Federal analysis method for non-independent gene data

Abstract

The invention discloses a federal analysis method for non-independent gene data. The federal analysis method comprises the following steps: S1, a client performs identity encryption and anonymization processing on a user gene sequence data set; s2, performing data preprocessing on the phenotype data and the genotype data, and performing principal component analysis on the standardized genotype data; s3, adopting a regression model, taking the specific phenotype determined according to the research purpose as a dependent variable, taking other required features and the first five principal components as covariables, taking the gene locus genotype as an independent variable, and respectively calculating a genetic effect value of each gene locus on the specific phenotype; s4, calculating an identification code list of the corresponding sample, and comparing a sample proportion and a case proportion; and S5, the main server adopts a federated average algorithm to calculate variance matrixes and weight vectors of different genetic effect values for each gene locus, and aggregate genetic effect values, so that the data privacy problem and the sample overlapping problem when multiple GWAS analyses are combined are solved.

Description

Technical Field

[0001] The present invention relates to a federated analysis method for gene data, and in particular to a federated analysis method for non-independent gene data, belonging to the technical field of federated learning. Background Art

[0002] First, Federated Learning is a machine learning setting in which multiple participants (which can be mobile devices or entire organizations) collaborate to train models without sharing their data. This approach is particularly suitable for scenarios with strict requirements on data privacy. Applying federated learning methods to genome-wide association studies (GWAS) can aggregate GWAS data from multiple sources.

[0003] Second, genome-wide association studies (GWAS) aim to identify genetic variants associated with specific phenotypes (such as diseases). A single GWAS study may not be able to detect small effects or statistically significant associations due to limited sample size. By combining the results of multiple GWAS studies, statistical power can be increased and the ability to discover true genetic variants can be improved.

[0004] However, genomic data is highly sensitive because it contains detailed descriptions of an individual's genetic information. Since these data are often subject to strict privacy protection regulations, it may not be feasible or permitted to directly share GWAS data for traditional centralized meta-analysis. At the same time, if there is sample overlap between multiple GWAS data to be merged, the GWAS statistics need to be adjusted for sample overlap, otherwise incorrect GWAS statistics will be obtained, which will affect the analysis results and cause false positive results.

[0005] Therefore, it is urgent to develop a federated analysis method for non-independent genetic data. Summary of the invention

[0006] In view of the above existing technical problems, the present invention provides a federated analysis method for non-independent genetic data to achieve the technical purpose of solving the data privacy problem and sample overlap problem when merging multiple GWAS analyses.

[0007] To achieve the above technical objectives, the present invention provides a federated analysis method for non-independent gene data, using a federated analysis system for non-independent gene data mainly composed of multiple clients and a main server, including the following steps: S1. The client uses an identification code generation algorithm to encrypt and anonymize the user's gene sequence dataset to obtain an anonymized dataset; S2. Based on the anonymized data set, the phenotypic data and genotypic data are preprocessed to obtain standardized phenotypic data sets and genotypic data sets, and the standardized genotypic data are subjected to principal component analysis to obtain the first five principal components. S3. Based on the standardized phenotype data set, the model is trained locally. Other required features except genotype data are selected in the model training. A regression model is used. The specific phenotype determined according to the research purpose is used as the dependent variable, other required features and the first five principal components are used as covariates, and the genotype of the gene locus is used as the independent variable. The genetic effect value of each gene locus on the specific phenotype is calculated to obtain the GWAS data of the specific phenotype. S4. Based on the GWAS data of a specific phenotype, calculate the corresponding sample identification code list, as well as the control sample ratio and the case ratio, obtain the client data set, and transmit it to the main server; S5. The main server uses the federated averaging algorithm to calculate the variance matrix and weight vector of different genetic effect values ​​for each gene locus, as well as the aggregated genetic effect value, based on the data sets from multiple clients and itself.

[0008] In the present invention, the client in S1 uses an identification code generation algorithm to perform identity encryption and anonymization processing on the user gene sequence data set, including the following steps: S1-1. Clean and preprocess the data set to remove information that can directly identify individuals; S1-2, group the data set according to non-sensitive attributes to ensure that each equivalence class meets the L-diversity requirement; S1-3. Generate a unique salt value for each equivalence class; S1-4. Combine each recorded personal ID with the corresponding salt value and apply a hash function to generate a unique identification code; S1-5. Use the generated identification code to replace the original personal ID.

[0009] In the present invention, the step of preprocessing the phenotypic data in S2 comprises the following steps: The genetic data were quality-controlled and screened, and the phenotypic data were stored in a .csv file with the identification code as the first column and each phenotype as the other columns, with missing values ​​represented by NA.

[0010] In the present invention, the genotype data is preprocessed in S2, comprising the following steps: S2-2-1, perform quality control screening on the genotype data to obtain a quality-controlled genotype data set; S2-2-2. Standardize the genotype data set after quality control by subtracting the mean value from the original value of the genotype of the gene locus and then dividing it by its standard deviation.

[0011] The present invention further provides that the quality control screening of the gene data in S2-2-1 comprises the following steps: S2-2-11, remove loci with undetected rates higher than 2%; S2-2-12, Remove the P value of Hardy-Weinberg test less than The gene loci; S2-2-13, remove loci with minor allele frequencies below 1%; S2-2-14, remove gene loci with gene filling quality lower than 0.3; S2-2-15. Remove samples with an overall gene locus deletion rate greater than 5%.

[0012] Furthermore, the present invention, the GWAS data of the specific phenotype in S3 is represented in the form of a data table about site information, and its fields include the site ID, coordinate information, genetic effect value and its standard deviation, regression intercept, and P value of each gene site.

[0013] In the present invention, the calculation of the variance matrix of different genetic effect values ​​for each gene locus in S5 comprises the following steps: The different genetic effect values ​​of any gene locus in multiple client and main server data sets are recorded as ,and , Indicates a total of Datasets, Indicates Datasets, , , Respectively indicate that the gene locus is in the first, , Genetic effect sizes for each dataset; The standard deviation of the corresponding different genetic effect values ​​is recorded as , , , Respectively indicate that the gene locus is in the first, , The standard deviation of the genetic effect size for each data set; The variance matrix corresponding to different genetic effect values ​​is recorded as ; Then the variance matrix Line The column value is calculated as: In the formula, This gene locus represents The genetic effect size in the dataset, and ; , , Respectively represent The number of cases, controls, and total samples for each dataset; , , Respectively represent The number of cases, controls, and total samples for each dataset; , Respectively represent and The number of controls and cases that overlap in each data set.

[0014] The present invention further comprises , The calculation method is as follows: Based on the identification code lists and case ratios transmitted by multiple clients, the identification codes are compared to see if they overlap. The total number of is then multiplied by the corresponding control sample proportion and case proportion to obtain and .

[0015] In the present invention, the weight vector of different genetic effect values ​​is calculated for each gene locus in S5, comprising the following steps: The weight vector of different genetic effect values ​​of any gene locus in multiple client and main server data sets is recorded as , , , Respectively indicate that the gene locus is in the first, , The weights of the genetic effect sizes for each dataset; The calculation formula of the weight vector is: In the formula, represents a unit vector; Represents the variance matrix corresponding to different genetic effect values.

[0016] In the present invention, the calculation of the aggregate genetic effect value for each gene locus in S5 comprises the following steps: The aggregate genetic effect value of any gene locus in multiple client and main server data sets is recorded as ; The calculation formula for the aggregate genetic effect value is: In the formula, Indicates that the gene locus is in The genetic effect size of the dataset is Indicates that the gene locus is in The weights of the genetic effect sizes for a dataset.

[0017] In summary, the present invention solves the data privacy problem and sample overlap problem when merging multiple GWAS analyses, and has the following technical advantages: 1. Privacy protection: The method of the present invention ensures the localization of data, that is, the data is processed at the original location (such as a hospital or research institution), and only the updates of the model (such as gradient information) are shared under encryption or privacy protection measures, thereby effectively protecting the privacy of users.

[0018] 2. Compliance: Since the data does not leave its original storage location, the method of the present invention helps to comply with data protection regulations, which makes cross-institutional and cross-national GWAS research possible and promotes international cooperation and data sharing.

[0019] 3. Improve statistical power: By integrating data from multiple studies, the method of the present invention can increase statistical power. This helps to increase the probability of identifying genetic variants associated with diseases, thereby accelerating the discovery and research process of disease-related genes.

[0020] 4. Heterogeneous data integration: Different studies may use different genotyping platforms, sample processing methods or analysis processes. The method of the present invention can handle this data heterogeneity because it allows local calculation and model updating, thereby achieving effective integration and analysis of data from different sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The figure is a flow chart of the steps of the method of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0023] like Figure 1 As shown, taking the binary phenotype data aggregation algorithm of the WeGene cohort genome-wide association study (GWAS) as an example, the present invention provides a federated analysis method for non-independent genetic data, using a federated analysis system for non-independent genetic data mainly composed of multiple clients and a main server, including the following specific steps.

[0024] S1. The client generates an algorithm based on the unique ID, encrypts and anonymizes the user's gene sequence dataset to ensure data privacy, and publishes the anonymized dataset to provide a secure data foundation for subsequent analysis.

[0025] It should be noted that the user's gene sequence data is the gene test result obtained based on chip detection or second-generation sequencing technology. The technicians encrypt and anonymize the sample information based on the gene sequence data set produced by the sequencer or chip detection instrument. The technicians who conduct subsequent analysis only have access to the encrypted gene sequence data, but not the specific identity information of the sample.

[0026] It should also be noted that in the process of implementing the identification code generation algorithm, the present invention adopts a composite method of L-diversity algorithm and hash function combined with salt value algorithm to ensure the privacy protection and anonymity of data when it is released, which is introduced as follows.

[0027] First, L-diversity is a privacy protection technology that aims to prevent individuals from being uniquely identified through query results, thereby protecting data privacy. In this step, the L-diversity algorithm is the core of privacy protection. It effectively prevents the inference of sensitive information by ensuring that each equivalence class in the dataset contains at least L different sensitive values. An equivalence class is a grouping of one or more records that are identical in non-sensitive attributes, such as age, gender, zip code, etc. In this way, even if an attacker can determine that an individual belongs to a specific equivalence class, it cannot easily infer the individual's specific sensitive information, such as health status, income level, etc. In addition, L-diversity can also protect minority groups from having their sensitive information leaked due to data release, which is crucial to maintaining social fairness and respecting individual privacy.

[0028] Secondly, the use of hash functions combined with salt algorithms provides an additional layer of security for the generation of identification codes. Hash functions are a one-way encryption technology that can convert input data (such as personal ID) into a string of fixed length. This process is irreversible, that is, the original data cannot be deduced from the hash value. The salt value is a randomly generated string that is combined with the original data and then hashed, further increasing the complexity and randomness of the hash value, so that even the same input data will produce completely different hash values ​​due to different salt values.

[0029] In specific implementation, the composite method of combining the L-diversity algorithm with the hash function and the salt value algorithm is described as follows.

[0030] S1-1. Data preprocessing: Clean and preprocess the user's gene sequence data set to remove information that can directly identify an individual. Specifically, remove all information directly related to the sample's identity, such as name, address, age, health status, income level, etc.

[0031] S1-2, Apply L-diversity: Group the dataset according to non-sensitive attributes and ensure that each equivalence class meets the L-diversity requirement. Specifically, ensure that there are at least L different sensitive values ​​in each equivalence class.

[0032] S1-3. Generate salt value: Generate a unique salt value for each equivalence class. Specifically, this salt value is random and confidential and will not be disclosed to the public.

[0033] S1-4, Hash processing: Combine the personal ID of each record with the corresponding salt value, and apply the hash function to generate a unique identification code. Specifically, a unique identification code is assigned to each sample. This identification code is randomly generated and does not contain any information that can be traced back to the identity of the sample. This identification code will serve as the unique identifier of each record in the data set and the only reference for all subsequent data processing and analysis. In addition, all clients use the same identification code generation algorithm, and the identification code of the same sample is the same.

[0034] S1-5, Data Release: Finally, the generated identification code is used to replace the original personal ID. Specifically, the AES encryption algorithm is used for data transmission to ensure the security of data during transmission.

[0035] In a specific implementation, an example of the encrypted unique identification code is shown in Table 1 below.

[0036] Table 1. Example of encrypted unique identification code As can be seen from the above, in order to ensure the privacy and security of users, this step uses advanced identity encryption technology to anonymize the user's genetic data. Specifically, the gene sequence is first desensitized, and a composite method of L-diversity algorithm and hash function combined with salt value algorithm is used to assign a unique identification code to each user. The identification code is completely separated from the user's real identity information and is only used for subsequent data storage, analysis and research. Then the anonymized data set is released, and the encryption algorithm is used to encrypt the data during transmission to ensure the security of the data during transmission.

[0037] In particular, through the above composite method, this step not only achieves L-diversity of the data set and protects sensitive information from being inferred, but also ensures the irreversibility and uniqueness of the identification code through the hash function and salt value algorithm, thereby protecting the privacy of the data to the greatest extent. This method is particularly important in the field of sensitive data such as medical data and demographic survey data. It helps organizations comply with data protection regulations while also providing valuable data resources for social research and policy making.

[0038] S2. Based on the anonymized data set, the phenotypic data and genotypic data were preprocessed to obtain standardized phenotypic data and genotypic data sets to ensure data consistency and comparability. The principal components (PC values) were calculated for the standardized genotypic data sets, and the first five principal components were selected as important features for subsequent analysis.

[0039] It should be noted that this step preprocesses the genotype data and phenotype data respectively, so as to achieve data quality control and standardization, which is described in detail as follows.

[0040] S2-1. Based on the anonymized data set, the phenotypic data were preprocessed and stored in a .csv file with the identification code as the first column and each phenotype as the other columns. Missing values ​​were represented by NA to obtain a standardized phenotypic data set.

[0041] In this embodiment, the standardized phenotype data set includes 2501 phenotypes such as gender, age, smoking, cancer, etc., of which gender and age (or date of birth) must be provided, and other phenotypes are collected and sorted according to the requirements of the research project. Table 2 below lists 31 phenotypes out of the 2501 phenotypes.

[0042] Table 2. 31 phenotypes among 2501 phenotypes S2-2. Based on the anonymized data set, the genotype data is preprocessed to obtain a standardized genotype data set to ensure the consistency and comparability of the data.

[0043] S2-2-1. Perform quality control screening on the genotype data to obtain a quality-controlled genotype data set, including the following steps: S2-2-11. Remove sites with too high undetected rates. Set the threshold to 2%, which means that sites with undetected rates higher than 2% will be removed, because a higher undetected rate may indicate low quality of genotype data.

[0044] S2-2-12. Remove sites with too small P values ​​in the Hardy-Weinberg test and set the threshold to , because such loci may indicate genotyping or sequencing errors.

[0045] S2-2-13. Remove sites with too low minor allele frequency (MAF) and set the threshold to 1%, because such sites may cause large errors in statistical tests due to too low variation.

[0046] S2-2-14. Remove sites with low gene filling quality and set the threshold to 0.3, because such sites may cause errors in statistical test results due to low filling quality.

[0047] S2-2-15. Remove samples with too high overall site missing rate and set the threshold to 5%, because a higher non-detection rate may indicate low quality of genotype data.

[0048] In the specific implementation, genotype data is used as the original input data, including single nucleotide polymorphisms (SNPs), insertions and deletions (indels) and other sites. In addition, the genotype dataset is stored in a .vcf file or a .bed file, including site location information, variant detection results (genotype information), and detection quality information. An example of the genotype dataset after quality control obtained in this step is shown in Table 3 below.

[0049] Table 3. Example of genotype dataset after quality control S2-2-2. Standardize the genotype data set after quality control, that is, standardize the site (column) data, subtract the mean from the original genotype value of the gene locus and divide it by its standard deviation to obtain a standardized genotype data set in preparation for subsequent analysis.

[0050] In specific implementation, an example of the standardized genotype data set obtained in this step is shown in Table 4 below.

[0051] Table 4. Examples of standardized genotype datasets S2-3. Calculate the principal components (PC values) for the standardized genotype data and obtain the first five principal components (PCs).

[0052] It should be noted that principal component analysis (PCA) is a commonly used technique to adjust subpopulation effects in data. The basic principle is to identify mathematical vectors that maximize the difference in gene frequencies, thereby obtaining the principal components (i.e., eigenvectors) of the data. Through these principal components, the population structure of the sample can be evaluated. The purpose of applying principal component analysis is to correct the systematic deviation of allele frequencies between different subpopulations within a population due to non-random mating. This deviation is an important confounding factor in genome-wide association studies and may lead to false positive results.

[0053] In the specific implementation, PLINK software was used to perform principal component analysis on the standardized genotype data, and the first five principal components were selected as important features for subsequent analysis.

[0054] S3. Based on the standardized phenotype data set, the model is trained locally. Other required features except genotype data are selected in the model training. A regression model is used. The specific phenotype determined according to the research purpose is used as the dependent variable, other required features and the first five principal components are used as covariates, and the genotype of the gene locus is used as the independent variable. The genetic effect value of each gene locus on the specific phenotype is calculated to obtain the GWAS data of the specific phenotype. The specific steps are described as follows.

[0055] S3-1. Feature selection: Select other required features in model training except genotype data.

[0056] In the specific implementation, gender and age are selected as other required features and used as covariates together with the first five principal components. When doing GWAS, it is customary to put gender, age and the five principal components into the model, and usually no other variables need to be added. If there is a special research purpose, other required variables can also be added.

[0057] S3-2. Model construction: A regression model is used, with the specific phenotype determined according to the research purpose as the dependent variable (such as smoking, cancer, etc.), other required characteristics (gender, age), and the first 5 principal components as covariates, and the gene locus genotype as the independent variable. After regression analysis of each gene locus, the genetic effect value (regression slope) and its standard deviation, regression intercept, and corresponding P value (used to evaluate the significance of the association) are obtained, and they are organized with other relevant information (such as the site ID and coordinate information of each gene locus) into GWAS data for the specific phenotype.

[0058] It should be noted that if the dependent variable is a continuous variable, such as height, weight, blood sugar level, etc., the regression model adopts an ordinary multiple linear regression model; if the dependent variable is a binary variable, such as whether the person has diabetes, myopia, or smoking, etc., the regression model adopts a logistic regression model.

[0059] In the specific implementation, this embodiment uses myopia as the dependent variable, age, gender, and the first five principal components as covariates, and gene locus genotype as the independent variable, and adopts a logistic regression model for local model training.

[0060] In particular, the GWAS data of the specific phenotype is represented in the form of a data table about site information, and its fields include the site ID, coordinate information, and corresponding genetic effect value, regression intercept, standard deviation of genetic effect value, and P value corresponding to each gene site. In addition, the local GWAS data is subsequently transmitted to the main server.

[0061] S4. Based on the GWAS data of a specific phenotype, calculate the corresponding sample identification code list, as well as the control sample ratio and case ratio, organize the client's data set, and securely transmit it to the main server.

[0062] It should be noted that the proportion of control samples and the proportion of case samples refer to the proportion of different samples of the dependent variable. For example, if the dependent variable is myopia, the control sample is a non-myopia sample, and the case sample is a myopia sample. The proportion of control samples and the proportion of case samples are obtained by dividing the number of samples by the total number of samples. In addition, since only the proportion of control and pathological samples in the entire sample is counted instead of directly using the phenotypic data corresponding to the sample, the sample information security is further protected.

[0063] During specific implementation, the sample identification code list is stored as a .csv file, represented by a single column, and each row represents the identification code of a sample. For myopia, the total number of samples in this embodiment is 6133, including 975 control samples and 5158 case samples. The proportion of control samples and the proportion of cases are counted, and the statistical data are stored in the form of a .csv table. Finally, the client transmits the local data set to the main server, including the GWAS data of the specific phenotype, the corresponding sample identification code list, and the control sample proportion and case proportion information.

[0064] S5. Based on the data sets from multiple clients and the main server, the main server uses the federated averaging algorithm to calculate the variance matrix and weight vector of different genetic effect values ​​for each gene locus, as well as the aggregated genetic effect value.

[0065] In specific implementation, the federal analysis system for non-independent genetic data of the present invention includes multiple clients and a main server. The data set output by each client includes GWAS data of a specific phenotype, a corresponding list of sample identification codes, and information on the proportion of control samples and the proportion of cases. The main server itself stores a data set in the same data format as the client, which is mainly data transmitted from various clients in the past, and also includes GWAS data of a specific phenotype, a corresponding list of sample identification codes, and information on the proportion of control samples and the proportion of cases. It's just that the client is a larger database that covers more phenotypes and more samples.

[0066] Furthermore, when the main server receives the data set transmitted by the client, it aggregates the client's data set with its own data set. Furthermore, in order to consider data correlation, the aggregation algorithm is optimized, and the federated average algorithm is used to perform weighted summation of the parameters transmitted by the client. However, when aggregating parameters transmitted by different clients, or merging the client's parameters with the aggregated parameters, it is necessary to consider the impact of repeated samples on the model, that is, it is necessary to make adjustments for non-independent issues and aggregate the parameters of the non-independent data processing model.

[0067] It should be noted that in federated learning, the Federated Averaging algorithm is a common method for collaborative computing between multiple clients without the need for centralized data storage. This method is particularly suitable for privacy-sensitive scenarios, such as genomic association studies. The following describes how to calculate the variance matrix, weight vector, and aggregated genetic effect of the genetic effect value of each gene locus based on the Federated Averaging algorithm.

[0068] S5-1. Calculate the variance matrix of different genetic effect values ​​for each gene locus from multiple client and master server datasets.

[0069] Specifically, the different genetic effect values ​​of any gene locus in multiple client and main server data sets are recorded as ,and , Indicates a total of Datasets, Indicates Datasets, , , Respectively indicate that the gene locus is in the first, , Genetic effect sizes for the dataset.

[0070] The standard deviation of the corresponding different genetic effect values ​​is recorded as , , , Respectively indicate that the gene locus is in the first, , The standard deviation of the genetic effect value of each dataset. In addition, the genetic effect value and its standard deviation are obtained from the corresponding fields in the GWAS data of the specific phenotype in the client and the main server.

[0071] The variance matrix corresponding to different genetic effect values ​​is recorded as .

[0072] Then the variance matrix Line The column value is calculated as: in, In the formula, This gene locus represents The genetic effect size in the dataset, and ; , , Respectively represent The number of cases, controls, and total samples for each dataset; , , Respectively represent The number of cases, controls, and total samples for each dataset; , Respectively represent and The number of controls and cases that overlap in each data set.

[0073] When it comes to specific implementation, , The calculation method is based on the identification code list and case ratio transmitted by multiple clients. Because the identification code and phenotypic type data are not included in the data transmission, the identification code can be compared to see if there is overlap. The total number of is then multiplied by the corresponding control sample ratio and case ratio, and finally the and .

[0074] S5-2. Calculate the weight vector of different genetic effect values ​​for each gene locus from multiple client and main server data sets.

[0075] Specifically, the weight vector corresponding to the different genetic effect values ​​of any site from multiple client and main server data sets is recorded as , , , Respectively indicate that the gene locus is in the first, , The weights of the genetic effect sizes for a dataset.

[0076] The calculation formula of the weight vector is: in, is a unit vector; Represents the variance matrix corresponding to different genetic effect values.

[0077] In specific implementation, when calculating the weight for each gene locus, Perform inverse operations and use parallel operations to improve overall computing efficiency.

[0078] S5-3. Calculate the aggregate genetic effect value for each gene locus separately.

[0079] Taking any gene locus as an example, the data of the gene locus in the data sets from multiple clients and the main server are aggregated to obtain the aggregated genetic effect value of the locus.

[0080] Specifically, the aggregate genetic effect value of any gene locus in multiple client and main server data sets is recorded as ; The calculation formula for the aggregate genetic effect value is: In the formula, Indicates that the gene locus is in The genetic effect size of the dataset is Indicates that the gene locus is in The weights of the genetic effect sizes for a dataset.

[0081] In the specific implementation, each gene locus is aggregated and calculated, and finally the aggregated genetic effect value of each gene locus is obtained. In addition, the aggregated genetic effect value is stored in the main server, and each client can also pull the data on the main server for research and analysis.

[0082] In summary, the entire calculation process of the present invention only uses summary statistics to protect the individual data of actual users, and it is impossible to obtain the individual data of each individual from the summary statistics. That is, the data is processed at the original location (such as a hospital or research institution), and only the update of the model (such as gradient information) is shared under encryption or privacy protection measures, which solves the problems of data privacy and sample overlap when merging multiple GWAS analyses. In addition, by integrating data from multiple studies, the federated learning method of the present invention can increase statistical power and improve the probability of identifying genetic variations associated with diseases.

[0083] The technical solutions provided by the embodiments of the present invention are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, the technical solutions recorded in the aforementioned embodiments can be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the idea and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A federated analysis method for non-independent gene data, characterized in that: The federated analysis system for non-independent genetic data, which is mainly composed of multiple clients and a main server, includes the following steps: S1. The client uses an identification code generation algorithm to encrypt and anonymize the user's gene sequence dataset to obtain an anonymized dataset; S2. Based on the anonymized data set, the phenotypic data and genotypic data are preprocessed to obtain standardized phenotypic data sets and genotypic data sets, and the standardized genotypic data are subjected to principal component analysis to obtain the first five principal components. S3. Based on the standardized phenotype data set, the model is trained locally. Other required features except genotype data are selected in the model training. A regression model is used. The specific phenotype determined according to the research purpose is used as the dependent variable, other required features and the first five principal components are used as covariates, and the genotype of the gene locus is used as the independent variable. The genetic effect value of each gene locus on the specific phenotype is calculated to obtain the GWAS data of the specific phenotype. S4. Based on the GWAS data of a specific phenotype, calculate the corresponding sample identification code list, as well as the control sample ratio and the case ratio, obtain the client data set, and transmit it to the main server; S5. The main server uses the federated averaging algorithm to calculate the variance matrix and weight vector of different genetic effect values ​​for each gene locus, as well as the aggregated genetic effect value, based on the data sets from multiple clients and itself.

2. A federated analysis method for non-independent gene data according to claim 1, characterized in that: In S1, the client uses an identification code generation algorithm to perform identity encryption and anonymization processing on the user's gene sequence data set, including the following steps: S1-1. Clean and preprocess the data set to remove information that can directly identify individuals; S1-2, group the data set according to non-sensitive attributes to ensure that each equivalence class meets the L-diversity requirement; S1-3. Generate a unique salt value for each equivalence class; S1-4. Combine each recorded personal ID with the corresponding salt value and apply a hash function to generate a unique identification code; S1-5. Use the generated identification code to replace the original personal ID.

3. A federated analysis method for non-independent gene data according to claim 1 or 2, characterized in that: The phenotypic data is preprocessed in S2, including the following steps: The phenotypic data were stored in a .csv file with the identification code as the first column and each phenotype as the other columns, with missing values ​​represented by NA.

4. A federated analysis method for non-independent gene data according to claim 1 or 2, characterized in that: The genotype data is preprocessed in S2, including the following steps: S2-2-1, perform quality control screening on the genotype data to obtain a quality-controlled genotype data set; S2-2-2. Standardize the genotype data set after quality control by subtracting the mean value from the original value of the genotype of the gene locus and then dividing it by its standard deviation.

5. A federated analysis method for non-independent gene data according to claim 4, characterized in that: The quality control screening of the gene data in S2-2-1 includes the following steps: S2-2-11, remove loci with undetected rates higher than 2%; S2-2-12, P value of removing Hardy-Weinberg test is less than The gene loci; S2-2-13, remove loci with minor allele frequencies below 1%; S2-2-14, remove gene loci with gene filling quality lower than 0.3; S2-2-15. Remove samples with an overall gene locus deletion rate greater than 5%.

6. A federated analysis method for non-independent gene data according to claim 1, characterized in that: The GWAS data of a specific phenotype in S3 is represented in the form of a data table about site information, and its fields include the site ID, coordinate information, genetic effect value and its standard deviation, regression intercept, and P value of each gene site.

7. A federated analysis method for non-independent gene data according to claim 1 or 6, characterized in that: The calculation of the variance matrix of different genetic effect values ​​for each gene locus in S5 includes the following steps: The different genetic effect values ​​of any gene locus in multiple client and main server data sets are recorded as ,and , Indicates a total of Datasets, Indicates Datasets, , , Respectively indicate that the gene locus is in the first, , Genetic effect sizes for each dataset; The standard deviation of the corresponding different genetic effect values ​​is recorded as , , , Respectively indicate that the gene locus is in the first, , The standard deviation of the genetic effect size for each data set; The variance matrix corresponding to different genetic effect values ​​is recorded as ; Then the variance matrix Line The column value is calculated as: In the formula, This gene locus represents The genetic effect size in the dataset, and ; , , Respectively represent The number of cases, controls, and total samples for each dataset; , , Respectively represent The number of cases, controls, and total samples for each dataset; , Respectively represent and The number of controls and cases that overlap in each data set.

8. A federated analysis method for non-independent gene data according to claim 7, characterized in that: Said , The calculation method is as follows: Based on the identification code lists and case ratios transmitted by multiple clients, the identification codes are compared to see if they overlap. The total number of is then multiplied by the corresponding control sample proportion and case proportion to obtain and .

9. A federated analysis method for non-independent gene data according to claim 8, characterized in that: Calculating the weight vector of different genetic effect values ​​for each gene locus in S5 includes the following steps: The weight vector of different genetic effect values ​​of any gene locus in multiple client and main server data sets is recorded as , , , Respectively indicate that the gene locus is in the first and , The weights of the genetic effect sizes for each dataset; The calculation formula of the weight vector is: In the formula, represents a unit vector; Represents the variance matrix corresponding to different genetic effect values.

10. A federated analysis method for non-independent gene data according to claim 9, characterized in that: Calculating the aggregate genetic effect value for each gene locus in S5 includes the following steps: The aggregate genetic effect value of any gene locus in multiple client and main server data sets is recorded as ; The calculation formula for the aggregate genetic effect value is: In the formula, Indicates that the gene locus is in The genetic effect size of the dataset is Indicates that the gene locus is in The weights of the genetic effect sizes for a dataset.

Images