A method, device and storage medium for screening SNP sites in the whole genome of pigs

Abstract

The present invention discloses a method, device and storage medium for screening SNP sites in the whole genome of pigs. The method comprises: extracting several sample sets from gene data samples, performing feature selection on all sample sets to obtain SNP subsets of preset sites; screening out a preset number of feature selection methods with the highest stability ranking level as the optimal feature selection method, screening out several SNP subsets selected by the optimal feature selection method as several machine learning model features; constructing a corresponding breeding value prediction model according to a specific machine learning algorithm; inputting the machine learning model features into the breeding value prediction model, and outputting the predicted breeding value; determining the final SNP site of the prediction accuracy; and generating a pig whole genome SNP site chip according to the final SNP site. The whole genome SNP sites of pigs are screened by machine algorithms to produce a low-cost whole genome SNP site chip of pigs.

Description

Technical Field

[0001] The present invention relates to the field of pig genome selection, and in particular to a method, device and storage medium for screening SNP sites in the whole pig genome. Background Art

[0002] Whole genome selection is the use of a large number of genetic markers within the genome to estimate genomic breeding values. It has many advantages, such as shortening generation intervals and accelerating genetic progress. However, the high cost of genotype sequencing limits the application of whole genome selection.

[0003] The most common existing whole-genome selection method is genome-wide association study (GWAS) site screening, which uses a mixed linear model to perform GAWS analysis on the reference population to detect SNP sites that are significantly associated with the trait, and cannot identify the nonlinear relationship between SNP sites very well; for phenotypic traits that are not directly measured, GWAS may not be able to screen out significant SNPs, and need to be converted according to other phenotypic traits that can be directly measured; since quantitative traits are jointly acted upon by multiple minor-effect genes and major-effect genes, there are many controlled SNP sites. When the candidate phenotypic trait is a quantitative trait, GWAS cannot fully consider the genetic characteristics of the quantitative trait, and therefore can only screen out a few SNPs; GWAS cannot fully consider the interaction effect between genes and the environment, and the stability of SNPs screened out from different populations is extremely poor; the principle of GWAS is based on the genetic linkage disequilibrium (LD) effect, and cannot fully consider the population stratification existing in the sample population. Summary of the invention

[0004] The present invention provides a method for screening SNP sites in the whole genome of pigs to solve the technical problem of high cost of SNP site chips.

[0005] In order to solve the above technical problems, the embodiment of the present invention provides a method for screening SNP sites in the whole genome of pigs, comprising:

[0006] Extracting several sample sets from the gene data samples, selecting several feature selection methods to perform feature selection on all sample sets, and obtaining the first SNP subsets of several preset sites;

[0007] By calculating the Hamming distance of the first SNP subset of each of the preset sites, the stability level of each of the feature selection methods is determined; the top N feature selection methods ranked from high to low in stability level are selected as the optimal feature selection method, and according to the optimal feature selection method, several second SNP subsets are selected from several sample sets as several machine learning model features; wherein N is a positive integer;

[0008] According to several machine learning algorithms, several corresponding initial breeding value models are constructed respectively; according to the nested cross-validation method, the parameters of each initial breeding value model are adjusted to determine the optimal parameters of each initial breeding value model; according to the optimal parameters, each corresponding breeding value prediction model is constructed;

[0009] Inputting each of the machine learning model features into each of the breeding value prediction models respectively, and outputting a number of predicted breeding values ​​corresponding to each breeding value prediction model;

[0010] According to the number of machine learning model features, predicted breeding value and true breeding value of each breeding value prediction model, the average prediction accuracy value of each machine learning model feature in several breeding value prediction models is calculated, and the final SNP site with the highest average prediction accuracy value is determined.

[0011] Compared with the existing technology, this preferred solution uses feature selection in machine learning, that is, selecting SNP subsets from SNP sites. This SNP site selection method can prevent overfitting of high-dimensional data. When the features have high redundancy, different feature selection methods are used to select key SNP sites in genetic data, and SNP sites with higher stability are selected. Therefore, the stability is higher and the number of feature screening is free, thereby improving the stability and accuracy of SNP sites.

[0012] This application also uses machine learning algorithms to construct breeding value prediction models. Machine learning algorithms include a variety of technologies and non-parametric models for prediction and identification of large data sets. These algorithms do not require advance assumptions about data distribution and models, so they can well identify the linear and nonlinear relationships of SNP sites and improve the accuracy of SNP site selection.

[0013] As a preferred solution, it also includes: gradually performing quality control, variance filtering and high collinearity operations on the genetic data, deleting some SNP sites and some genetic data, and obtaining screened SNP sites and genetic data samples; wherein the screened SNP sites are SNP sites on the genetic sequence in the genetic data sample.

[0014] Compared with the existing technology, this preferred solution also performs quality control, variance filtering and high collinearity operations on the genetic data before screening SNP sites through feature selection in machine learning. According to the distribution of the genetic data, some SNP sites and some genetic data that cause large errors in the sample are deleted, thereby reducing the errors caused by the extreme distribution of genetic data, reducing the adverse effects on the subsequent feature selection screening of SNP sites, and improving the accuracy of SNP site screening.

[0015] As a preferred solution, several sample sets are extracted from the gene data samples, and several feature selection methods are selected to perform feature selection on all sample sets to obtain the first SNP subsets of several preset sites, specifically:

[0016] A number of samples are randomly selected from the genetic data samples as sample sets, and the random selection is repeated several times with replacement to obtain several sample sets; feature selection is performed on all sample sets using a first filtering feature selection method, a second filtering feature selection method, a first embedded feature selection method, a second embedded feature selection method and a random feature selection method, respectively, so that each feature selection method selects several first SNP subsets of preset sites in each sample set.

[0017] Compared with the existing technology, this preferred solution randomly extracts several sample sets from the genetic data, and uses different feature selection methods to perform feature selection on the sample sets, thereby screening out the key SNP sites from a huge number of SNP sites, greatly reducing the dimensionality of the SNP sites and improving the accuracy of screening SNP sites.

[0018] As a preferred solution, the parameters of each of the initial breeding value models are adjusted according to the nested cross-validation method to determine the optimal parameters of each of the initial breeding value models, specifically:

[0019] Dividing the data set of each of the initial breeding value models into a plurality of first training sets and a plurality of corresponding test sets, and dividing each of the first training sets into a plurality of second training sets and a plurality of corresponding validation sets;

[0020] The preset parameters of each of the breeding value initial models are set, and the accurate value of each corresponding breeding value initial model is calculated using the data set; and the optimal parameters of each of the breeding value initial models are determined according to the accurate value of each breeding value initial model with different preset parameters.

[0021] Compared with the prior art, in the process of constructing the breeding value prediction model, this preferred solution trains the model according to the model training set, tests the model on the test set, and then uses the error on the test set to approximate the generalization error of the model in the real scene. For each model, the model itself must be evaluated and the hyperparameters must be continuously adjusted until the optimal model parameters are selected and the most optimized model is trained. This reduces the error brought by the model itself to the data set, making the model applicable to the model data set and improving the model prediction accuracy.

[0022] As a preferred solution, each of the machine learning model features is input into each of the breeding value prediction models, and a number of predicted breeding values ​​corresponding to each breeding value prediction model are output, specifically:

[0023] Calculating the gene data, pedigree data and phenotypic value data to obtain the breeding value of the phenotypic trait, and using the breeding value of the phenotypic trait as a prediction target of a breeding value prediction model;

[0024] The machine learning model features are input into the breeding value prediction model so that the breeding value prediction model outputs the corresponding predicted breeding value according to the prediction target.

[0025] Compared with the existing technology, this preferred scheme combines genetic data, pedigree data, and phenotypic data to calculate the breeding value of phenotypic traits as the prediction target, and screens out key SNP sites that are strongly correlated with phenotypic shapes through the breeding value prediction model, eliminating the influence of environmental effects, population stratification, and whether direct measurement is used, thereby improving the accuracy and universality of SNP site screening.

[0026] Accordingly, the present application also proposes a method for preparing a pig whole genome SNP locus chip, comprising:

[0027] Select key SNP sites and design probes to make SNP site chips;

[0028] The method for screening the key SNP sites uses the pig whole genome SNP site screening method described in the present invention.

[0029] Compared with the existing technology, this preferred solution selects key SNP sites to make SNP chips, which reduces costs and can be applied to different pig groups with similar prediction effects and strong stability.

[0030] Accordingly, the present application also proposes a pig whole genome SNP site screening device, comprising: a feature selection module and a model construction module;

[0031] The feature selection module is used to extract several sample sets from the gene data sample, select several feature selection methods to perform feature selection on all sample sets, and obtain the first SNP subsets of several preset sites;

[0032] By calculating the Hamming distance of the first SNP subset of each of the preset sites, the stability level of each of the feature selection methods is determined; the top N feature selection methods ranked from high to low in stability level are selected as the optimal feature selection method, and according to the optimal feature selection method, several second SNP subsets are selected from several sample sets as several machine learning model features; wherein N is a positive integer;

[0033] The model building module is used to construct a number of corresponding initial breeding value models according to a number of machine learning algorithms; adjust the parameters of each initial breeding value model according to a nested cross-validation method to determine the optimal parameters of each initial breeding value model; and construct each corresponding breeding value prediction model according to the optimal parameters;

[0034] The site determination module is used to input each of the machine learning model features into each of the breeding value prediction models, and output a number of predicted breeding values ​​corresponding to each breeding value prediction model;

[0035] According to the number of machine learning model features, predicted breeding value and true breeding value of each breeding value prediction model, the average prediction accuracy value of each machine learning model feature in several breeding value prediction models is calculated, and the final SNP site with the highest average prediction accuracy value is determined.

[0036] Compared with the prior art, the feature selection module in the pig whole genome SNP site screening device of the preferred scheme uses feature selection in machine learning, that is, selecting SNP subsets from SNP sites. This SNP site selection method can prevent overfitting of high-dimensional data. When the feature has a high redundancy, different feature selection methods are used to select key SNP sites in the genetic data, and SNP sites with higher stability are selected. Therefore, the stability is higher and the number of feature screening is free, thereby improving the stability and accuracy of the SNP site.

[0037] The model building module and site determination module use machine learning algorithms to build breeding value prediction models and determine key SNP sites through model calculations. Machine learning algorithms include a variety of technologies and non-parametric models for prediction and identification of large data sets. These algorithms do not need to assume data distribution and models in advance, so they can well identify the linear and nonlinear relationships of SNP sites and improve the accuracy of SNP site selection.

[0038] As a preferred solution, the feature selection module includes: a site deletion unit and a site selection unit;

[0039] The site deletion unit is used to gradually perform quality control, variance filtering and high collinearity operations on the gene data, delete some SNP sites and some gene data, and obtain the screened SNP sites and gene data samples; the screened SNP sites are SNP sites on the gene sequence in the gene data sample;

[0040] Randomly extract several samples from the gene data sample as sample sets, repeat the random extraction several times with replacement, and obtain several sample sets; perform feature selection on all sample sets respectively using the first filtering feature selection method, the second filtering feature selection method, the first embedded feature selection method, the second embedded feature selection method, and the random feature selection method, so that each feature selection method selects several first SNP subsets of preset sites in each sample set;

[0041] The site selection unit is used to determine the stability level of each of the feature selection methods by calculating the Hamming distance of the first SNP subset of each of the preset sites; screen out the top N feature selection methods ranked from high to low in stability level as the optimal feature selection method, and according to the optimal feature selection method, select several second SNP subsets from several sample sets as several machine learning model features; wherein N is a positive integer.

[0042] Compared with the prior art, the site deletion unit of the feature selection module of the pig whole genome SNP site screening device of the preferred scheme also performs quality control, variance filtering and high collinearity operations on the genetic data before screening SNP sites through feature selection in machine learning. According to the distribution of the genetic data, some SNP sites and some genetic data that cause large errors in the sample are deleted, thereby reducing the errors caused by the extreme distribution of genetic data, reducing the adverse effects on the subsequent feature selection screening of SNP sites, and improving the accuracy of SNP site screening.

[0043] The site selection unit uses different feature selection methods to perform feature selection on several sample sets, and screens out key SNP sites from a huge number of SNP sites, which greatly reduces the dimension of SNP sites and improves the accuracy of SNP site screening.

[0044] As a preferred solution, each of the machine learning model features is input into each of the breeding value prediction models, and a number of predicted breeding values ​​corresponding to each breeding value prediction model are output, specifically:

[0045] Calculating the gene data, pedigree data and phenotypic value data to obtain the breeding value of the phenotypic trait, and using the breeding value of the phenotypic trait as a prediction target of a breeding value prediction model;

[0046] Inputting the machine learning model features into the breeding value prediction model so that the breeding value prediction model outputs a corresponding predicted breeding value according to the prediction target;

[0047] Compared with the prior art, the site determination module of the pig whole genome SNP site screening device of the preferred solution calculates the breeding value of the phenotypic trait as the prediction target by processing the gene data, pedigree data, and phenotypic data, and outputs the prediction value according to the prediction target through the machine learning algorithm. This data processing method can eliminate the environmental effect of the phenotypic trait, the influence of population stratification and whether it is directly measured, and is applicable to the breeding value prediction and site screening of any phenotypic trait, improving the universality and the accuracy of site screening.

[0048] Accordingly, the present application also proposes a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute a pig whole genome SNP site screening method as described in any of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a schematic diagram of an embodiment of the method for screening SNP sites in the whole genome of pigs provided by the present invention;

[0050] Figure 2 This is a schematic flow chart of an embodiment of a method for preparing a pig whole genome SNP site chip provided by the present invention;

[0051] Figure 3 It is a schematic structural diagram of an embodiment of a pig whole genome SNP site screening device provided by the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 creative work are within the scope of protection of the present invention.

[0053] Embodiment 1

[0054] Please refer to Figure 1 , a method for screening SNP sites in the whole genome of pigs provided in an embodiment of the present invention, comprising the following steps S101-S105:

[0055] Step S101: extracting a plurality of sample sets from the gene data sample, selecting a plurality of feature selection methods to perform feature selection on all the sample sets, and obtaining a plurality of first SNP subsets of preset sites;

[0056] In this embodiment, M samples are randomly selected from the gene data sample as a sample set, and the random selection is repeated X times with replacement to obtain several sample sets; the f_regression filtering feature selection method, the mutual_info_regression filtering feature selection method, the Lasso embedded feature selection method, the FR embedded feature selection method and the random feature selection method are used to perform feature selection on all sample sets, so that each feature selection method selects several first SNP subsets of 250, 500, 750, 1000, 1250, and 1500 sites in each sample set. Wherein, M and X are positive integers.

[0057] In this embodiment, it also includes: gradually performing quality control, variance filtering and high collinearity operations on the genetic data, deleting some SNP sites and some genetic data, and obtaining screened SNP sites and genetic data samples; wherein the screened SNP sites are SNP sites on the genetic sequence in the genetic data sample.

[0058] Step S102: determining the stability level of each of the feature selection methods by calculating the Hamming distance of the first SNP subset of each of the preset sites; selecting the top N feature selection methods ranked from high to low in stability level as the optimal feature selection method, and selecting a plurality of second SNP subsets from a plurality of the sample sets as a plurality of machine learning model features according to the optimal feature selection method; wherein N is a positive integer;

[0059] In this embodiment, according to the several first SNP subsets selected in each sample set by each feature selection method, the Hamming distances of the several first SNP subsets in each sample set are calculated;

[0060] The formula for Hamming distance is:

[0061]

[0062] Among them, h represents the Hamming distance, N represents the number of sample sets, S represents the first SNP subset, and k represents the number of sites in the SNP subset. The value range of h is 0 to 1. The closer it is to 0, the more stable the feature selection method is.

[0063] Step S103: constructing a number of corresponding initial breeding value models according to a number of machine learning algorithms; adjusting the parameters of each initial breeding value model according to a nested cross-validation method to determine the optimal parameters of each initial breeding value model; and constructing each corresponding breeding value prediction model according to the optimal parameters;

[0064] In this embodiment, quality control, variance filtering and high collinearity operations are performed step by step on the gene data, specifically:

[0065] The call rate threshold of quality control is set to 0.9, the MAF threshold is set to 0.05, and the HWEP threshold is set to 10-6, and some SNP sites and some gene data are deleted; the variance filtering parameter is set to 95 / 5, and variance filtering is performed to delete some SNP sites; the high collinearity parameter cut_off is set to 0.8 to delete high collinearity SNP sites. Quality control, variance filtering and high collinearity operations are existing technologies and will not be repeated here.

[0066] In this embodiment, several corresponding initial breeding value models are constructed according to several machine learning algorithms; wherein the machine learning algorithms include but are not limited to ridge regression algorithm, support vector machine algorithm, gradient boosting tree algorithm and extreme gradient boosting algorithm.

[0067] In this embodiment, the parameters of each of the initial breeding value models are adjusted according to the nested cross-validation method to determine the optimal parameters of each of the initial breeding value models, specifically:

[0068] The data set of each initial breeding value model is divided into several first training sets and corresponding several test sets, and each of the first training sets is divided into several second training sets and corresponding several verification sets; a double loop is implemented, the several first training sets are the outer loop to provide data for the inner loop, and the several second training sets are the inner loop, and the verification is performed by looping the several second training sets to determine the optimal parameters of each algorithm.

[0069] The preset parameters of each of the breeding value initial models are set, and the accurate value of each corresponding breeding value initial model is calculated using the data set; and the optimal parameters of each of the breeding value initial models are determined according to the accurate value of each breeding value initial model with different preset parameters.

[0070] Step S104: inputting each of the machine learning model features into each of the breeding value prediction models respectively, and outputting a number of predicted breeding values ​​corresponding to each breeding value prediction model;

[0071] Step S105: Calculate the average prediction accuracy of each machine learning model feature in several breeding value prediction models based on the number of machine learning model features, predicted breeding value and true breeding value of each breeding value prediction model, and determine the final SNP site with the highest average prediction accuracy.

[0072] In this embodiment, according to the number of machine learning model features, predicted breeding value and true breeding value of each breeding value prediction model, the average prediction accuracy value of each machine learning model feature in several breeding value prediction models is calculated, specifically:

[0073] Calculate the coefficient of determination R 2 and root mean square error RMSE;

[0074] The formulas for the coefficient of determination R2 and the root mean square error RMSE are:

[0075]

[0076] Where m represents the number of machine learning model features. represents the predicted breeding value, y represents the mean breeding value, and y represents the true breeding value.

[0077] R 2 The value range is 0 to 1. The closer it is to 1, the higher the prediction accuracy. The RMSE value range is 0 to infinity. The closer it is to 0, the higher the prediction accuracy.

[0078] Implementing the embodiments of the present invention has the following effects:

[0079] This application uses feature selection in machine learning, that is, selecting SNP subsets from SNP sites. This SNP site selection method can prevent overfitting of high-dimensional data. When the feature has a high degree of redundancy, different feature selection methods are used to select key SNP sites in genetic data, and SNP sites with higher stability are selected. Therefore, the stability is higher and the number of feature screening is free, which improves the stability and accuracy of SNP sites. Several sample sets are randomly selected from the genetic data, and different feature selection methods are used to select features for several sample sets, respectively. The huge number of SNP sites are screened out to select key SNP sites, which greatly reduces the dimension of the SNP sites and improves the accuracy of screening SNP sites.

[0080] This application also uses machine learning algorithms to construct breeding value prediction models. Machine learning algorithms include a variety of technologies and non-parametric models for the prediction and identification of large data sets. These algorithms do not need to assume data distribution and models in advance, so they can well identify the linear and nonlinear relationships of SNP sites and improve the accuracy of SNP site selection. This application combines genetic data, pedigree data, and phenotypic data to calculate the breeding value of phenotypic traits as the prediction target. The breeding value prediction model screens out key SNP sites that are strongly correlated with phenotypic shapes, eliminates the effects of environmental effects, population stratification, and whether direct measurement is used, and improves the accuracy and universality of SNP site screening.

[0081] Before screening SNP sites through feature selection in machine learning, this application also performs quality control, variance filtering and high collinearity operations on the genetic data, and deletes some SNP sites and some genetic data that cause large errors in the sample according to the distribution of the genetic data, thereby reducing the errors caused by the extreme distribution of genetic data, reducing the adverse effects on subsequent feature selection screening of SNP sites, and improving the accuracy of SNP site screening.

[0082] Embodiment 2

[0083] like Figure 2 As shown, the present invention also provides a method for preparing a pig whole genome SNP site chip, comprising the following steps:

[0084] Step S201: Select key SNP sites and design probes.

[0085] Step S202: Prepare a SNP site chip based on the probe.

[0086] The method for screening the key SNP sites uses the pig whole genome SNP site screening method as described in Example 1.

[0087] Implementing the embodiments of the present invention has the following effects:

[0088] This application selects key SNP sites to make SNP chips, which reduces costs and can be applied to different pig groups with similar prediction effects and strong stability.

[0089] Embodiment 3

[0090] like Figure 3 As shown, the present invention also provides a pig whole genome SNP site screening device, including: a feature selection module 301, a model construction module 302 and a site determination module 303.

[0091] The feature selection module 301 is used to extract several sample sets from the gene data sample, select several feature selection methods to perform feature selection on all sample sets, and obtain the first SNP subsets of several preset sites; determine the stability level of each feature selection method by calculating the Hamming distance of the first SNP subset of each preset site; select the top N feature selection methods ranked from high to low in stability level as the optimal feature selection method, and select several second SNP subsets from the several sample sets according to the optimal feature selection method as several machine learning model features; wherein N is a positive integer;

[0092] The model building module 302 is used to build several corresponding initial breeding value models according to several machine learning algorithms; adjust the parameters of each initial breeding value model according to the nested cross-validation method to determine the optimal parameters of each initial breeding value model; and build each corresponding breeding value prediction model according to the optimal parameters;

[0093] The site determination module 303 is used to input each of the machine learning model features into each of the breeding value prediction models respectively, and output several predicted breeding values ​​corresponding to each breeding value prediction model; according to the number of machine learning model features, predicted breeding values ​​and true breeding values ​​of each breeding value prediction model, the average prediction accuracy value of each machine learning model feature in several breeding value prediction models is calculated, and the final SNP site with the highest average prediction accuracy value is determined.

[0094] Among them, the feature selection module includes: a site deletion unit and a site selection unit;

[0095] The site deletion unit is used to gradually perform quality control, variance filtering and high collinearity operations on the gene data, delete some SNP sites and some gene data, and obtain the screened SNP sites and gene data samples; the screened SNP sites are SNP sites on the gene sequence in the gene data sample;

[0096] Randomly extract a number of samples from the gene data sample as sample sets, repeat the random extraction several times with replacement, and obtain a number of sample sets; perform feature selection on all sample sets respectively using a first filtering feature selection method, a second filtering feature selection method, a first embedded feature selection method, a second embedded feature selection method, and a random feature selection method, so as to obtain SNP subsets corresponding to five preset sites in each sample set;

[0097] The site selection unit is used to determine the stability level of each of the feature selection methods by calculating the Hamming distance of the first SNP subset of each of the preset sites; screen out the top N feature selection methods ranked from high to low in stability level as the optimal feature selection method, and according to the optimal feature selection method, select several second SNP subsets from several sample sets as several machine learning model features; wherein N is a positive integer.

[0098] Among them, the site determination module also calculates the gene data, pedigree data and phenotypic value data to obtain the breeding value of the phenotypic trait, and uses the breeding value of the phenotypic trait as the prediction target of the breeding value prediction model; the machine learning model features are input into the breeding value prediction model, so that the breeding value prediction model outputs the corresponding predicted breeding value according to the prediction target.

[0099] The above-mentioned pig whole genome SNP site screening device can implement the pig whole genome SNP site screening method of the above-mentioned method embodiment. The options in the above-mentioned method embodiment are also applicable to this embodiment and will not be described in detail here. The rest of the contents of the embodiment of this application can refer to the contents of the above-mentioned method embodiment, and will not be repeated in this embodiment.

[0100] Implementing the embodiments of the present invention has the following effects:

[0101] The feature selection module in the pig whole genome SNP site screening device of the present application uses feature selection in machine learning, that is, selecting SNP subsets from SNP sites. This SNP site selection method can prevent overfitting of high-dimensional data. When the features have high redundancy, different feature selection methods are used to select key SNP sites in the genetic data, and SNP sites with higher stability are selected. Therefore, the stability is higher and the number of feature screening is free, thereby improving the stability and accuracy of SNP sites.

[0102] The model building module and site determination module use machine learning algorithms to build breeding value prediction models and determine key SNP sites through model calculations. Machine learning algorithms include a variety of technologies and non-parametric models for prediction and identification of large data sets. These algorithms do not need to assume data distribution and models in advance, so they can well identify the linear and nonlinear relationships of SNP sites and improve the accuracy of SNP site selection.

[0103] The site deletion unit of the feature selection module of the pig whole genome SNP site screening device of the present application also performs quality control, variance filtering and high collinearity operations on the genetic data before screening SNP sites through feature selection in machine learning. According to the distribution of the genetic data, some SNP sites and some genetic data that cause large errors in the sample are deleted, thereby reducing the errors caused by the extreme distribution of genetic data, reducing the adverse effects on the subsequent feature selection screening of SNP sites, and improving the accuracy of SNP site screening.

[0104] The site selection unit uses different feature selection methods to perform feature selection on several sample sets, and screens out key SNP sites from a huge number of SNP sites, which greatly reduces the dimension of SNP sites and improves the accuracy of SNP site screening.

[0105] Embodiment 4

[0106] Accordingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the pig whole genome SNP site screening method as described in any one of the above embodiments.

[0107] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the terminal device.

[0108] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0109] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0110] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile terminal, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0111] Wherein, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0112] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for screening SNP sites in the whole genome of pigs, characterized in that: include: Extracting several sample sets from the gene data samples, selecting several feature selection methods to perform feature selection on all sample sets, and obtaining the first SNP subsets of several preset sites; By calculating the Hamming distance of the first SNP subset of each of the preset sites, the stability level of each of the feature selection methods is determined; the top N feature selection methods ranked from high to low in stability level are selected as the optimal feature selection method, and according to the optimal feature selection method, several second SNP subsets are selected from several sample sets as several machine learning model features; wherein N is a positive integer; According to several machine learning algorithms, several corresponding initial breeding value models are constructed respectively; according to the nested cross-validation method, the parameters of each initial breeding value model are adjusted to determine the optimal parameters of each initial breeding value model; according to the optimal parameters, each corresponding breeding value prediction model is constructed; Inputting each of the machine learning model features into each of the breeding value prediction models respectively, and outputting a number of predicted breeding values ​​corresponding to each breeding value prediction model; According to the number of machine learning model features, predicted breeding value and true breeding value of each breeding value prediction model, the average prediction accuracy value of each machine learning model feature in several breeding value prediction models is calculated, and the final SNP site with the highest average prediction accuracy value is determined.

2. A method for screening SNP sites in the whole pig genome as claimed in claim 1, characterized in that: Also includes: The genetic data is gradually subjected to quality control, variance filtering and high collinearity operations, some SNP sites and some genetic data are deleted, and screened SNP sites and genetic data samples are obtained; wherein the screened SNP sites are SNP sites on the genetic sequence in the genetic data sample.

3. A method for screening SNP sites in the whole pig genome as claimed in claim 1, characterized in that: The method extracts several sample sets from the gene data sample, selects several feature selection methods to perform feature selection on all sample sets, and obtains the first SNP subsets of several preset sites, specifically: A number of samples are randomly selected from the genetic data samples as sample sets, and the random selection is repeated several times with replacement to obtain several sample sets; feature selection is performed on all sample sets using a first filtering feature selection method, a second filtering feature selection method, a first embedded feature selection method, a second embedded feature selection method and a random feature selection method, respectively, so that each feature selection method selects several first SNP subsets of preset sites in each sample set.

4. A method for screening SNP sites in the whole genome of pigs as claimed in claim 1, characterized in that: The method of adjusting the parameters of each initial breeding value model according to the nested cross-validation method to determine the optimal parameters of each initial breeding value model is as follows: Dividing the data set of each of the initial breeding value models into a plurality of first training sets and a plurality of corresponding test sets, and dividing each of the first training sets into a plurality of second training sets and a plurality of corresponding validation sets; Setting preset parameters of each of the initial breeding value models, and calculating the accurate value of each corresponding initial breeding value model using the data set; According to the accurate value of each breeding value initial model with different preset parameters, the optimal parameters of each breeding value initial model are determined.

5. A method for screening SNP sites in the whole pig genome as claimed in claim 4, characterized in that: The steps of respectively inputting each of the machine learning model features into each of the breeding value prediction models, and outputting a number of predicted breeding values ​​corresponding to each breeding value prediction model are as follows: Calculating the gene data, pedigree data and phenotypic value data to obtain the breeding value of the phenotypic trait, and using the breeding value of the phenotypic trait as a prediction target of a breeding value prediction model; The machine learning model features are input into the breeding value prediction model so that the breeding value prediction model outputs the corresponding predicted breeding value according to the prediction target.

6. A pig whole genome SNP site screening device, characterized in that: include: Feature selection module, model building module and site determination module; The feature selection module is used to extract several sample sets from the gene data sample, select several feature selection methods to perform feature selection on all sample sets, and obtain the first SNP subsets of several preset sites; By calculating the Hamming distance of the first SNP subset of each of the preset sites, the stability level of each of the feature selection methods is determined; the top N feature selection methods ranked from high to low in stability level are selected as the optimal feature selection method, and according to the optimal feature selection method, several second SNP subsets are selected from several sample sets as several machine learning model features; wherein N is a positive integer; The model building module is used to construct a number of corresponding initial breeding value models according to a number of machine learning algorithms; adjust the parameters of each initial breeding value model according to a nested cross-validation method to determine the optimal parameters of each initial breeding value model; and construct each corresponding breeding value prediction model according to the optimal parameters; The site determination module is used to input each of the machine learning model features into each of the breeding value prediction models, and output a number of predicted breeding values ​​corresponding to each breeding value prediction model; According to the number of machine learning model features, predicted breeding value and true breeding value of each breeding value prediction model, the average prediction accuracy value of each machine learning model feature in several breeding value prediction models is calculated, and the final SNP site with the highest average prediction accuracy value is determined.

7. A pig whole genome SNP site screening device according to claim 6, characterized in that: The feature selection module includes: a site deletion unit and a site selection unit; The site deletion unit is used to gradually perform quality control, variance filtering and high collinearity operations on the gene data, delete some SNP sites and some gene data, and obtain the screened SNP sites and gene data samples; the screened SNP sites are SNP sites on the gene sequence in the gene data sample; Randomly extract several samples from the gene data sample as sample sets, repeat the random extraction several times with replacement, and obtain several sample sets; perform feature selection on all sample sets respectively using the first filtering feature selection method, the second filtering feature selection method, the first embedded feature selection method, the second embedded feature selection method, and the random feature selection method, so that each feature selection method selects several first SNP subsets of preset sites in each sample set; The site selection unit is used to determine the stability level of each of the feature selection methods by calculating the Hamming distance of the first SNP subset of each of the preset sites; screen out the top N feature selection methods ranked from high to low in stability level as the optimal feature selection method, and according to the optimal feature selection method, select several second SNP subsets from several sample sets as several machine learning model features; wherein N is a positive integer.

8. The pig whole genome SNP site screening device according to claim 6, characterized in that: The steps of respectively inputting each of the machine learning model features into each of the breeding value prediction models, and outputting a number of predicted breeding values ​​corresponding to each breeding value prediction model are as follows: Calculating the gene data, pedigree data and phenotypic value data to obtain the breeding value of the phenotypic trait, and using the breeding value of the phenotypic trait as a prediction target of a breeding value prediction model; The machine learning model features are input into the breeding value prediction model so that the breeding value prediction model outputs the corresponding predicted breeding value according to the prediction target.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program; wherein, when the computer program is run, it controls the device where the computer-readable storage medium is located to execute a method for screening SNP sites in the whole genome of pigs as described in any one of claims 1 to 5.

Images