait x, The variance of can be expressed as Statistics For inspection of significance.

[0094] The GSMR method (Zhu Z., et al. Causal associations between risk factors and common diseases inferred from GWAS summary data. Nat Commun, 2018) is an extension of the SMR method, which is used to estimate the causal effect of exposure factors on outcomes, while taking into account the influence of multiple variants and linkage disequilibrium between variants. It is a general method that is applicable to any type of exposure. The effects of exposure factors and outcome variables estimated by the GSMR model can be expressed as:

[0095]

[0096] in, is the SNP effect estimate of the GWAS for a trait y, is the SNP effect estimate of the GWAS for a trait x, The variance of can be expressed as Statistics For inspection of significance.

[0097] As another preferred embodiment, S3's multiple complementary two-stage causal inference methods also include the Waldratio method (Hemani G., et al. The MR-Base platform supports systematic causalinference across the human phenome. Elife, 2018).

[0098] The Wald ratio method calculates the change in disease risk per standard deviation change in gene expression through the instrumental cis-eQTL for a specific gene. It is a general method that is applicable to any type of exposure and only requires one SNP as an instrumental variable.

[0099] A gene is considered causal for a trait in a given cell type if it meets the following criteria:

[0100] 1) Significant after false discovery rate (FDR) correction in at least one single instrumental variable method (corrected P < 0.05);

[0101] 2) significant after FDR correction in at least one of the multiple instrumental variable methods (corrected P < 0.05);

[0102] 3) Shows consistent direction of causal effect across all four approaches.

[0103] In this embodiment, the robustness and repeatability of the causal relationship are tested, and the causal relationship is screened according to the test results, including the following steps:

[0104] Step 1: Use HEIDI test to check whether there is heterogeneity in the results of SMR method (Zhu Z., et al. Integration of summary data from GWAS and eQTL studies predicts complextrait gene targets. Nat Genet, 2016); if yes, eliminate the causal relationship; if no, jump to step 2;

[0105] Step 2: Use HEIDI-outlier test to test whether there is heterogeneity in the results of GSMR method (Zhu Z., et al. Causal associations between risk factors and common diseases inferred from GWAS summary data. Nat Commun, 2018); if yes, eliminate the causal relationship; if not, jump to step 3;

[0106] Step 3: Use the Cragg-Donald statistic to calculate the statistic F of each tool and determine whether the validity of each instrumental variable is reliable (Cragg JG, Donald SG. Testing identifiability and specification in instrumental variable models. Econometric theory, 1993); if not, eliminate the causal relationship; if yes, jump to step 4.

[0107] Step 4: Verify the repeatability of genes in specific cell types corresponding to causal relationships through external databases such as BrainMeta, PsychENCODE, and NHGRI-EBI GWAS Catalog and use hypergeometric tests to obtain verification results; the verification results include qualified and unqualified;

[0108] Step 5: Eliminate the causal relationships with unqualified verification results.

[0109] When the HEIDI test is used to test whether there is heterogeneity in the results of the SMR method, the test statistic is:

[0110] in, d i =b xy(i) -b xy(top) It represents the difference between the SMR results of the SNP included in the analysis and the surrounding adjacent SNPi;

[0111] Among them, cod(d i ,d j )=cov(b xy(i) ,b xy(j) )-cov(b xy(i) ,b xy(top) )-cov(b xy(j) ,b xy(top) )+var(b xy(top) ).

[0112] In step 2, the HEIDI-outlier test is used to test whether there is heterogeneity in the results of the GSMR method. The test statistic involved in this process is:

[0113] in,

[0114] In step three, the Cragg-Donald statistic is used to calculate the statistic F of each tool to evaluate the effectiveness of a single SNP as an instrumental variable. The specific formula is:

[0115]

[0116] In the formula, represents the proportion of exposure factors explained by the instrumental variable, N represents the sample size of the eQTL summary data, β is the effect size of the instrumental variable of the eQTL summary data, se is the standard error of β, and k is the number of tools used in the MR estimation. If the F value of a SNP used as an instrumental variable is less than the empirical value of 10, it means that the strength of the instrumental variable is insufficient and the analysis result is unreliable.

[0117] The external databases in step 4 include but are not limited to BrainMeta (Qi T., et al. Genetic control of RNA splicing and its distinct role in complex trait variation. Nat Genet, 2022), PsychENCODE (Gandal MJ., et al. Transcriptome-wide isoform-level dysregulation in ASD, schizophrenia, and bipolar disorder. Science, 2018), and NHGRI-EBI GWAS Catalog (Sollis E., et al. The NHGRI-EBI GWAS Catalog: knowledge base and description resource. Nucleic Acids Res, 2023).

[0118] S5. Cell type-specific gene-IDP-DB pathway analysis: A variety of Mendelian randomization analysis methods based on multiple instrumental variables were used to identify the bidirectional causal relationship between IDP and DB; among them, the bidirectional causal relationship between brain imaging features and brain diseases / behaviors was jointly determined based on the inference methods of multiple instrumental variables: GSMR, PMR-Egger, IVW, MR-Egger, Weightedmedian, Simple mode, and Weighted mode;

[0119] By integrating gene-IDP pairs and gene-DB pairs, cell type-specific gene-IDP-DB pathways and cell type-specific gene-DB-IDP putative causal pathways were constructed to reveal the regulatory role of these causal genes in brain diseases and behavioral traits, thereby providing a scientific basis for the development of new therapeutic targets.

[0120] The overall framework of this embodiment is as follows Figure 1 The overall results of the application of the method are shown in Figure 2 (a) with Figure 2 (b) As shown. This method predicted 254 and 217 genes that have causal effects on 112 IDPs and 26 DBs from eight cell types, respectively. Moreover, the analysis results showed extremely high cell type specificity, with more than 90% of gene-IDP associations and 80% of gene-DB associations existing only in a single cell type. In addition, up to 76.3% of the predicted gene-IDP pairs and 23.8% to 52.9% of the gene-DB pairs could be successfully replicated in external databases ( Figure 3), indicating that the causal markers identified by this method have high reproducibility.

[0121] This patent invention successfully identified many potential causal genes with pleiotropic effects (i.e. genes that have causal relationships with multiple phenotypes at the same time) ( Figure 4 ). The present invention found that within the same trait category (i.e., DB or IDP), the brain region volume group and the white matter microstructure group showed the greatest overlap (a total of 42 shared causal genes, accounting for up to 37.5%) ( Figure 4 a). For the DB category, this method revealed four cell type-specific genes that are causally related to the phenotypes of neurological diseases, behavioral cognitive phenotypes, and psychiatric disease groups: DDHD2 in excitatory neurons, MAP2K5 and XKR6 in inhibitory neurons, and LIN28B in excitatory neurons. These genes have been widely reported as potential causal factors for brain diseases (Dang X., et al. Mendelian Randomization Study Using Dopaminergic Neuron-Specific eQTL Identifies Novel Risk Genes for Schizophrenia. Mol Neurobiol, 2023) or have risks across psychiatric diseases (Nishioka M., et al. Systematic analysis of exonicgermline and postzygotic de novo mutations in bipolar disorder. Nat Commun. 2021; Mulvey B., et al. Transcriptional-regulatory convergence acrossfunctional MDD risk variants identified by massively parallel reporter assays. Transl Psychiatry. 2021). In addition, the genes shared within the DB category showed higher genetic conservation compared to the IDP category. At the same time, the genes shared between the brain region volume group and the DB category groups (neurological diseases, behavioral cognitive phenotypes, and psychiatric disorders) were more genetically conserved than the genes shared between the white matter microstructure group and the DB category groups ( Figure 4 b). Further gene enrichment analysis showed that the causal genes shared among these brain-related complex traits were mainly involved in a series of key biological processes, including memory and cognitive function, neurotransmitter regulation, protein binding, cell maintenance and signal transduction, and extracellular matrix dynamics ( Figure 5 ).

[0122] The results of this study found that the shared causal genes between DB and IDP categories showed different patterns in brain cell types ( Figure 6 For example, SCZ and brain region volume groups share genes extensively across multiple cell types, whereas neurodevelopmental disorders such as autism spectrum disorder (ASD), post-traumatic stress disorder (PTSD), and attention deficit hyperactivity disorder (ADHD) show only limited cell-type-specific gene sharing.

[0123] This patent successfully constructed four potential biological causal relationship pathways of cell type specific gene-IDP-DB, and seven reverse potential biological causal relationship pathways of cell type specific gene-DB-IDP ( Figure 7 , Table 1). Typical examples include: (1) ICA1L gene in excitatory neurons - anterior coronary radiation (ACR) - multiple sclerosis (MS) pathway, indicating that the expression of ICA1L gene in excitatory neurons may regulate the microstructure of ACR white matter, thereby affecting executive control ability and ultimately increasing the risk of MS; (2) ZSCAN31 gene in astrocytes - SCZ - posterior thalamic radiation (PTR) pathway, revealing that the transcription level of ZSCAN31 in astrocytes may affect the disease progression of SCZ and thus cause causal changes in the microstructure of PTR. The pairwise relationships between these variables have been widely supported by the literature, and this patented method further explores the potential and deeper causal biological connections between them.

[0124]

[0125]

[0126] Table 1

[0127] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0128] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for identifying causal genes of brain cells based on a Mendelian randomization framework, characterized in that: The following steps are involved: S1: Preliminarily collect, clean, and organize the eQTL data of the required brain cell types and the genome-wide association analysis data of brain diseases and behavioral traits, and preliminarily screen out genetic variants related to brain diseases and behavioral traits as initial instrumental variables; S2: Download the required prediction weights and covariance files from the external data repository and perform whole transcriptome association analysis to obtain the association between gene expression in specific cell types and brain diseases; S3: Based on the correlation between gene expression of specific cell types and brain diseases, multiple complementary two-stage causal inference methods are used to infer the causal relationship between gene expression of specific cell types and brain diseases; the multiple complementary two-stage inference methods include: SMR method, PMR-Egger method and GSMR method; S4: Test the robustness and repeatability of the causal relationship, and screen the causal relationship based on the test results to obtain relevant causal relationships; S5: Based on the relevant causal relationships, potential causal pathways between cell type-specific genes, brain imaging features, and brain diseases / behaviors are constructed; The robustness and repeatability of the causal relationship are tested, and the causal relationship is screened according to the test results, including the following steps: Step 1: Use the HEIDI test to test whether there is heterogeneity in the results of the SMR method; if yes, eliminate the causal relationship; if not, jump to step 2; Step 2: Use the HEIDI-outlier test to test whether there is heterogeneity in the results of the GSMR method; if yes, eliminate the causal relationship; if not, jump to step 3; Step 3: Use the Cragg-Donald statistic to calculate the statistic F of each tool and determine whether the validity of each tool variable is reliable; if not, eliminate the causal relationship; if yes, jump to step 4; Step 4: Verify the repeatability of genes of specific cell types corresponding to the causal relationship through an external database and a hypergeometric test to obtain verification results; wherein the verification results include qualified and unqualified; Step 5: Eliminate the causal relationships with unqualified verification results.

2. A method for identifying causal genes of brain cells based on a Mendelian randomization framework according to claim 1, characterized in that: The model in the PMR-Egger method is expressed as: ； in, Represented as an exposure factor vector of n1 dimension in gene expression data, represents a genotype matrix of n1×p dimensions, Represented as an n2-dimensional result factor vector, represents an n2×p genotype matrix, and are the intercepts on the x-axis and y-axis respectively, represents the exposure factor vector of n2 dimension in GWAS data, p-dimensional vector representing the effect size of the instrumental variable on the exposure factor, A scalar indicating the causal relationship between the exposure variable and the outcome variable, p-dimensional vector representing the horizontal multi-effect of instrumental variables on the outcome variable, , , Both represent residual vectors.

3. The method for identifying causal genes of brain cells based on a Mendelian randomization framework according to claim 1, characterized in that: In the GSMR method, the effects of exposure factors and outcome variables obtained by the GSMR model are expressed as: ； in, is the SNP effect estimate of the GWAS for a trait y, is the SNP effect estimate of the GWAS for a trait x, The variance of can be expressed as , statistics For inspection of significance.

4. The method for identifying causal genes of brain cells based on a Mendelian randomization framework according to claim 1, characterized in that: In S3, the multiple complementary two-stage causal inference methods also include the Wald ratio method.

5. The method for identifying causal genes of brain cells based on a Mendelian randomization framework according to claim 1, characterized in that: In S5, the bidirectional causal relationship between brain imaging features and brain diseases / behaviors is determined based on the inference methods of multiple instrumental variables: GSMR, PMR-Egger, IVW, MR-Egger, Weighted median, Simple mode, and Weighted mode.

Images