High-order SNP (Single Nucleotide Polymorphism) interaction detection method based on hierarchical clustering and multi-task harmony search algorithm, and electronic equipment

Abstract

The invention relates to a high-order SNP (Single Nucleotide Polymorphism) interaction detection method based on hierarchical clustering and a multi-task harmony search algorithm, and electronic equipment, and belongs to the technical field of biogenetics. The method comprises the following steps: constructing a simulation data set of high-order SNP interaction; clustering the simulation data set by using hierarchical clustering; on the basis of clustering, inter-cluster and intra-cluster clustering initialization is carried out on the harmony library; iteration is started, and knowledge migration among different tasks is achieved through a multi-task framework; in the iteration process, the relation between the high-order SNP combination and the disease state is described from a plurality of aspects through a plurality of objective functions, and the optimal solution is liberated into a candidate solution set; screening the candidate solution set after iteration is finished through G-test to obtain a high-order SNP interaction set; according to the method, experiments are performed on 46 simulation data sets, a comparison algorithm is adopted for comparison with the method, and the optimal effect is achieved through the method.

Description

Technical Field

[0001] The invention relates to a high-order SNP interaction detection method and electronic equipment based on hierarchical clustering and multi-task harmonious search algorithm, and belongs to the technical field of biogenetics. Background Art

[0002] As the importance of genome-wide association studies (GWAS) continues to increase, the need to explore the genetic basis of complex diseases has become more urgent. However, there are still many problems in this field. Although single nucleotide polymorphisms (SNPs) are important markers for assessing disease susceptibility, traditional GWAS methods have made progress in identifying single risk SNPs, but they face the challenge of missing heritability because they fail to consider the interactions between multiple sites that cause many complex diseases. In terms of detecting high-order SNP interactions, although exhaustive search methods such as (MDR, Bi Force, GMDR) have been applied, the time consumed by this method increases exponentially when detecting high-order SNPs. Filter-based methods such as (BOOST, FastChi) can reduce the combinations to be searched, but may lose key information in the initial screening stage. At the same time, many machine learning-based methods, such as (epiForest, AdaBoost, ts-RF, support vector machines, Bayesian models), still face the difficulty of building accurate classifiers and perform poorly on data sets with no marginal effects. The above methods all face the problem of combinatorial explosion when detecting high-order SNP interactions, which consumes a lot of computing time. Although swarm intelligence algorithms such as (MACOED, FHSA-SED, IOBLPSO, DE, MTHSA-DHEI, LDHS) can search for a large number of SNPs, they often do not fully utilize the characteristics of genome-wide association analysis data and do not model the correlation of SNPs, resulting in insufficient efficiency and accuracy in discovering high-order SNP interactions. To solve the above problems, we proposed a high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm. Summary of the invention

[0003] The present invention provides a high-order SNP interaction detection method and electronic device based on hierarchical clustering and multi-task harmonious search algorithm, which are used to alleviate the problem that traditional methods are difficult to comprehensively and efficiently detect high-order SNP interactions. The present invention can be used to detect high-order SNP interactions, and the present invention has achieved good results on data sets with no marginal effects.

[0004] The technical solution of the present invention is: a high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm, the method comprising:

[0005] Step 1, construct a simulated data set of high-order SNP interactions;

[0006] Step 2, cluster the simulated data set using hierarchical clustering;

[0007] Step 3, based on clustering, initialize the inter-cluster and intra-cluster clustering of the harmony library;

[0008] Step 4: Start iteration and realize knowledge transfer between different tasks through a multi-task framework;

[0009] Step 5: In the iterative process, multiple objective functions are used to characterize the relationship between high-order SNP combinations and disease status from several aspects, and the better solutions are released into a set of candidate solutions;

[0010] Step 6: Use G-test to screen the candidate solution set after the iteration to obtain the high-order SNP interaction set.

[0011] Furthermore, the tool for constructing the simulation data set of high-order SNP interactions in Step 1 includes the simulation software GAMETES2.1.

[0012] Furthermore, the Step 2 includes:

[0013] Step 2.1, for the high-order SNP set S = {SNP 2 ,SNP 2 ,…,SNP n}, define a correlation measurement function r(SNP i ,SNP j ) to represent the correlation between two high-order SNPs; SNP i represents the i-th SNP in the SNP set;

[0014] Step 2.2, using the hierarchical clustering algorithm, calculate the correlation matrix R based on the correlation metric function, where the element R in the correlation matrix R ij , R ij = r(SNP i ,SNP j );

[0015] Step 2.3: Strengthen the correlation by square coefficient to get the strengthened correlation matrix R ′ , where the enhanced correlation matrix R ′ The element R i ′ j , R i ′ j =[r(SNP i ,SNPj )] 2 ;

[0016] Step 2.4, according to the enhanced correlation matrix R ′ , clustering high-order SNPs into different clusters C 1 ,C 2 ,…,C m , where m is the number of classes.

[0017] Furthermore, the Step 3 includes:

[0018] The initialization parameter IN is introduced to control the generation of candidate solutions. If IN = 1, candidate solutions are generated by randomization within the class. a , randomly select an element from Generate candidate solutions If IN = 0, generate candidate solutions through inter-class randomization, and randomly select different classes C p and class C q , from class C p and class C q Randomly select elements from Generate candidate solutions pq .

[0019] Furthermore, the step 4 includes:

[0020] (1) Set a certain number of iterations k and start iteration;

[0021] (2) The current task set is T = {T 1 ,T 2 ,…,T b}, each task T i1 Corresponding to different orders of high-order SNP interaction detection; let the current iteration number be k, for task T i1 , and its solution at the kth iteration is The objective function value is

[0022] (3) In each iteration, the solution is updated by transferring knowledge between multiple tasks: Where Update is a function that updates the solution according to the rules of the harmony search algorithm;

[0023] The knowledge transfer process between multiple tasks is expressed as: For the knowledge transfer matrix K, where K i1j1 Indicates that from task T j1 To Task T i1 The amount of knowledge transfer for task T i1, its new solution is affected by the transfer of knowledge from other tasks:

[0024]

[0025] in, represents the solution of the j1-th task after the k-th iteration.

[0026] Furthermore, the Step 5 includes:

[0027] (1) Use the K2 objective function to characterize the association between high-order SNP combinations and diseases. The K2 objective function is defined as:

[0028]

[0029] Where I represents the total number of gene association combinations. For example, for a k1-order high-order SNP combination, the total number of combinations is I=3. k1 , j2 represents the number of sample states, which is usually 2, namely normal and diseased. j2 represents the number of sample states, n i2 represents the number of samples with the i2th gene combination, n i2j2 Indicates the number of samples with the i2th gene combination in the j2th state; s is an index of the number of samples, with values ​​ranging from 1 to n i2j2 ;

[0030] (2) Use the LR objective function to model sensitivity and specificity, cover unobserved samples, and reflect the likelihood differences between high-order gene associations under different conditions; the LR objective function is defined as:

[0031]

[0032] Among them, e i2j2 represents the expected number of samples with the i2th gene combination in the j2th state, obtained by the Hardy-Weinberg law;

[0033] (3) The ME-score objective function is used to reflect the differences in the distribution of high-order gene associations under different disease states. The ME-score objective function is defined as:

[0034] ME-score=H(Y)+H(x 1 ,...,x c )-H(Y,x 1 ,...,x c )

[0035] Among them, H(Y) represents the Shannon entropy of Y, H(x 1 ,...x c ) represents x 1 ,...xc The joint entropy of

[0036]

[0037] p(y) represents the probability that the sample state is y, y is diseased or normal, p(x 1 ,...,x c ) means taking genotypes (x 1 ,...,x c ), x 1 ,...x c Indicates genotype, H(Y,x 1 ,...,x c ) represents Y and x 1 ,...,x c The joint entropy of

[0038] (4) Using the NJ objective function to reflect the diversity of high-order gene associations helps to more quickly mine pathogenic gene associations, accelerate the optimization process, and reduce the computational redundancy caused by combinatorial explosion. The NJ objective function is specifically defined as:

[0039] NJ-score=ND(X) / JE(X d )

[0040] Among them, ND(X) represents the distance between each gene combination site under different disease states, JE(X d ) represents the joint entropy of normal samples and is used to normalize the distance objective function.

[0041] Furthermore, the specific steps of Step 6 are:

[0042] Step 6.1. After the iteration is finished, the candidate high-order SNP interaction combinations are screened by G-test. For the candidate solution set D of the candidate high-order SNP interaction combinations, for each combination Combination in the candidate solution set D i3 Perform G-test statistics: G-test statistics are expressed as:

[0043]

[0044] Among them, O i3 is the observed frequency, E i3 is the expected frequency, and the candidate solution set D is expressed as: D = {Combination 1 ,Combination 2 ,…,Combination i3 ,…Combination m1};

[0045] Step 6.2. For each combination in the candidate solution set D of the candidate high-order SNP interaction combination, i3 Calculate the degrees of freedom df:

[0046] df = (r-1)(l-1);

[0047] Where r is the number of rows in the contingency table, l is the number of columns in the contingency table, the rows of the contingency table represent the number of elements in the candidate solution set D, and the columns correspond to the number of disease states 2, pathogenic or normal;

[0048] Step 6.3, according to G i3 Calculate the P value with degrees of freedom df:

[0049] P i3 =P(G i3 |df);

[0050] G i3 Represents Combination i3 G-test statistical scores;

[0051] Step 6.5, set the threshold to P threshold , select the solution set with a P value less than the threshold as the final solution:

[0052] SelectedSet = {Combination j3 ∈C|P j3 <P threshold};

[0053] SelectedSet represents the final candidate solution set, Combination j3 represents the j3rd element of the candidate solution set D, P j3 Represents the P value of the j3rd element of the candidate solution set D.

[0054] The present invention provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, a high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm is implemented.

[0055] The present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm is implemented.

[0056] The present invention provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm is implemented.

[0057] The beneficial effects of the present invention are:

[0058] 1. The present invention initializes the potential correlation between SNPs through hierarchical clustering to accurately capture the interaction characteristics between different SNPs, and uses a multi-task framework to achieve knowledge transfer between tasks. It can detect multiple high-order SNP interactions at the same time, alleviating the difficulty of traditional methods in comprehensively and efficiently detecting high-order SNP interactions. It can comprehensively and efficiently detect high-order SNP interactions. The present invention has achieved good results on data sets without marginal effects;

[0059] 2. This invention plays an important role in revealing the genetic mechanisms of complex diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is the overall flow chart of the present invention;

[0061] Figure 2 It is the multi-task knowledge transfer graph of the present invention. DETAILED DESCRIPTION

[0062] Example 1: Figure 1-Figure 2 As shown, a high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm, the method comprising:

[0063] Step 1, constructing a simulation data set of high-order SNP interactions; the tool for constructing the simulation data set of high-order SNP interactions in Step 1 includes simulation software GAMETES2.1, 46 generating simulation data sets with high-order SNP interactions, including 12 second-order 100 SNPs marginal effect data sets, 14 second-order 100 SNPs non-marginal effect data sets, 8 second-order 1000 SNPs marginal effect data sets and 12 second-order 1000 SNPs non-marginal effect data sets;

[0064] Step 2, cluster the simulated data set using hierarchical clustering;

[0065] Step 3, based on clustering, initialize the inter-cluster and intra-cluster clustering of the harmony library;

[0066] Step 4: Start iteration and realize knowledge transfer between different tasks through a multi-task framework;

[0067] Step 5: In the iterative process, multiple objective functions are used to characterize the relationship between high-order SNP combinations and disease status from several aspects, and the better solutions are released into a set of candidate solutions;

[0068] Step 6: Use G-test to screen the candidate solution set after the iteration to obtain the high-order SNP interaction set.

[0069] Furthermore, the Step 2 includes:

[0070] Step 2.1, for the high-order SNP set S = {SNP 1 ,SNP 2 ,…,SNP n}, define a correlation measurement function r(SNP i ,SNP j ) to represent the correlation between two high-order SNPs; SNP i represents the i-th SNP in the SNP set;

[0071] Step 2.2, using the hierarchical clustering algorithm, calculate the correlation matrix R based on the correlation metric function, where the element R in the correlation matrix R ij , R ij = r(SNP i ,SNP j );

[0072] Step 2.3: Strengthen the correlation by square coefficient to get the strengthened correlation matrix R ′ , where the enhanced correlation matrix R ′ The element R i ′ j , R i ′ j =[r(SNP i ,SNP j )] 2 ;

[0073] Step 2.4, according to the enhanced correlation matrix R ′ , clustering high-order SNPs into different clusters C 1 ,C 2 ,…,C m , where m is the number of classes.

[0074] Furthermore, the Step 3 includes:

[0075] The initialization parameter IN is introduced to control the generation of candidate solutions. If IN = 1, candidate solutions are generated by randomization within the class. a, randomly select an element from Generate candidate solutions If IN = 0, generate candidate solutions through inter-class randomization, and randomly select different classes C p and class C q , from class C p and class C q Randomly select elements from Generate candidate solutions pq .

[0076] Furthermore, the step 4 includes:

[0077] (1) Set a certain number of iterations k and start iteration;

[0078] (2) The current task set is T = {T 1 ,T 2 ,…,T b}, each task T i1 Corresponding to different orders of high-order SNP interaction detection; let the current iteration number be k, for task T i1 , and its solution at the kth iteration is The objective function value is

[0079] (3) In each iteration, the solution is updated by transferring knowledge between multiple tasks: Where Update is a function that updates the solution according to the rules of the harmony search algorithm;

[0080] The knowledge transfer process between multiple tasks is expressed as: For the knowledge transfer matrix K, where K i1j1 Indicates that from task T j1 To Task T i1 The amount of knowledge transfer for task T i1 , its new solution is affected by the transfer of knowledge from other tasks:

[0081]

[0082] in, represents the solution of the j1-th task after the k-th iteration.

[0083] Furthermore, the Step 5 includes:

[0084] (1) The K2 objective function is used to characterize the association between high-order SNP combinations and diseases, which has a higher discrimination accuracy. The K2 objective function is defined as:

[0085]

[0086] Where I represents the number of all combinations of gene associations, j2 represents the number of sample states, and n i2 represents the number of samples with the i2th gene combination, n i2j2 represents the number of samples with the i2th gene combination in the j2th state;

[0087] s is an index of the number of samples, ranging from 1 to n i2j2 ;

[0088] (2) Use the LR objective function to model sensitivity and specificity, covering unobserved samples; the LR objective function is defined as:

[0089]

[0090] Among them, e i2j2 represents the expected number of samples with the i2th gene combination in the j2th state, obtained by the Hardy-Weinberg law;

[0091] (3) The ME-score objective function is used to reflect the differences in the distribution of high-order gene associations under different disease states. The ME-score objective function is defined as:

[0092] ME-score=H(Y)+H(x 1 ,...,x c )-H(Y,x 1 ,...,x c )

[0093] Among them, H(Y) represents the Shannon entropy of Y, H(x 1 ,...x c ) represents x 1 ,...x c The joint entropy of

[0094]

[0095] p(y) represents the probability that the sample state is y, y is diseased or normal, p(x 1 ,...,x c ) means taking genotypes (x 1 ,...,x c ), x 1 ,...x c Indicates genotype, H(Y,x 1 ,...,x c ) represents Y and x 1 ,...,x c The joint entropy of

[0096] (4) The NJ objective function is used to reflect the diversity of high-order gene associations. The NJ objective function is specifically defined as:

[0097] NJ-score=ND(X) / JE(X d )

[0098] Among them, ND(X) represents the distance between each gene combination site under different disease states, JE(X d ) represents the joint entropy of normal samples.

[0099] Furthermore, the specific steps of Step 6 are:

[0100] Step 6.1. After the iteration, for the candidate solution set D of the candidate high-order SNP interaction combination, the candidate high-order SNP interaction combination is screened by G-test; for each combination in the candidate solution set D, i3 Perform G-test statistics: G-test statistics are expressed as:

[0101]

[0102] Among them, O i3 is the observed frequency, E i3 is the expected frequency, and the candidate solution set D is expressed as: D = {Combination 1 ,Combination 2 ,…,Combination i3 ,…Combination m1};

[0103] Step 6.2. For each combination in the candidate solution set D of the candidate high-order SNP interaction combination, i3 Calculate the degrees of freedom df:

[0104] df = (r-1)(l-1);

[0105] Where r is the number of rows in the contingency table, l is the number of columns in the contingency table, the rows of the contingency table represent the number of elements in the candidate solution set D, and the columns correspond to the number of disease states 2, pathogenic or normal;

[0106] Step 6.3, according to G i3 Calculate the P value with degrees of freedom df:

[0107] P i3 =P(G i3 |df);

[0108] G i3Represents Combination i3 G-test statistical scores;

[0109] Step 6.5, set the threshold to P threshold , select the solution set with a P value less than the threshold as the final solution:

[0110] SelectedSet = {Combination j3 ∈C|P j3 <P threshold};

[0111] SelectedSet represents the final candidate solution set, Combination j3 represents the j3rd element of the candidate solution set D, P j3 Represents the P value of the j3rd element of the candidate solution set D.

[0112] The present invention provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, a high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm is implemented.

[0113] The present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm is implemented.

[0114] The present invention provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm is implemented.

[0115] In order to illustrate the detection effect of the present invention, a comparison algorithm was used to compare with the present invention. Table 1 shows the comparative experimental results of the present method on a second-order 100SNPs data set with marginal effect after 3600 iterations, and Table 2 shows the comparative experimental results of the present method on a second-order 100SNPs data set without marginal effect after 3600 iterations.

[0116] Table 1 shows the comparative experimental results on the second-order 100 SNPs marginal effect data set under 3600 iterations.

[0117]

[0118] Table 2 shows the comparative experimental results on the second-order 100SNPs no marginal effect dataset under 3600 iterations.

[0119]

[0120] The values ​​in the table represent the detection results of the algorithm on a certain data set. For example, 23 in the second column of the third row means that the MACOED algorithm detected 23 disease models on the first data set with marginal effects (each data set without marginal effects has 100 pathogenic models), and the last column represents the average value of the disease models detected by the algorithm on each data set. As can be seen from Table 1, this method achieved relatively balanced results on 12 data sets, but the average performance was the best. The average value of this method is 70.6, which is 0.2 higher than the second-ranked epiACO. At the same time, Table 2 shows the experimental results of this method on 14 data sets without marginal effects. It can be observed that in most cases, the value of this method far exceeds that of other algorithms, and the average performance is the best, reaching 57, which is nearly 10 higher than the second-best algorithm MP-HS-DHSI, which shows the accuracy of this method.

[0121] In order to illustrate the detection effect of the present invention, a comparison algorithm was used for comparison with the present invention. Table 3 shows the comparative experimental results of the present method on a second-order 1000SNPs data set with marginal effect after 100,000 iterations. Table 4 shows the comparative experimental results of the present method on a second-order 1000SNPs data set without marginal effect after 100,000 iterations.

[0122] Table 3 shows the comparative experimental results on the second-order 1000SNPs marginal effect data set under 100,000 iterations.

[0123]

[0124] Table 4 shows the comparative experimental results on the second-order 1000SNPs no marginal effect dataset under 100,000 iterations

[0125]

[0126] Tables 3 and 4 show the experimental results of our method on 8 1000SNPs datasets with marginal effects and 12 1000SNPs datasets without marginal effects, respectively. The results show that the task of detecting second-order SNP interactions in 1000SNPs is significantly more challenging than that in 100SNPs, and the performance of all methods has dropped significantly. In addition, although MTHSA-DHEII does not perform well in detecting second-order SNP interactions in 100SNPs, it performs well in the 1000SNPs setting, achieving the second highest average performance on both datasets with and without marginal effects. This may be due to the fact that MTHSA-DHEII fully exploits the performance of its algorithm at a higher number of iterations. In the datasets with marginal effects, our method performs best in DME-1, DME-2, DME-6, and DME-7, and also achieves the best average performance. In most disease models without marginal effects, the values ​​of our method surpass those of other algorithms, achieving the best average performance of 24.7, 3.9 higher than the second-ranked MTHSA-DHEI.

[0127] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm, characterized in that: The method comprises: Step 1, construct a simulated data set of high-order SNP interactions; Step 2, cluster the simulated data set using hierarchical clustering; Step 3, based on clustering, initialize the inter-cluster and intra-cluster clustering of the harmony library; Step 4: Start iteration and realize knowledge transfer between different tasks through a multi-task framework; Step 5: In the iterative process, multiple objective functions are used to characterize the relationship between high-order SNP combinations and disease status from several aspects, and the better solutions are released into a set of candidate solutions; Step 6: Use G-test to screen the candidate solution set after the iteration to obtain the high-order SNP interaction set.

2. The method for detecting high-order SNP interactions based on hierarchical clustering and multi-task harmonious search algorithm according to claim 1, characterized in that: The tool for constructing the simulation data set for high-order SNP interactions in Step 1 includes the simulation software GAMETES2.

1.

3. The method for detecting high-order SNP interactions based on hierarchical clustering and multi-task harmonious search algorithm according to claim 1, characterized in that: The Step 2 includes: Step 2.1, for the high-order SNP set S = {SNP1, SNP2, ..., SNP n }, define a correlation measurement function r(SNP i , SNP j ) to represent the correlation between two high-order SNPs; SNP i represents the i-th SNP in the SNP set; Step 2.2, using the hierarchical clustering algorithm, calculate the correlation matrix R based on the correlation metric function, where the element R in the correlation matrix R ij , R ij = r(SNP i , SNP j ); Step 2.3, strengthen the correlation by square coefficient, and obtain the strengthened correlation matrix R', where the element R' in the strengthened correlation matrix R' ij , R′ ij =[r(SNP i , SNP j )] 2 ; Step 2.4: According to the enhanced correlation matrix R′, the high-order SNPs are clustered into different classes C1, C2, ..., C m , where m is the number of classes.

4. The method for detecting high-order SNP interactions based on hierarchical clustering and multi-task harmonious search algorithm according to claim 1, characterized in that: The Step 3 includes: The initialization parameter IN is introduced to control the generation of candidate solutions. If IN = 1, candidate solutions are generated by randomization within the class. a , randomly select an element from Generate candidate solutions If IN = 0, generate candidate solutions through inter-class randomization, and randomly select different classes C p and class C q , from class C p and class C q Randomly select elements from Generate candidate solution SolutiOn pq .

5. The method for detecting high-order SNP interactions based on hierarchical clustering and multi-task harmonious search algorithm according to claim 1, characterized in that: The step 4 includes: (1) Set a certain number of iterations k and start iteration; (2) The current task set is T = {T1, T2, ..., T b }, each task T i1 Corresponding to different orders of high-order SNP interaction detection; let the current iteration number be k, for task T i1 , and its solution at the kth iteration is The objective function value is (3) In each iteration, the solution is updated by transferring knowledge between multiple tasks: Where Update is a function that updates the solution according to the rules of the harmony search algorithm; The knowledge transfer process between multiple tasks is expressed as: For the knowledge transfer matrix K, where K i1j1 Indicates that from task T j1 To Task T i1 The amount of knowledge transfer for task T i1 , its new solution is affected by the transfer of knowledge from other tasks: in, represents the solution of the j1-th task after the k-th iteration.

6. The method for detecting high-order SNP interactions based on hierarchical clustering and multi-task harmonious search algorithm according to claim 1, characterized in that: The Step 5 includes: (1) Use the K2 objective function to characterize the association between high-order SNP combinations and diseases. The K2 objective function is defined as: Where I represents the number of all combinations of gene associations, j2 represents the number of sample states, and n i2 represents the number of samples with the i2th gene combination, n i2j2 Indicates the number of samples with the i2th gene combination in the j2th state; s is an index of the number of samples, with values ​​ranging from 1 to n i2j2 ; (2) Use the LR objective function to model sensitivity and specificity, covering unobserved samples; the LR objective function is defined as: Among them, e i2j2 represents the expected number of samples with the i2th gene combination in the j2th state, obtained by the Hardy-Weinberg law; (3) The ME-score objective function is used to reflect the differences in the distribution of high-order gene associations under different disease states. The ME-score objective function is defined as: ME-score=H(Y)+H(x1,...,x c )-H(Y,x1,...,x c ) Among them, H(Y) represents the Shannon entropy of Y, H(x1,...x c ) represents x1,...x c The joint entropy of p(y) represents the probability that the sample state is y, y is diseased or normal, p(x1, ..., x c ) means taking genotypes (x1, ..., x c ), x1, ...x c represents genotype, H(Y, x1, ..., x c ) represents Y and x1, ..., x c The joint entropy of (4) The NJ objective function is used to reflect the diversity of high-order gene associations. The NJ objective function is specifically defined as: NJ-score=ND(X) / JE(X d ) Among them, ND(X) represents the distance between each gene combination site under different disease states, JE(X d ) represents the joint entropy of normal samples.

7. The method for detecting high-order SNP interactions based on hierarchical clustering and multi-task harmonious search algorithm according to claim 1, characterized in that: The specific steps of Step 6 are: Step 6.

1. After the iteration is finished, for each combination in the candidate solution set D of the candidate high-order SNP interaction combination, i3 Perform G-test statistics: G-test statistics are expressed as: Among them, O i3 is the observed frequency, E i3 is the expected frequency, and the candidate solution set D is expressed as: D = {Combination1, Combination2, ..., Combination i3 ,…Combination m1 }; Step 6.

2. For the candidate solution set D of the candidate high-order SNP interaction combinations, for each combination in the candidate solution set D, i3 Calculate the degrees of freedom df: df = (r - 1)(l - 1); Where r is the number of rows in the contingency table, l is the number of columns in the contingency table, the rows of the contingency table represent the number of elements in the candidate solution set D, and the columns correspond to the number of disease states 2, pathogenic or normal; Step 6.3, according to G i3 Calculate the P value with degrees of freedom df: P i3 =P(G i3 |df); G i3 Represents Combination i3 G-test statistical scores; Step 6.5, set the threshold to P threshold , select the solution set with a P value less than the threshold as the final solution: SelectedSet={Combination j3 ∈C|P j3 <P threshold }; SelectedSet represents the final candidate solution set, Combination j3 represents the j3rd element of the candidate solution set D, P j3 Represents the P value of the j3rd element of the candidate solution set D.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, a high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm as described in any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements a high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements a high-order SNP interaction detection method based on hierarchical clustering and multi-task harmonious search algorithm as described in any one of claims 1 to 7.

Images