A gene selection system based on homomorphic encryption technology

Abstract

This invention discloses a gene selection system based on homomorphic encryption technology, comprising: an input module for inputting data; a selection module for allocating the input data to corresponding prediction models and obtaining all information from the data and pedigree files in a one-step process; an execution prediction module for reconstructing the computation array according to the configuration information provided by the selection module to implement the algorithms for each stage of homomorphic encryption; a display module for storing the initial plaintext, intermediate ciphertext, and plaintext obtained after decryption of the homomorphic encryption algorithm in the execution prediction module; and a cloud output module for outputting the encrypted data to a cloud platform. This invention also develops a local client, HEcryptor, based on the homomorphic encryption algorithm to work with the HEGS platform. Users can encrypt data locally before uploading to ensure complete data security. The encrypted data can be used for both single-field and cross-field genome selection, meeting various user needs.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of encryption algorithm, and particularly discloses a gene selection system based on homomorphic encryption technology. BACKGROUND

[0002] Genomic data privacy threats

[0003] (1) Individual identification: Many studies using genomic data also contain many other types of data such as demographic information, behavioral information, molecular or clinical phenotype information, but because genomic data itself contains a large amount of information that can identify individuals, even if the identity information is deleted, genomic data still has the possibility of combining with other information to remove anonymization. For example, data users can infer phenotype information from genomic data to ultimately identify individuals. In one study, researchers used genomic data to predict eye and skin color traits and re-identified individuals in the dataset. Similarly, individual information can be re-identified through phenotypes such as diseases, facial structures, or pedigree relationships. By combining other easily accessible demographic information databases, individual information can also be inferred. Sweeney et al. linked names and contact information with publicly available information in the Personal Genome Project, and by linking demographic data with public records such as voter lists and digging up names hidden in additional files, they correctly identified 84% to 97% of the individual information. Gymrek et al. re-analyzed the identity information of multiple participants in the public genome project by analyzing short tandem repeats on the Y chromosome (Y-strs) and genetic pedigree databases, relying on completely free and publicly accessible Internet resources.

[0004] (2) Member inference: In genome-wide association studies (GWAS), usually only statistical data useful for meta-analyses are published by most researchers. However, Homer et al. proved that allele frequencies or genotype counts cannot hide the identity of individuals in genome-wide association studies. And using statistical information such as correlation and regression coefficients can further improve the accuracy of inference. Even the parameters of machine learning models can expose the genotypes and identities of participants. The identification of participants in GWAS data can lead to the leakage of various sensitive information such as diseases.

[0005] (3) Genotype inference and kinship search: Because the genomes of relatives have similarities, even if someone's genome has never been shared or sequenced, their genomic information and even certain phenotypes such as disease risk can be inferred from the genomic data of their relatives. Erlich et al. tested a database of 1.28 million personal genomes and found that about 60% of Europeans would match a third cousin or closer in the database. Although these databases are not currently open to individuals, they raise concerns about individual privacy.

[0006] (4) Commercial data leakage: In addition to concerns about individual privacy, genomic data or phenotypic data is often an important asset for businesses, especially breeding businesses, where a wealth of information can be mined, such as the performance of the company's pigs, population structure, introduction history, and other information related to business secrets. Cross-population genomic selection hopes to integrate data scattered among various pig breeding companies to improve selection accuracy, but for a company itself, its data is directly exposed to the platform, and there is a risk of leakage. The existence of this risk has to some extent affected the initiative and enthusiasm of companies to submit data, and some even deliberately submit tampered data, which seriously affects the overall data quality. SUMMARY

[0007] The purpose of the present application is to provide a gene selection system based on homomorphic encryption technology to solve the problems raised in the background art.

[0008] To achieve the above purpose, the present application provides the following technical solution: a gene selection system based on homomorphic encryption technology, comprising

[0009] An input module for inputting data;

[0010] A selection module for distributing the data of the input unit to the corresponding prediction model to obtain all information of the data and pedigree file;

[0011] An execution prediction module for reconstructing the operation array according to the configuration information provided by the selection module to implement the algorithms of each stage in homomorphic encryption;

[0012] And a display module for storing the initial plaintext of the homomorphic encryption algorithm of the execution prediction module, the intermediate result ciphertext, and the plaintext obtained by decrypting after homomorphic operation;

[0013] A cloud output module for outputting encryption to the cloud platform.

[0014] Preferably, the prediction model of the selection module includes BLUP, GBLUP, and ssGBLUP.

[0015] Preferably, the execution prediction module can perform at least single-population genomic prediction, multi-population genomic prediction, homomorphic encryption single-population genomic prediction, and homomorphic encryption multi-population genomic prediction.

[0016] Preferably, the system further comprises a phenotype processing module 600, which is used for conveniently and quickly filtering and screening a plurality of original phenotype data and performing visual processing.

[0017] Preferably, in the GBLUP model, y=Xb+Zg+e (1), wherein y is a phenotype vector, b is a fixed effect vector, X is a design matrix of the fixed effect, Z is a design matrix of an additive effect, i.e., EBV (g), e is a random error vector, g and e respectively obey and wherein is an additive genetic effect variance, is a residual variance, I is an identity matrix, and G is a kinship matrix calculated based on genomic information, and the formula is: wherein H is a normalized genotype matrix of an individual, and the calculation formula is:

[0018] wherein M is an original genotype matrix coded by 0, 1 and 2, is the jth SNP of the ith individual, and is the minor allele frequency of the site. At this time, the variance of the phenotype y in the model is:

[0019]

[0020] A random orthogonal matrix P is established as a secret key, and thus P -1 =P T The original phenotype y, the fixed effect b and the genotype data H are changed as follows to perform encryption:

[0021]

[0022] The relationship between the genomic kinship matrix calculated using the encrypted data and the original G matrix is:

[0023]

[0024] At this time, the variance of the encrypted phenotype z is:

[0025] A gene selection method based on homomorphic encryption technology, comprising the following steps:

[0026] S1, data uploading, the input module accepts configuration information during initialization;

[0027] S1.1, inputting the gene data collected by the terminal;

[0028] S1.2, obtain corresponding phenotypes by processing gene data;

[0029] S1.3, preset the phenotypes into a module, check the correctness of the data through the phenotypes, if the data is correct, proceed to the next step, if the data is incorrect, return to S1.1 after prompting;

[0030] S2, select a model, process and visualize the data through a phenotype processing module;

[0031] S3, execute prediction, the prediction module checks the data, generates a random orthogonal matrix as a key according to the selected data, and performs homomorphic encryption on the data through prediction;

[0032] S4, send a report, upload the encrypted data to the homomorphic encryption technology-based gene selection system through the processing module and the output module.

[0033] Compared with the prior art, the beneficial effects of the present application are: the present application develops a local client HEcryptor based on a homomorphic encryption algorithm to cooperate with the use of HEGS platform, the user can completely guarantee the safety of the data by encrypting the data locally before uploading, the encrypted data can be used for single-site genomic selection or cross-site genomic selection, and the user's various needs are met. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The system architecture of the present application is shown in the figure;

[0035] Figure 2 The method flowchart of embodiment 2 of the present application is shown in the figure. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0037] Embodiment 1

[0038] Please refer to Figure 1 The present application provides a technical solution: a homomorphic encryption technology-based gene selection system, comprising

[0039] The input module 100 is used for inputting data;

[0040] The selection module 200 distributes the data of the input unit to the corresponding prediction model, and obtains all the information of the data and the pedigree file through a one-step method.

[0041] An execution prediction module 300 is configured to reconstruct the operation array according to the configuration information provided by the selection module 200 to implement the algorithm of each stage in the homomorphic encryption;

[0042] A display module 400 is configured to store the initial plaintext of the homomorphic encryption algorithm of the execution prediction module 300, the intermediate result ciphertext, and the plaintext obtained by decryption after the homomorphic operation;

[0043] A cloud output module 500 is configured to output the encryption to the cloud platform.

[0044] Specifically, the prediction model of the selection module 200 includes BLUP, GBLUP, and ssGBLUP.

[0045] Specifically, the execution prediction module 300 can at least perform single-population genomic prediction, multi-population genomic prediction, homomorphic encryption single-population genomic prediction, and homomorphic encryption multi-population genomic prediction.

[0046] Specifically, the method further includes a phenotype processing module 600, which is configured to filter and screen a plurality of original phenotype data and perform visual processing for the user.

[0047] Specifically, in the GBLUP model, y=Xb+Zg+e (1), where y is a phenotype vector; b is a fixed effect vector; X is a design matrix of the fixed effect; Z is a design matrix of the additive effect, i.e., EBV(g); e is a random error vector; g and e respectively obey and wherein is the variance of the additive genetic effect, is the residual variance, I is an identity matrix, and G is a kinship matrix calculated based on genomic information, and the formula is: wherein H is a standardized genotype matrix of an individual, and the calculation formula is:

[0048] wherein M is an original genotype matrix coded with 0, 1, and 2, and is the corresponding jth SNP of the ith individual, and is the minor allele frequency of the site. At this time, the variance of the phenotype y in the model is:

[0049]

[0050] A random orthogonal matrix P is established as a secret key, and then P -1 =P T The original phenotype y, the fixed effect b, and the genotype data H are changed as follows to perform encryption:

[0051]

[0052] The relationship between the genomic relationship matrix calculated using the encrypted data and the original G matrix is:

[0053]

[0054] At this time, the variance of the encrypted phenotype z is:

[0055] Therefore, it can be found that after the original phenotype, fixed effect and genotype are encrypted by the orthogonal matrix P, but the estimated variance components remain consistent with the original data.

[0056] For multiple populations, we can construct a random orthogonal matrix P for each simulated population i , and let:

[0057]

[0058] where y i is the original phenotype of population i, z i is the encrypted phenotype vector, b i is the encoding matrix corresponding to the original environmental factor, W i is the encoding matrix corresponding to the encrypted environmental factor, H i is the standardized population genotype matrix, where the genotype value of the jth individual of the kth marker of population i is M ijk is the corresponding element of the original genotype matrix of the jth individual of the kth marker of population i coded by 0, 1 and 2, p ik is the minimum allele frequency of the kth locus. It is not difficult to see that although each population independently uses the secret key P i which is unknown to others, it will not change the homomorphic encryption feature of the data when joint genomic selection is performed, because:

[0059]

[0060] wherein is a diagonal block matrix composed of orthogonal matrices, that is, the matrix is a new orthogonal matrix, and the homomorphic encryption property is not broken. It can be seen that independent orthogonal transformation can still be used in the practice of joint genomic selection.

[0061] For ssGBLUP method and traditional BLUP method, because the encryption method cannot homomorphically encrypt pedigree information, joint breeding based on these two methods is not supported at present, but we have realized the genomic prediction of a single population based on homomorphic encryption in this chapter. Because for BLUP and ssGBLUP, the only difference from GBLUP is the difference in the use of the kinship matrix, that is, the A matrix based on pedigree construction and the H matrix based on both. When the user calculates the A matrix or the H matrix locally, we can obtain PB by performing Cholesky decomposition on it, such as A = BB T , and then encrypting the B matrix as described above, which can be equivalent to the homomorphic encryption method of GBLUP, that is:

[0062] PBB T P T = PAP T #(10)

[0063] Embodiment 2

[0064] A gene selection method based on homomorphic encryption technology, based on the above-mentioned gene selection system based on homomorphic encryption technology, comprising the following steps:

[0065] S1, data uploading, the input module 100 accepts configuration information at initialization;

[0066] S1.1, input the gene data collected by the terminal;

[0067] S1.2, obtain the corresponding phenotype by processing the gene data;

[0068] S1.3, preset the phenotype to the input module 100, and check the correctness of the input data through the phenotype. If the data is correct, go to the next step, and if the data is incorrect, return to S1.1 after prompting;

[0069] S2, selection model, process and visualize the data through the phenotype processing module 600;

[0070] S3, execute prediction, the prediction execution module 300 checks the data, generates a random orthogonal matrix as a key according to the selected data, and performs homomorphic encryption on the data through prediction;

[0071] S4, send report, upload the encrypted data to the gene selection system based on homomorphic encryption technology through the processing module 500 and the output module 500.

[0072] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A homomorphic encryption technology-based gene selection system, characterized by comprising: Comprising an input module (100) for inputting data; a selection module (200) for assigning data of the input unit to a corresponding prediction model to obtain all information of data and pedigree files by one-step method; an execution prediction module (300) for reconstructing an operation array according to configuration information provided by the selection module (200) to implement algorithms of each stage in homomorphic encryption; a processing module (400) for storing initial plaintext of the execution prediction module (300) homomorphic encryption algorithm, intermediate result ciphertext, and plaintext obtained by decryption after homomorphic operation; and a cloud output module (500) for outputting encryption to a cloud platform; the prediction model of the selection module (200) comprises GBLUP; In the GBLUP model, y = Xb + ZG + e (1), where y is the phenotypic vector; b is the fixed effect vector; X is the design matrix of the fixed effect; Z is the design matrix of the additive effect, i.e., EBV(g); e is the random error vector, and g and e are subject to and where is the variance of the additive genetic effect, is the residual variance, I is the unit matrix, and G is the kinship matrix calculated based on the genomic information, and the formula is: where H is the normalized genotype matrix of the individual, and the calculation formula is: where M is the original genotype matrix coded as 0, 1, 2, M ij for the jth SNP of the ith individual, respectively. p j is the frequency of the alternative allele at this locus. At this point, the variance of the phenotype y in the model is: A random orthogonal matrix P is established as a secret key, then we have: P -1 = P T , the original phenotype y, fixed effects b, genotype data H are encrypted by the following changes: z=Py W=Pb#(5) F=PH Then the relationship between the genomic relationship matrix calculated using encrypted data and the original G matrix is: At this time, the variance of the encrypted phenotype z is:

2. The homomorphic encryption technology based gene selection system according to claim 1, wherein: The execution prediction module (300) can at least perform single-population genomic prediction, multi-population genomic prediction, homomorphic encryption single-population genomic prediction, and homomorphic encryption multi-population genomic prediction.

3. The homomorphic encryption technology based gene selection system according to claim 2, wherein: It also comprises a phenotype processing module (600) for users to conveniently and quickly filter and screen multiple original phenotype data and perform visual processing.

4. A homomorphic encryption technology-based gene selection method, implemented by the homomorphic encryption technology-based gene selection system of any one of claims 1 to 3, characterized in that Comprising the following steps: S1, data uploading, the input module (100) accepts configuration information during initialization; S1.1, inputting gene data collected by a terminal; S1.2, obtaining corresponding phenotypes by processing gene data; S1.3, preloading the phenotypes into the input module (100), checking the correctness of the input data through phenotype inspection, if the data is correct, proceeding to the next step, if the data is incorrect, returning to S1.1 after prompting; S2, selecting a model, processing data and visualizing through the phenotype processing module (600); S3, executing prediction, the execution prediction module (300) checks data, generates a random orthogonal matrix as a key according to selected data, and performs homomorphic encryption on data through prediction; S4, sending a report, uploading encrypted data to the gene selection system based on homomorphic encryption technology through the processing module (500) and the output module (500).

Images