A method for mining stroke treatment targets

Abstract

The application provides a stroke treatment target mining method, aiming at solving the technical problems that the existing stroke target mining method lacks a systematic evaluation paradigm, is easily affected by confounding bias and ignores target safety, and the method comprises the following steps: obtaining a stroke prognosis GWAS data set and a target GWAS data set, identifying protein targets that have a causal relationship with stroke prognosis through two-sample Mendelian randomization analysis, and generating candidate target data; performing co-localization analysis, collision bias exclusion and independent queue verification reliability verification processing on the candidate target, generating verified target data; performing whole-phenotype group safety screening on the verified target, generating target safety evaluation information, and outputting stroke intervention target data based on the information; the application integrates genetic evidence and multi-level verification, improves the accuracy and reliability of target screening, and prospectively evaluates the clinical safety of the target, thereby providing an effective target mining scheme for stroke treatment.

Description

Technical Field

[0001] This invention relates to the fields of biomedicine and bioinformatics, and specifically to a method for identifying targets for stroke treatment. Background Technology

[0002] Stroke is characterized by high incidence, high recurrence rate, high mortality, and high disability rate, with a particularly prominent disease burden in my country. The mechanisms of stroke occurrence and development are complex and highly heterogeneous, and even with standardized treatment based on existing clinical guidelines, it remains difficult to effectively control its adverse prognosis. Therefore, in-depth analysis of the heterogeneous mechanisms of stroke and identification of precise intervention targets are of great significance for improving the prognosis of stroke patients.

[0003] Defining causal relationships is crucial for accurately screening therapeutic targets. While randomized controlled trials are considered the "gold standard" in causal inference, they often face challenges such as long-term follow-up, high costs, and ethical restrictions, making large-scale application in target discovery difficult. In recent years, Mendelian randomization (MR) has emerged as an alternative strategy, utilizing genetic variation as an instrumental variable. This approach effectively overcomes confounding bias and reverse causality issues in traditional observational studies, providing a new technical pathway for inferring the causal association between exposure factors and disease outcomes.

[0004] While Mendelian randomization has been widely applied to etiological studies of various complex diseases, its application in the specific context of stroke prognosis—systematically screening and validating therapeutic targets—remains immature. Existing research largely consists of fragmented, single-target analyses, lacking a unified, systematic, and standardized paradigm for target discovery and evaluation. For example, current methods often fail to adequately consider potential collision biases between targets and stroke risk, and do not systematically assess potential off-target effects or systemic adverse reactions after target intervention. Therefore, a systematic approach that integrates multidimensional evidence and balances target efficacy and safety is urgently needed to overcome the shortcomings of existing research paradigms in stroke prognosis target discovery and provide reliable technical support for clinical translation. Summary of the Invention

[0005] To address the aforementioned technical limitations, this invention proposes a method for identifying stroke treatment targets, enabling the systematic screening of intervention targets that combine effectiveness and safety from large-scale genetic data.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for identifying targets for stroke treatment, the method comprising: Step 1, Candidate Target Identification: Obtain the stroke prognosis GWAS dataset and target GWAS dataset, and identify protein targets that are causally associated with stroke prognosis through two-sample Mendelian randomization analysis to generate candidate target data. Step 2, Target Screening and Verification: Perform reliability verification processing on the targets in the candidate target data to generate verified target data; the reliability verification processing includes co-localization analysis verification, collision bias elimination, and independent queue verification; Step 3, Target Safety Assessment: Perform full phenotypic safety screening on the targets in the validated target data, generate target safety assessment information, and output stroke intervention target data based on the target safety assessment information.

[0007] Step 1 includes: Step 11: Construct instrumental variable dataset: Obtain target GWAS dataset, and screen out genetic variation sites that are independently related to the expression level of target protein from the target GWAS dataset to construct instrumental variable dataset; Step 12: Obtain prognostic outcome effect estimate data: Extract the effect estimate and standard error of each genetic locus and stroke prognosis from the stroke prognosis GWAS dataset to obtain the prognostic outcome effect estimate data. Step 13: Perform causal effect inference: Integrate the instrumental variable dataset with the prognostic outcome effect estimate data, use the two-sample Mendelian randomization method to calculate the causal effect estimate of each protein target on stroke prognosis, and screen candidate targets according to the preset significance threshold to generate candidate target data.

[0008] Furthermore, step 11, constructing the instrumental variable dataset, includes: Step 111: Identify conditionally independent cis-pQTLs: For each target protein, based on its corresponding cis-pQTL data, use precise stepwise conditional analysis to identify genetic variation sites that are independently associated with the expression level of the target protein, and generate an independent site dataset. Step 112: Screening overlapping sites: Integrate the independent site dataset with the stroke prognosis GWAS dataset, screen out the genetic variation sites that coexist in the two datasets as candidate instrumental variables, and generate a candidate instrumental variable dataset. Step 113, Allele Consistency Verification: Perform allele consistency verification on the genetic variation sites in the candidate instrumental variable dataset, screen out valid instrumental variables, and complete the construction of the instrumental variable dataset.

[0009] Further, step 13 includes: Step 131: Set the significance threshold; Step 132: Perform two-sample Mendelian randomization analysis based on the integrated data, and screen to obtain candidate target data that have a significant causal association with stroke prognosis. Specifically, for each protein, if the protein contains only one independent pQTL, the Wald ratio method is used to calculate the causal effect estimate; if the protein contains multiple independent pQTLs, the multiplicative random effects inverse variance weighted method is used to calculate the causal effect estimate.

[0010] As a preferred embodiment, step 13 further includes: performing causal effect inference on the prognostic outcome data of the total stroke population and different stroke subtype populations, and identifying subtype-specific targets according to preset judgment criteria to generate subtype-specific candidate target data.

[0011] Step 2 includes: Step 21: For each target in the candidate target data, perform colocalization analysis between the target GWAS dataset and the stroke prognosis GWAS data to determine whether they share the same causal genetic variant, and update the validation status field of the candidate target data according to the validation results to obtain the first target validation data. Step 22: Perform collision bias evaluation on the candidate targets in the first target verification data, and update the verification status field of the candidate target data according to the evaluation results to obtain the second target verification data. Step 23: Using stroke population cohort data independent of the stroke prognosis GWAS dataset, perform Mendelian randomization analysis on the candidate targets. If the causal effect direction of the candidate target in the independent cohort is consistent with the result of Step 1 and is statistically significant, then the target is determined to have passed the independent cohort verification.

[0012] Step 3 includes: Step 31: Obtain and map phenotypic data: Obtain the full phenotypic GWAS dataset. For each candidate target in the validated target data, extract the effect estimate and standard error of each genetic locus on each phenotype in the full phenotypic GWAS dataset to generate target phenotype effect mapping data. Step 32: Perform off-target effect screening: Based on the target phenotypic effect mapping data, perform Mendelian randomization analysis on each phenotype in the candidate target and the full phenotypic group GWAS dataset, calculate the estimated causal association effect and P-value, and generate off-target risk assessment data. Step 33: Generate and output stroke intervention target data: Integrate the target information that has passed Step 2 and whose off-target risk assessment is low to generate stroke intervention target data.

[0013] Compared with the prior art, the present invention has the following advantages: (1) Improved the accuracy and reliability of target screening, effectively eliminated non-causal targets through colocalization analysis and collision bias assessment, and significantly reduced the false positive rate; (2) The extrapolation and robustness of the technical solution are enhanced. The dual queue design of "discovery-verification" is adopted, which effectively reduces queue-specific bias. (3) It ensures the clinical safety of the intervention targets. Through systematic whole phenome screening, it prospectively identifies targets that may cause adverse reactions, thereby enhancing the clinical translational value. (4) It provides subtype-specific precision intervention strategies, which can identify targets that only play a role in specific stroke subtypes, and help promote the precision development of stroke treatment.

[0014] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0015] Figure 1 This is a flowchart of a stroke treatment target discovery method provided in an embodiment of the present invention.

[0016] Figure 2 A flowchart for candidate target identification provided in an embodiment of the present invention.

[0017] Figure 3 A flowchart illustrating the set of construction tool variables provided in an embodiment of the present invention.

[0018] Figure 4 A flowchart for target screening and verification provided in an embodiment of the present invention.

[0019] Figure 5 The figure shows animal experimental data on the efficacy of sIL6R in improving functional prognosis of ischemic stroke, as provided in the embodiments of the present invention. Detailed Implementation

[0020] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. To further understand the present invention, the present invention will be further described in detail below with reference to the preferred embodiments.

[0021] The following are explanations and definitions of the terms used in the embodiments of this invention: Genome-wide association study (GWAS) is a hypothetical-free research method that systematically examines millions of single nucleotide polymorphisms (SNPs) and other genetic variations across the entire genome in a specific population (such as patients and healthy controls) and uses rigorous statistical tests (with multiple validations) to identify genetic loci that are significantly associated with specific complex diseases or quantitative traits.

[0022] Protein quantitative trait loci (pQTLs) are specific genomic genetic variation sites that are statistically significantly associated with differences in protein expression or abundance levels among individuals, as identified at the population level through genome-wide association analysis (GWAS). They reveal the regulatory role of genetic variation on protein concentration and serve as an important functional bridge connecting genomic variation and phenotype.

[0023] The China Kadoorie Biobank (CKB) is a large-scale international prospective collaborative research project on chronic diseases, jointly conducted by the Chinese Center for Disease Control and Prevention, the Chinese Academy of Medical Sciences, Peking University, and the University of Oxford. The project covers over 510,000 adults aged 30-79 in 10 regions of China (5 urban and 5 rural areas). Through long-term follow-up and systematic collection of questionnaire information, anthropometric data, and biological samples, it aims to conduct in-depth research on the pathogenic factors, epidemiological patterns, and pathogenesis of various major chronic diseases (such as cardiovascular disease, cancer, and diabetes) affecting the health of the Chinese population from multiple dimensions, including genetics, environment, and lifestyle. The CKB dataset is based on East Asian population samples and was obtained through standardized genotyping, quality control, genotype imputation, and proteomic analysis. Protein quantitative trait loci are divided into cis-pQTLs (within ±1 Mb) and trans-pQTLs (outside ±1 Mb) according to their positional relationship with structural genes.

[0024] The Third China National Stroke Registry (CNSR-III) was a large-scale, nationwide, multicenter prospective cohort study conducted from August 2015 to March 2018. The study enrolled 15,166 patients with ischemic stroke or transient ischemic attack (TIA) and followed them for 3 months, 6 months, and 1-5 years. Three core prognostic outcomes were collected: stroke recurrence, functional scores, and death. Whole-genome sequencing and standardized quality control were performed on all enrolled patients. To control for the impact of hospitalization on prognosis, patients who received thrombolysis or thrombectomy were excluded before the analysis. Patients were classified into subtypes such as large artery atherosclerosis (LAA) and small vessel occlusion (SVO) according to the Causative Classification System (CCS) classification criteria. Genetic variant quality control retained biselenate variants with minor allele count (MAC) >20 and minor allele frequency (MAF) >0.001. A linear mixed model was used to assess the association between genetic variants and various prognostic outcomes, adjusting for age, sex, stroke history, hospital admission NIHSS score, and principal components of genetics.

[0025] Exact step-wise conditional analysis is a statistical method used for fine-grained localization in genetic association studies. Its core principle is that when multiple highly linked genetic variant signals are detected within a genomic region, this method iteratively incorporates the most significant association signal as a covariate into the statistical model at each step. This allows for conditional analysis of the remaining variants within the region, distinguishing and identifying multiple independent, statistically significant associated loci, rather than signals represented by a single linkage disequilibrium block.

[0026] The coloc method, short for Colocalization Analysis, is an algorithm based on a Bayesian statistical framework used to assess whether two associated genetic traits (such as a disease phenotype and a gene expression level) are driven by the same (or the same group of) causal variants within the same genomic region. This method integrates aggregated statistical data (such as GWAS and QTL results) of the two traits, calculates and compares the posterior probabilities of five mutually exclusive hypotheses to quantify the likelihood of a shared causal mechanism. The five hypotheses are: H0 (no association with either trait), H1 (associate only with trait one), H2 (associate only with trait two), H3 (different associated variants exist between the two traits), and H4 (the two traits share the same causal variant). Its core outputs are the posterior probabilities PP4 (the posterior probability of H4) and PP3 (the posterior probability of H3), thus providing statistical evidence for "pleiotropy" or "linkage disequilibrium," and is a key bioinformatics tool for elucidating the genetic mechanisms of complex traits and screening for priority disease-causing candidate genes.

[0027] CHANCE2 is a randomized controlled trial for ischemic stroke. Follow-up for stroke outcomes at 3 months and 1 year has been completed, with outcome definitions consistent with CNSR-III. CHANCE2 has completed whole-genome sequencing of 1,668 patients with laryngeal atrial fibrillation (LAA) and 1,582 patients with spontaneous ventricular vomiting (SVO). The sequencing platform and quality control standards are the same as those used in CNSR-III.

[0028] BioBank Japan (BBJ) is a national biobank project in Japan, initiated in 2003 and led by the Institute of Medical Science, University of Tokyo. It systematically collects, stores, and manages DNA samples and clinical information from approximately 270,000 patients and approximately 200,000 serum samples from collaborating medical institutions nationwide. The BBJ's whole-phenome GWAS data includes 220 GWAS phenotypes, categorized into three main groups: disease endpoints, biomarkers, and drug use. Among these, 159 are disease endpoints, covering infectious diseases, malignant tumors, hematological diseases, endocrine and metabolic diseases, neuropsychiatric diseases, cardiovascular diseases, respiratory diseases, digestive diseases, skin diseases, rheumatic and immune diseases, musculoskeletal diseases, urinary system diseases, and obstetric and gynecological diseases. 38 are biomarkers, including anthropometric indicators such as height, BMI, and blood pressure, as well as clinical laboratory indicators such as blood lipids, blood glucose, liver and kidney function, and complete blood count. Additionally, 23 categories of drug use history are included as phenotypes.

[0029] The implementation of this invention will be described in detail below with reference to the aforementioned terminology.

[0030] This invention provides a method for identifying stroke treatment targets, which identifies protein targets associated with stroke prognosis from large-scale genetic data, and performs multi-level information processing and evaluation on their effectiveness and safety. The method is executed by a digital data processing system, which includes at least: a causal target discovery module, a target screening and validation module, and a target safety assessment module. (Reference) Figure 1 The method specifically includes the following steps.

[0031] Step 1: Candidate Target Identification: The causal target discovery module performs a two-sample Mendelian randomization analysis process to process the stroke prognosis GWAS dataset and the target GWAS dataset, identify protein targets that have a potential causal relationship with stroke prognosis, and generate candidate target data.

[0032] Specifically, refer to Figure 2 Step 1 includes the following sub-steps.

[0033] Step 11: Construct the instrumental variable dataset. The causal target discovery module acquires the target GWAS dataset, which contains pQTL information for the target population. From the target GWAS dataset, genetic variant sites independently associated with the expression level of the target protein are screened out to construct the instrumental variable dataset.

[0034] In one embodiment, the target GWAS dataset uses protein quantitative trait locus data from the CKB dataset. The causal target discovery module selects genetic variation sites independently correlated with the expression levels of several target proteins from the CKB dataset as instrumental variables, forming an instrumental variable dataset. (Reference) Figure 3 Specific methods include: Step 111: Identify conditionally independent cis-pQTLs and generate an independent site dataset: For each target protein, the causal target discovery module uses the cis-pQTL data in its corresponding CKB dataset to call the precise stepwise conditional analysis method to identify genetic variation sites that are independently related to the expression level of the target protein and generate an independent site dataset.

[0035] Each record in the independent locus dataset includes at least: locus identifier, effect allele, non-effect allele, effect allele frequency, effect estimate, standard error, and p-value.

[0036] The specific method of the precise stepwise conditional analysis is as follows: (1) Using protein expression level as the dependent variable and allele dose at each genetic variation locus as the independent variable, the association between each cis-pQTL and protein expression level was examined one by one. The association with the smallest P value (representing the most significant association) and reaching the genome-wide significance threshold (P<5×10) was selected.-8 The site of ) is labeled as the first independent instrumental variable; (2) The allele dose of the identified site was included as a covariate in the regression model. Under the premise of controlling for the effect of the site, the association between the remaining cis-pQTL and the protein expression level was re-examined to identify the second independent site that reached the significance threshold. (3) Iteratively execute the above process: In each iteration, the allele doses of all previously identified loci are included as covariates in the regression model. Under the premise of controlling for the effects of these identified loci, the association between the remaining loci and protein expression levels is tested until no new loci reach the genome-wide significance threshold. Through the above iterations, a set of conditionally independent genetic variation loci is obtained and stored in the independent loci dataset.

[0037] Preferably, the regression model described in steps (2) and (3) above adopts a joint linear model, that is, the identified independent sites are simultaneously included as covariates in the linear regression equation for joint fitting. Since the genetic variation sites obtained through the above iterative process are relatively independent (i.e., conditionally independent), their original effect estimates can accurately reflect their independent contributions to protein expression levels. Therefore, their original effect estimates can be directly stored in the independent site dataset without additional effect size refitting.

[0038] In another embodiment, when only summary statistics for each pQTL are available but individual-level genotype data are unavailable, the precise stepwise conditional analysis method can employ the GCTA-COJO (Genome-wide Complex Trait Analysis - Conditional and Joint analysis) method in conjunction with reference population genomic data to perform stepwise regression analysis. Preferably, in the GCTA-COJO analysis, the conditional analysis window size is set to 10,000 kb (i.e., ±5,000 kb), the collinearity removal threshold is set to r² < 0.9, and the remaining parameters use the default settings of the cojo-slct program.

[0039] The number of proteins and cis-pQTLs screened in step 111 above depends on the specific content of the target GWAS dataset used (the CKB dataset in this embodiment). In this embodiment, as an example, based on the above processing, a total of 2,577 independent cis-pQTLs for 1,264 proteins were screened, forming the initial independent site dataset.

[0040] In practical applications, the number of proteins and pQTLs screened will vary depending on the dataset or proteomics platform used, but all follow the core processing logic of the precise stepwise conditional analysis method described above.

[0041] Step 112: Integrate with the stroke prognosis GWAS dataset and screen overlapping sites to generate a candidate instrumental variable dataset: The causal target discovery module integrates the independent site dataset obtained in step 111 with the stroke prognosis GWAS dataset, screens out the genetic variation sites that coexist in the two datasets as candidate instrumental variables, and generates a candidate instrumental variable dataset.

[0042] Each record in the candidate instrumental variable dataset includes at least: site identifier, target identifier, effect allele, non-effect allele, effect allele frequency, effect estimate and standard error in the target GWAS dataset, and effect allele, non-effect allele, effect allele frequency, effect estimate and standard error in the prognostic GWAS dataset.

[0043] In this embodiment, the stroke prognosis GWAS dataset uses the CNSR-III dataset.

[0044] After screening, 1,882 pQTLs of 1,038 proteins were retained as candidate instrumental variables, forming the candidate instrumental variable dataset.

[0045] Step 113, Allele Consistency Verification: The causal target discovery module performs allele consistency verification on the candidate instrumental variables selected in Step 112, selects valid instrumental variables, and completes the construction of the instrumental variable dataset.

[0046] Specifically, the causal target discovery module compares the effect alleles (i.e., alleles associated with protein expression levels or prognostic outcomes) and non-effect alleles of each genetic variant locus in the target GWAS dataset and the stroke prognosis GWAS dataset. If the allele information in the two datasets is inconsistent, the locus is excluded from the candidate instrumental variable dataset; if the locus is a palindromic SNP (A / T or C / G) that cannot distinguish between forward and reverse strands, it is also excluded. After validation, a valid instrumental variable dataset is obtained.

[0047] After this verification, 1,877 pQTLs of 1,038 proteins were finally included as valid instrumental variables, completing the construction of the instrumental variable dataset.

[0048] In this embodiment, the instrumental variable dataset is stored in a structured table format, and each record contains at least the following fields: target identifier, site identifier, effect allele, non-effect allele, protein effect estimate, protein effect standard error, protein effect P-value, prognostic effect estimate, prognostic effect standard error, and prognostic effect P-value.

[0049] Step 12: Obtain prognostic outcome effect estimates. The causal target discovery module extracts the effect estimates and standard errors of each genetic locus and stroke prognosis from the instrumental variable dataset constructed in Step 11 from the stroke prognosis GWAS dataset, thus obtaining prognostic outcome effect estimates.

[0050] Each record of the prognostic outcome effect estimate data includes at least: locus identifier, effect allele, non-effect allele, effect allele frequency, prognostic outcome effect estimate, and standard error of the prognostic outcome effect estimate.

[0051] The effect estimate reflects the strength and direction of the association between the genetic variant site and stroke prognosis. The standard error is used to calculate weights and confidence intervals in subsequent Mendelian randomization analysis.

[0052] In a preferred embodiment, the stroke prognosis GWAS dataset uses the CNSR-III dataset.

[0053] Step 13: Perform causal effect inference and generate candidate target data. The causal target discovery module integrates the instrumental variable dataset generated in Step 11 with the prognostic outcome effect estimates obtained in Step 12. Using a two-sample Mendelian randomization method, it calculates the causal effect estimate of each protein target on stroke prognosis and filters candidate targets based on a preset significance threshold, generating candidate target data. Specific methods include: Step 131: Set the significance threshold.

[0054] In one embodiment, the causal target discovery module sets the significance threshold to P<0.05 for subsequent significance judgment of causal effect inference.

[0055] It should be noted that the significance threshold is not fixed. Those skilled in the art can use other conventional statistical methods (such as Bonferroni correction, false positive rate (FDR) control) or set other reasonable significance levels based on factors such as the number of multiple tests, tolerance for false positive risk, and sample size in specific application scenarios to achieve the same purpose of this invention. All such adjustments do not depart from the scope of the technical solution of this invention.

[0056] Step 132: Perform two-sample Mendelian randomization analysis based on the integrated data, and screen to obtain candidate target data that have a significant causal association with stroke prognosis.

[0057] Specifically, for each protein in the instrumental variable dataset constructed in step 11, the causal target discovery module performs the following causal effect estimation and screening process: Step 1321: If the protein contains only one independent pQTL, the causal target discovery module uses the Wald ratio method to calculate the causal effect estimate, as shown in the following formula:

[0058]

[0059] In the formula, This is an estimate of the causal effect; and The values ​​and standard errors of the effect estimates of the instrumental variable and stroke prognosis obtained in step 12 are respectively: This is the effect estimate of the instrumental variable on protein expression level obtained in step 11.

[0060] Step 1322: If the protein contains multiple independent pQTLs, the causal target discovery module uses the multiplicative random effects inverse variance weighted method (IVW). First, it calculates the causal effect estimate of each instrumental variable based on the Wald ratio method as in step 1321, and then calculates the corresponding weights. The calculation method is as follows:

[0061] Then, the weighted average effect estimate is calculated as follows:

[0062] In the formula, For the first j The instrumental variables are estimated based on the Wald ratio method to obtain causal effect values. The corresponding weights are denoted as .

[0063] Step 1323: Compare the calculated causal effect estimate and its corresponding P value with the significance threshold (P<0.05) determined in step 131. If the threshold condition is met, the protein target is marked as a candidate target with a significant causal association with stroke prognosis, and candidate target data is generated.

[0064] The candidate target data includes at least: target identifier (such as protein name or ID), causal effect estimate, standard error of effect estimate, p-value, confidence interval, and significance indicator.

[0065] Preferably, based on step 13, the causal target discovery module can further perform the above-mentioned causal effect inference on the prognostic outcome data of the total stroke population and different subtype populations, and identify subtype-specific targets according to preset judgment criteria to generate subtype-specific candidate target data. The subtype-specific candidate target data, based on the candidate target data, adds a subtype identification field (such as LAA, SVO), and records the effect estimate and significance status of the target in each subtype and the total population.

[0066] Specifically, it includes: The causal target discovery module applies the analysis method described in step 132 to the prognostic outcome data of different stroke subtypes (including large artery atherosclerosis type LAA and small vessel occlusion type SVO), calculates the estimated causal effect of each protein target on the prognostic outcome in each subtype, and uses the same significance threshold determined in step 131 for screening.

[0067] After completing the causal effect inference for the total population and each subtype, if a certain protein target shows a significant causal association with prognostic outcome only in a single stroke subtype ( P <0.05), but no significant effect was observed in other subtypes or the total population. P If the value is greater than 0.05, the target is identified as a specific candidate target for that subtype, and corresponding identification information is generated, along with subtype-specific candidate target data. If the target is significantly associated in both the total population and the subtype population, it is considered a broad-spectrum target and does not possess subtype specificity.

[0068] In another embodiment, when individual-level genetic and phenotypic data are available, the causal target discovery module can also use a two-stage least squares method based on individual data to infer causal effects and generate corresponding candidate target data.

[0069] Specifically, it includes: (1) Based on the instrumental variable set constructed in step 11, calculate the polygenic score (PGS) of protein targets for each sample. The calculation formula is as follows:

[0070] In the formula, Let j be the genotype of the j-th instrumental variable (coded as 0, 1, or 2). The effect estimate for this instrumental variable (obtained in step 11).

[0071] (2) Using multi-gene scores as exposure variables, fit a model of association between them and prognostic outcomes.

[0072] Specifically, for survival outcomes such as recurrence and death, a Cox proportional hazards model (R survival package) was used for analysis. The model used polygenic scores as the primary exposure variable, adjusting for covariates such as age, sex, and principal genetic components. For binary functional scores (e.g., good and bad scores on the modified Rankin Scale), a logistic regression model (R glm function) was used, similarly adjusting for the aforementioned covariates. The causal effect of protein targets on stroke prognosis was estimated using these models, and the results were compared with the significance threshold determined in step 131 to screen candidate targets.

[0073] Step 2: The target screening and verification module performs reliability verification processing on the candidate targets identified in Step 1 to filter out false positive targets that are not causally related, and generates verified target data.

[0074] The reliability verification process includes co-location analysis verification, collision bias elimination, and independent queue verification. Specifically, refer to... Figure 4 Step 2 includes the following sub-steps.

[0075] Step 21: Colocalization analysis and verification to generate the first target verification data: For each target in the candidate target data, the target screening and verification module performs colocalization analysis between the target GWAS dataset and the stroke prognosis GWAS data to determine whether the two share the same causal genetic variant. Based on the verification results, the verification status field of the candidate target data is updated to obtain the first target verification data.

[0076] Specifically, the target screening and validation module extracts all genetic variations within a 100 kb range upstream and downstream of the target cis-pQTL, and performs Bayesian colocalization analysis using the coloc method. This analysis uses the default coloc parameters, where the prior probabilities of H1 and H2 are both 1 × 10⁻⁶. -4 The prior probability of H4 is 1×10. -5 The posterior probability of each hypothesis is calculated using Bayesian methods.

[0077] A screening threshold is set to retain only candidate targets with an H4 posterior probability greater than 0.5. Targets that fail the colocalization analysis are marked as "colocalization failed" in the target screening and verification module and excluded from the candidate target data. Targets that pass the verification are updated to "colocalization passed" in their verification status field, forming the first target verification data.

[0078] It should be noted that the H4 posterior probability threshold (0.5) used in the colocalization analysis above is a conventional conservative threshold. In practical applications, those skilled in the art can adjust this threshold appropriately based on the research objectives and the trade-off between false positives and false negatives (e.g., setting it to 0.8 when higher confidence is needed, and setting it to 0.2 for exploratory analyses) to achieve the best screening results.

[0079] Step 22: Collision bias elimination and generation of second target validation data: The target screening and validation module performs collision bias evaluation on the candidate targets in the first target validation data obtained in step 21, eliminates false positive associations between targets and stroke prognosis caused by the influence of targets on stroke susceptibility, and updates the validation status field of the candidate target data according to the evaluation results to obtain the second target validation data.

[0080] Understandably, since stroke occurrence may be a collider between target proteins and stroke prognosis, if a causal relationship exists between the target protein and stroke occurrence, collider bias may be introduced when conducting analyses using stroke patients as research subjects. To avoid such bias interfering with the determination of causal relationships, collider bias assessment of candidate targets is necessary.

[0081] Specifically, the target screening and validation module acquires the stroke occurrence GWAS dataset, which contains stroke risk association information for the target population. For each candidate target in the first target validation data, the target screening and validation module uses the same Mendelian randomization analysis method as in step 132 to evaluate the causal association between the target and the risk of stroke occurrence.

[0082] If the target has a significant causal association with the risk of all-cause stroke or any stroke subtype ( P If the value is less than 0.05, the target screening and verification module marks its verification status field as "collision bias risk" and excludes it from the candidate target data; if the analysis results are not significant ( P If the value is ≥0.05, the target point is retained, and the verification status field is updated to "collision bias exclusion passed", forming the second target point verification data.

[0083] It should be noted that the collision bias assessment above uses P<0.05 as the threshold for determining a significant causal association. This criterion is relatively sensitive and may lead to many candidate targets being judged as having a risk of bias and thus being eliminated, especially those targets that have real biological effects on both stroke occurrence and prognosis. In practical applications, those skilled in the art can make adaptive adjustments to the significance threshold based on the specific research objectives, balancing the control of collision bias risk with the retention of potentially effective targets. For example, when it is necessary to retain candidate targets that may simultaneously affect stroke occurrence and prognosis, a more conservative significance threshold can be used (such as P<0.05 / total number of targets assessed, or P<0.05 / number of stroke-related datasets, after Bonferroni correction) to improve the target retention rate and achieve the best screening effect.

[0084] In one embodiment, GWAS data on stroke incidence in East Asian populations from the GIGASTRAKOGE Consortium (GWASCatalog numbers GCST90104544 to GCST90104548) were used, covering all-cause stroke and subtypes such as ischemic stroke, large artery stroke, small vessel stroke, and cardioembolic stroke.

[0085] Optionally, to more comprehensively correct for collision bias, a multivariate Mendelian randomization method can be used in step 22 to reassess the causal effects of the target and prognosis after adjusting for stroke occurrence status, in order to verify the stability of the results. Specifically, the implementation of this multivariate Mendelian randomization method includes: (1) Constructing a joint instrumental variable dataset: Based on the instrumental variable dataset constructed in step 11, further incorporate independent genetic variations associated with the risk of stroke as instrumental variables; wherein, the independent genetic variations exclude collinearity among instrumental variables by setting a linkage disequilibrium parameter (e.g., r² < 0.01) to ensure conditional independence among instrumental variables; (2) Update the prognostic outcome effect estimates: Based on the joint instrumental variable dataset constructed in step (1), re-extract or calculate the effect estimates of each instrumental variable on stroke prognostic outcomes and their standard errors, so as to update the prognostic outcome effect estimates data obtained in step 12; (3) Perform multivariate causal inference: Based on the updated effect estimates, the main analysis is performed using the multivariate inverse variance weighted method (MVMR-IVW) during the causal inference process in step 13. Optionally, when there are many ineffective instrumental variables, the sensitivity analysis can be performed using the MVMR-cML method based on the mixed likelihood framework. The maximum number of iterations in the MVMR-cML analysis is set to 1,000, and the convergence tolerance is 1×10⁻⁶. -6The optimal mixture likelihood model is selected using the Bayesian Information Criterion (BIC). Under conditions that jointly adjust for stroke occurrence status, the causal effect between the target and stroke prognosis is re-inferred, thereby further correcting for collision bias and improving the reliability and robustness of the causal inference results.

[0086] Step 23: Independent cohort validation and generation of validated target data: To reduce the impact of cohort-specific bias and random error on target selection, the target selection validation module uses a stroke population cohort independent of the stroke prognosis GWAS dataset used in Step 1 to perform Mendelian randomization analysis on candidate targets to further confirm the robustness and extrapolation potential of target associations and generate validated target data.

[0087] In one embodiment, the CHANCE2 cohort is used as an independent validation cohort. Based on the stroke prognostic GWAS data from the CHANCE2 cohort, the target screening and validation module uses the same instrumental variable screening method and Mendelian randomization analysis method as in steps 11 to 13 to evaluate the association between candidate targets in the second target validation data and stroke prognostic outcomes in this independent cohort.

[0088] The target screening and validation module judges based on validation criteria: if the causal effect direction of the candidate target in the independent cohort is consistent with the result obtained in step 132, and the meta-analysis results are statistically significant (e.g., P If the value is less than 0.05, the target is determined to have passed the independent queue verification, and its verification status field is updated to "Independent queue verification passed"; otherwise, its verification status field is marked as "Independent queue verification failed", and it is excluded from the candidate target data; the candidate target data that has passed this step is stored in the verified target data.

[0089] The reliability of the target verified through this step is further confirmed.

[0090] In another embodiment, when the sample size of the independent cohorts allows, the above validation can be performed on the total population and each subtype separately to ensure the consistency of the effect of subtype-specific targets in independent subtype populations.

[0091] Step 3: The target safety assessment module performs a safety screening step on the candidate targets that have passed the validity verification in Step 2, and generates target safety assessment information.

[0092] The target safety assessment module performs a full phenome safety scan on the candidate targets validated in step 2. By systematically evaluating the potential causal associations between candidate targets and multiple clinical phenotypes, it identifies targets that may cause adverse reactions or off-target effects and generates target safety assessment information.

[0093] The safety assessment in this step can provide genetic safety evidence for the clinical translation of the target.

[0094] Specifically, step 3 includes the following sub-steps.

[0095] Step 31: Acquisition and mapping of phenotypic data to generate target phenotypic effect mapping data. The target safety assessment module acquires the full phenotypic GWAS dataset. For each candidate target in the validated target data, it extracts the effect estimates and standard errors of each genetic locus on each phenotype in the full phenotypic GWAS dataset from the instrumental variable dataset constructed in Step 11, generating target phenotypic effect mapping data, thereby establishing the effect mapping relationship between the target and each phenotype. If a genetic locus is not included in the full phenotypic GWAS dataset, it is removed from the instrumental variable set of the current target and will no longer be included in the safety assessment of that target. Optionally, a surrogate SNP with high linkage disequilibrium (e.g., r²>0.8) can be used as a substitute, and the substitution relationship and linkage disequilibrium parameters are noted in the assessment record.

[0096] The whole-phenome GWAS dataset contains genetic association information for multiple clinically relevant phenotypes.

[0097] In a preferred embodiment, the full phenotypic GWAS dataset uses GWAS data from the Japan Biobank.

[0098] Step 32: Perform off-target effect screening and generate off-target risk assessment data.

[0099] The target safety assessment module performs Mendelian randomization analysis on each phenotype in the candidate target and the full phenotype GWAS dataset based on the target phenotypic effect mapping data to assess the potential off-target effects or adverse reactions of intervention in the target. Specifically, using the same two-sample Mendelian randomization analysis method as in step 132, the target safety assessment module calculates the estimated potential causal association effect and corresponding p-value for each candidate target and each phenotype in the full phenotype GWAS dataset, generating off-target risk assessment data.

[0100] The off-target risk assessment data shall include at least: target identifier, phenotypic identifier, causal effect estimate, P-value, and risk level identifier.

[0101] The target safety assessment module sets a multiple test correction threshold to control the false positive rate. In one embodiment, the full phenotypic GWAS dataset uses GWAS data from the Japan Biobank, including 220 phenotypes. The target safety assessment module uses Bonferroni correction, setting the significance threshold to [value missing]. P <0.05 / 220 ( P<2.27×10 -4 If the association analysis results of a target with any phenotype reach the correction threshold, the target is judged to have potential off-target risk or adverse reaction risk, and its risk level is marked as "high risk"; if the association analysis results of a target with all phenotypes do not reach the correction threshold, the target is judged to have good genetic safety, and its risk level is marked as "low risk".

[0102] Optionally, to more granularly assess safety, differentiated correction thresholds can be set based on the importance of the phenotype or clinical concern, for example, using stricter thresholds for phenotypes associated with serious adverse reactions (e.g., P <0.05), a relatively lenient threshold is used for minor phenotypes (e.g., Bonferroni correction).

[0103] In another embodiment, the False Detection Rate (FDR) control method can be used instead of Bonferroni correction to improve detection efficiency while controlling the false positive rate and identifying potential safety risk signals.

[0104] In another embodiment, when a target has a nominally significant association with a specific phenotype (such as liver function indicators, cardiovascular events, etc.) P Even if the value is less than 0.05 but fails the multiple validation correction, a risk warning message will still be output for further attention during the subsequent drug development stage.

[0105] It should be noted that the above-described multiple test correction methods and thresholds are merely exemplary schemes. Those skilled in the art can adopt other reasonable correction strategies based on the number of phenotypes in the full phenotypic GWAS dataset, the clinical importance of each phenotype, and the tolerance for false positives / false negatives.

[0106] Step 33: Generate and output stroke intervention target data.

[0107] The target safety assessment module integrates target information that simultaneously meets the following conditions to generate stroke intervention target data: (1) It was identified in step 1 as having a significant causal association with stroke prognosis (i.e., the significance level in the candidate target data is marked as "significant"). (2) Through the colocalization analysis, collision bias elimination and independent queue verification in step 2 (i.e., the verification status field in the target data after verification is all in the pass status). (3) No obvious off-target effect was shown in step 32 (the risk level of the off-target risk assessment data was marked as "low risk").

[0108] The output target information should include at least: target identifier, stroke prognostic outcome identifier, causal effect estimate and confidence interval of target and stroke prognostic outcome, target subtype specificity identifier, and target safety assessment results.

[0109] The safety assessment results include at least: the total number of phenotypes assessed for the target, the number and list of phenotypes that reached the nominal significance level (P<0.05), the number and list of phenotypes that remained significant after multiple test correction, and a risk level conclusion (i.e., 'high risk' or 'low risk') that comprehensively determines whether the target has potential off-target risks.

[0110] The safety assessment results provide comprehensive information support for subsequent drug development and precision intervention strategies.

[0111] Through the above information processing flow, this invention can systematically mine potential targets for stroke intervention from genetic data, comprehensively evaluate their effectiveness and safety, and generate structured stroke intervention target data, providing reliable information support for clinical translation.

[0112] To make the objectives, technical solutions, and advantages of this invention clearer, the following uses sIL6R as a representative target to elaborate on the complete processing flow of the stroke treatment target discovery method described in this invention. Those skilled in the art should understand that this specific embodiment is only for explaining this invention and is not intended to limit the scope of protection of this invention.

[0113] Example: Stroke treatment target discovery process targeting sIL6R Step 1: Candidate Target Identification Step 11: Construct the instrumental variable dataset Input the GWAS dataset of sIL6R (set the significance threshold P<5×10). -8 The dataset contains 579 cis-pQTLs. After conditional analysis, the output is an instrumental variable dataset containing 3 independent cis-pQTLs (as shown in Table 1).

[0114] Table 1. Dataset of independent instrumental variables for target sIL6R

[0115] Note: SNP, EA, NEA, and FREQ represent the locus identifier, effect allele, non-effect allele, and effect allele frequency, respectively; BETA, SE, and P represent the effect estimate, standard error, and P-value, respectively.

[0116] Subsequently, the GWAS dataset of modified Rankin Scale (mRS) scores (3-6 points vs. 0-2 points) 3 months after small vessel occlusive stroke in the CNSR-III cohort was input, and two overlapping sites were selected as candidate instrumental variables (as shown in Table 2).

[0117] Table 2. Candidate instrumental variable dataset for target sIL6R

[0118] Note: SNP, EA, NEA, and FREQ represent the locus identifier, effect allele, non-effect allele, and effect allele frequency, respectively; BETA, SE, and P represent the effect estimate, standard error, and P-value, respectively.

[0119] For the above candidate instrumental variables, the allelic direction consistency between the target and outcome GWAS was checked, and the final output dataset contained two valid instrumental variables, rs12083537 and rs2228145 (as shown in Table 3).

[0120] Table 3. Effective instrumental variable dataset for target sIL6R

[0121] Note: SNP, EA, NEA, and FREQ represent the locus identifier, effect allele, non-effect allele, and effect allele frequency, respectively; BETA, SE, and P represent the effect estimate, standard error, and P-value, respectively.

[0122] Step 12: Obtain prognostic outcome effect estimates data Step 13: Causal Effect Inference The causal target discovery module inferred causal effects based on the aforementioned effective instrumental variable dataset, outputting a causal effect value of -0.391 (P=0.027) between sIL6R expression level and poor functional outcomes (mRS 3-6) in small vessel occlusive stroke patients at 3 months. Based on this significant result, sIL6R was labeled as a candidate target.

[0123] Step 2: Target screening and verification Step 21: Colocalization Analysis Colocalization analysis was performed using the sIL6R GWAS dataset and GWAS data of mRS scores at 3 months post-stroke in the CNSR-III cohort. The results showed that the posterior probability of hypothesis H4 (i.e., the target and outcome share a causal variant) was 0.576, and the colocalization was deemed "passed".

[0124] Step 22: Collision Bias Elimination We input GWAS data on stroke incidence in the East Asian population from the GIGASTROKE Consortium (GWAS Catalog numbers GCST90104544 to GCST90104548) to assess the causal association between sIL6R and the risk of various stroke types. The results showed that the p-values ​​for causal analysis of sIL6R with all-cause stroke and each subtype of stroke were all greater than 0.05 (as shown in Table 4), indicating that there were no false positive associations caused by the target influencing stroke susceptibility, and the collision bias was ruled out.

[0125] Table 4. Causal relationship between target sIL6R and stroke occurrence

[0126] Step 23: Independent Queue Verification Independent validation was performed using the sIL6R GWAS dataset and GWAS data of 3-month mRS scores for small vessel occlusion stroke from the CHANCE2 study. The results showed a causal effect value of -0.237, consistent with the effect direction in the CNSR-III cohort. Meta-analysis yielded a pooled effect value of -0.360 and a p-value of 0.023, indicating successful independent validation.

[0127] Step 3: Target safety assessment GWAS data for 220 phenotypes were input from the BBJ database. The target safety assessment module identified a potential causal association between sIL6R and 14 phenotypes (P<0.05, as shown in Table 5). However, after Bonferroni multiple test correction, none of the above associations reached the significance threshold (P>0.05 / 220). This result suggests that intervention with sIL6R does not pose a risk of serious adverse reactions, and the system outputs a risk level of "low risk" for sIL6R.

[0128] Table 5. Results of Mendelian randomization analysis showing nominally significant associations (P<0.05) between the target sIL6R and each phenotype.

[0129] Based on the above analysis, this algorithm ultimately determined sIL6R to be a candidate therapeutic target for improving the 3-month functional prognosis of patients with small vessel occlusion stroke.

[0130] To further verify the reliability of the algorithm's mining results, animal experiments were conducted. The experiments confirmed that sIL6R can significantly promote the recovery of cerebral blood flow, promote the recovery of neurological function, and improve survival rate in mice with cerebral ischemia-reperfusion injury (see appendix). Figure 5 ).

[0131] Among them, the appendix Figure 5 Animal studies have confirmed that sIL6R improves functional prognosis in ischemic stroke: Figure A shows the laser speckle cerebral blood flow before modeling, after embolization, after embolization removal, and 1 day (1d) and 3 days (3d) after injection of recombinant sIL6R protein; Figure B shows the comparison of laser speckle cerebral blood flow before and after modeling; Figure C shows the comparison of laser speckle cerebral blood flow 1d and 3d after drug administration; Figure D shows the changes in neurological function scores from 0 to 7 days after modeling; Figure E shows the survival rate curve from 0 to 7 days after modeling (**** represents P<0.0001; *** represents P<0.001; ** represents P<0.01; * represents P<0.05; MCAO: middle cerebral artery embolism mouse model).

[0132] The experimental evidence above fully demonstrates that the sIL6R target discovered by this algorithm has a reliable prognostic effect on stroke function.

[0133] As one embodiment, the method described in this invention can be implemented in software and / or a combination of software and hardware, for example, using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device.

[0134] The method described in this invention can be implemented as a software program, which can be executed by a processor to achieve the steps or functions described above. Similarly, the software program (including associated data structures) can be stored in a computer-readable recording medium, such as RAM memory, magnetic or optical drives, floppy disks, and similar devices.

[0135] In addition, some steps or functions of the method described in this invention can be implemented in hardware, for example, as a circuit that works with a processor to perform the various steps or functions.

[0136] Furthermore, a portion of the methods described in this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. The program instructions invoking the methods described in this invention can be stored in a fixed or removable recording medium, and / or transmitted via a data stream in a broadcast or other signal carrying medium, and / or stored in the working memory of a computer device operating according to the program instructions.

[0137] As one embodiment, the present invention also provides an apparatus comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the apparatus is triggered to run a method and / or technical solution based on the foregoing plurality of embodiments.

[0138] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0139] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0140] Furthermore, the technical solutions of the various embodiments can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0141] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for identifying target points in stroke treatment, characterized in that, The method includes: Step 1, Candidate Target Identification: Obtain the stroke prognosis GWAS dataset and target GWAS dataset, and identify protein targets that are causally associated with stroke prognosis through two-sample Mendelian randomization analysis to generate candidate target data. Step 2, Target Screening and Verification: Perform reliability verification processing on the targets in the candidate target data to generate verified target data; the reliability verification processing includes co-localization analysis verification, collision bias elimination, and independent queue verification; Step 3, Target Safety Assessment: Perform full phenotypic safety screening on the targets in the validated target data, generate target safety assessment information, and output stroke intervention target data based on the target safety assessment information.

2. The method according to claim 1, characterized in that, Step 1 includes: Step 11: Construct instrumental variable dataset: Obtain target GWAS dataset, and screen out genetic variation sites that are independently related to the expression level of target protein from the target GWAS dataset to construct instrumental variable dataset; Step 12: Obtain prognostic outcome effect estimate data: Extract the effect estimate and standard error of each genetic locus and stroke prognosis from the stroke prognosis GWAS dataset to obtain the prognostic outcome effect estimate data. Step 13: Perform causal effect inference: Integrate the instrumental variable dataset with the prognostic outcome effect estimate data, use the two-sample Mendelian randomization method to calculate the causal effect estimate of each protein target on stroke prognosis, and screen candidate targets according to the preset significance threshold to generate candidate target data.

3. The method according to claim 2, characterized in that, Step 11 involves constructing the instrumental variable set, including: Step 111: Identify conditionally independent cis-pQTLs: For each target protein, based on its corresponding cis-pQTL data, use precise stepwise conditional analysis to identify genetic variation sites that are independently associated with the expression level of the target protein, and generate an independent site dataset. Step 112: Screening overlapping sites: Integrate the independent site dataset with the stroke prognosis GWAS dataset, screen out the genetic variation sites that coexist in the two datasets as candidate instrumental variables, and generate a candidate instrumental variable dataset. Step 113, Allele Consistency Verification: Perform allele consistency verification on the genetic variation sites in the candidate instrumental variable dataset, screen out valid instrumental variables, and complete the construction of the instrumental variable dataset.

4. The method according to claim 2, characterized in that, Step 13 further includes: Step 131: Set the significance threshold; Step 132: Perform two-sample Mendelian randomization analysis based on the integrated data, and screen to obtain candidate target data that have a significant causal association with stroke prognosis.

5. The method according to claim 4, characterized in that, In step 132, for each protein, if the protein contains only one independent pQTL, the Wald ratio method is used to calculate the causal effect estimate; if the protein contains multiple independent pQTLs, the multiplicative random effects inverse variance weighted method is used to calculate the causal effect estimate. In step 132, the calculated causal effect estimate and its corresponding P value are compared with the significance threshold determined in step 131. If the threshold condition is met, the protein target is marked as a candidate target with a significant causal association with stroke prognosis, and candidate target data is generated.

6. The method according to claim 2, characterized in that, Step 13 further includes: performing causal effect inference on the prognostic outcome data of the total stroke population and different stroke subtype populations, and identifying subtype-specific targets according to preset judgment criteria to generate subtype-specific candidate target data.

7. The method according to claim 2, characterized in that, In step 13, when individual-level genetic and phenotypic data are available, a two-stage least squares method based on individual data is used to infer causal effects. Specifically, this includes: calculating a multigene score for the protein target for each sample based on the constructed set of instrumental variables, fitting the multigene score as an exposure variable to a model associated with the prognostic outcome, and estimating the causal effect of the protein target on the prognostic outcome of stroke.

8. The method according to claim 1, characterized in that, Step 2 includes: Step 21: For each target in the candidate target data, perform colocalization analysis between the target GWAS dataset and the stroke prognosis GWAS data to determine whether they share the same causal genetic variant, and update the validation status field of the candidate target data according to the validation results to obtain the first target validation data. Step 22: Perform collision bias evaluation on the candidate targets in the first target verification data, and update the verification status field of the candidate target data according to the evaluation results to obtain the second target verification data. Step 23: Using stroke population cohort data independent of the stroke prognosis GWAS dataset, perform Mendelian randomization analysis on the candidate targets. If the causal effect direction of the candidate target in the independent cohort is consistent with the result of Step 1 and is statistically significant, then the target is determined to have passed the independent cohort verification.

9. The method according to claim 1, characterized in that, The stroke prognosis GWAS dataset used is the dataset from the Third National Stroke Registry Study of China; The target GWAS dataset used is the dataset from the Chinese Chronic Disease Prospective Study. The independent queue verification uses the CHANCE2 queue dataset; The whole-phenome safety screening used whole-phenome GWAS data from the Japan Biobank.

10. The method according to claim 1, characterized in that, Step 3 includes: Step 31: Obtain and map phenotypic data: Obtain the full phenotypic GWAS dataset. For each candidate target in the validated target data, extract the effect estimate and standard error of each genetic locus on each phenotype in the full phenotypic GWAS dataset to generate target phenotype effect mapping data. Step 32: Perform off-target effect screening: Based on the target phenotypic effect mapping data, perform Mendelian randomization analysis on each phenotype in the candidate target and the full phenotypic group GWAS dataset, calculate the estimated causal association effect and P-value, and generate off-target risk assessment data. Step 33: Generate and output stroke intervention target data: Integrate the target information that has passed Step 2 and whose off-target risk assessment is low to generate stroke intervention target data.

Images