TWI786916B - Genomic estimated breeding value for predicting a trait and its application - Google Patents

Abstract

The present disclosure provides a method for establishing a formula of a genomic estimated breeding value for prediction a trait, which includes performing genome wide association analysis and machine learning on a database of single nucleotide polymorphism of the whole genome and a quantified database of a trait for establishing the formula.

Description

Breeding Value of Predicted Traits and Its Application

The disclosure relates to a method for predicting traits, specifically, a method for predicting traits using molecular biology techniques.

Economic animals are very important to the supply of human food. According to the statistics of FAO in 2018, the total number of pigs in the world is about 1.4 billion, surpassing beef, mutton and other poultry, ranking first in the livestock and poultry industry, and also the most important meat protein in the market Among the suppliers, the annual production and consumption in Asia are more than 50% of the global total, and they are the main production and consumption regions of edible pork in the world.

Traditional breeding farms most often use appearance traits to select pigs, but this method is slow in selection speed and low in accuracy, and usually requires many generations of screening to have a chance to select suitable breeds; with the development of molecular biotechnology, Single-gene DNA molecular marker-assisted breeding has become an important auxiliary breeding method. At present, the application of molecular markers to livestock breed improvement is mainly based on the genetic and molecular metabolic pathway information of the trait, and then find out the candidate gene that affects the trait, and then use the genetic Marking and locating the locus region that affects the important economic trait in the chromosome, or testing the influence of individual genes on the trait, this method must have a deep understanding of the genetic and metabolic pathways of the target trait in order to effectively improve the accuracy of breeding for the target trait Spend.

Growth, meat quality, and reproduction are the three important economic traits of pigs, and they are also the focus of breeding research. However, the main genetic mechanisms of these traits are mostly unclear, and most of them are complex genetic traits controlled by polygenes. Due to the technical limitations of traditional breeding methods and single-gene marker-assisted selection, it is extremely difficult to identify major genes and polymorphic loci, and it is difficult to select pigs with such excellent traits stably. Therefore, a complete genome-wide molecular marker is established It is very urgent and important to use the database as the basis for pig breeding.

In order to improve at least one or more of the aforementioned problems, this disclosure establishes a method for predicting traits as a basis for economic trait breeding.

The disclosure provides a method for establishing a calculation formula for predicting a trait breeding value (genomic estimated breeding value, GEBV) in a population, which includes: establishing genome-wide single nucleotide polymorphism (SNP) data of the population library, which includes the genome-wide SNP of each individual in the population; establishes a trait quantification database for the trait in the population, which includes the trait quantification value and the average trait quantification value of the trait in each individual in the population; Genome-wide association study (GWAS) was performed between the genome-wide single nucleotide polymorphism database and the trait quantification database, and each polymorphism of each SNP site was given a weight representing the degree of influence value (e value); and select n SNPs that affect the trait by machine learning methods, and establish the calculation formula of formula (1): GEBV _x = Mean _x + e ₁ ÍSNP ₁ + e ₂ ÍSNP ₂ +...+ e _n ÍSNP _n formula (1) In formula (1): GEBV _x is the breeding value of the trait; Mean _x is the quantified value of the average trait; e ₁ to e _n are the weight values of each SNP site; and SNP ₁ to SNP _n is a number, which is given according to the genotype of each SNP site. If the genotype is a heterozygote, it is given as 1; if the genotype is a homozygote that promotes the quantitative value of the trait, it is given is 2; if the genotype is an isozygote that reduces the quantified value of the trait, it is given a value of 0.

In an embodiment of the present disclosure, the genome-wide association analysis is performed using a fixed and random model circulating probability unification model.

In a specific embodiment of the present disclosure, the machine learning includes supervised learning, unsupervised learning and/or semi-supervised learning.

In an embodiment of the present disclosure, the numbers converted from SNP ₁ to SNP _n are 0, 1 or 2.

In specific embodiments of the present disclosure, the population is pigs. Examples of pigs include, but are not limited to, Duroc, Yorkshire, Lambrew, or hybrids thereof.

In particular embodiments of the present disclosure, examples of such traits include, but are not limited to, pig growth traits, meat quality traits, or reproductive traits. Examples of such pig growth traits include, but are not limited to, average daily gain, backfat thickness, age at 100 kg, or feed efficiency. Examples of such pork quality traits include, but are not limited to, meat color, marbling, loin area, loin depth, backfat depth, crude fat, live weight, carcass weight, or cooking loss. Examples of such swine reproductive traits include, but are not limited to, average total litter size, average live piglet size, average number of malformed/weak/stillborn births, average fetal distance, average days difference between due dates, average litter weight at birth, average three week old live piglets, or Average litter weight at three weeks.

This disclosure also provides a calculation formula for predicting the breeding value of a trait of an individual, which is represented by formula (1), and established by the method for establishing a predictive formula of a trait in a population as disclosed in this disclosure .

This disclosure also provides a method for predicting the breeding value of a trait of an individual to be tested using the aforementioned calculation formula, which includes detecting n SNP polymorphisms affecting the trait in the genome of the individual to be tested, and using the formula (1 ) to calculate the breeding value of the individual to be tested.

With the rapid development of big data, artificial intelligence learning and genome sequencing technology, this disclosure uses the concept of genomic selection (GS) to collect all the molecular markers scattered in the genome of the sample to be tested, and estimate their Breeding value (genomic estimated breeding value; GEBV), according to the level of its breeding value as the standard of breeding selection, the accuracy and efficiency of genomic selection are significantly higher than the traditional phenotype selection and single-gene molecular marker-assisted selection.

In order to improve the competitiveness of the industry and increase the overall income of the breeders, this disclosure provides a method for establishing a calculation formula for predicting the breeding price of a trait in a population, which includes: Establishing the whole genome SNP of the population A phenotypic (SNP) database, which includes the genome-wide SNP of each individual in the population; establishes a trait quantitative database for the trait in the population, which includes the trait quantification value and the average value of the trait for each individual in the population Quantitative value of traits; Genome-wide association analysis is performed between the genome-wide single nucleotide polymorphism database and the trait quantitative database, and each polymorphism of each SNP site is given a weight value representing the degree of influence ( e value); and select n SNPs affecting the character by machine learning method, and establish the calculation formula of formula (1): GEBV _x = Mean _x + e ₁ ÍSNP ₁ + e ₂ ÍSNP ₂ +...+ e _n ÍSNP _n formula (1) In formula (1): GEBV _x is the breeding value of the trait; Mean _x is the quantified value of the average trait; e ₁ to e _n are the weight values of each SNP locus; and SNP ₁ to SNP _n They are numbers respectively, given according to the genotype of each SNP locus, where if the genotype is a heterozygote, a reference value is given; if the genotype is a homozygote that promotes the quantitative value of the trait, it is given as A number greater than the reference value; if the genotype is an isozygote that reduces the quantitative value of the trait, a number smaller than the reference value is given.

This disclosure also provides a calculation formula for predicting the breeding value of a trait of an individual, which is represented by formula (1), and established by the method for establishing a predictive formula of a trait in a population as disclosed in this disclosure .

This disclosure also provides a method for predicting the breeding value of a trait of an individual to be tested using the aforementioned calculation formula, which includes detecting n SNP polymorphisms affecting the trait in the genome of the individual to be tested, and using the formula (1 ) to calculate the breeding value of the individual to be tested.

In one embodiment of the present disclosure, the concept of genome selection is used, combined with genome-wide correlation analysis and machine learning methods, to analyze the genome-wide single nucleotide polymorphic loci, and find out the corresponding genes according to the traits polymorphic sites. Finally, different weights are given according to the degree of influence of each marker on the aforementioned traits, and then an evaluation system for various important economic traits is established, and the test results are quantified into comparable values, which are used as a comparison index and provided to users as a reference index for breeding screening.

The "group" mentioned in this disclosure can be a collection formed by multiple biological individuals, including animal groups or plant groups. In a specific embodiment of the present disclosure, each individual in the group belongs to the same species. In another specific embodiment of the present disclosure, each individual in the group belongs to the same species. In a specific embodiment of the present disclosure, each individual in the group belongs to the same species. , each individual in this group belongs to the species that grow in the same area. In a specific embodiment of the present disclosure, the organism is an economic animal, including but not limited to pigs, cows, chickens, ducks, fish, and shrimps.

The pigs described in this disclosure are not limited to their breed or sex. For example, the pig can be a boar or a sow. In addition, examples of such pigs are Duroc, Yorkshire, Lambrace and crossbred lines thereof.

The traits disclosed herein include but are not limited to growth traits, meat quality traits or reproductive traits. In order to enable analysis, the traits according to the disclosure are quantified, including but not limited to giving a score to quantify a trait.

The formula (1) disclosed in this disclosure can be used to estimate the breeding value of a trait, so as to evaluate whether a test individual can be used as a breeding parent.

In formula (1), e ₁ to e _n are the weight values of each site, and SNP ₁ to SNP _n are numbers respectively, which are given according to the genotype of each SNP site, wherein if the genotype is heterozygous, then A base value is given; if the genotype is an isozygote that promotes the quantitative value of the trait, it is given as a number greater than the base value; if the genotype is an isozygous that reduces the quantitative value of the trait, it is given as A number that is less than the benchmark value.

In a specific embodiment of the present disclosure, if the genotype of a SNP site is Aa, the given reference value is 1, and if the genotype AA of the SNP site is a quantitative value for promoting a trait, then it is given as 2, If the genotype aa of this SNP locus reduces the quantitative value of a trait, it is set as 0. In another specific embodiment of the present disclosure, if the genotype of a SNP site is Bb, the given reference value is 1, and if the genotype bb of the SNP site is a quantitative value for promoting a trait, then it is given as 2 , if the genotype BB of this SNP locus is to reduce the quantitative value of a trait, it is given as 0.

In a specific embodiment of the present disclosure, the genome-wide association analysis is performed using a fixed and random model circulating probability unification model, which can effectively reduce false positives compared with other tools. The problem of positives and false negatives (false positives and false negatives) is divided into a fixed effect model (fixed effect model, FEM) and a random effect model (random effect model, REM). The two models are used to repeatedly detect alternately to improve statistics Accuracy, but also can effectively increase the computing speed and use lower computer resource computing.

In specific embodiments of the present disclosure, growth traits may include average daily gain, backfat thickness, age at 100 kg or feed efficiency. Pork quality traits may include flesh color, marbling, loin area, loin depth, fat depth, crude fat, live weight, carcass weight, or cooking loss. Pig reproductive traits can include average total litter size, average number of live piglets, average number of abnormal/weak/dead fetuses, average fetal distance, average days difference between expected farrowing dates, average birth litter weight, average three-week live piglets or average three-week litter weight .

In a specific embodiment of the present disclosure, it combines machine learning and whole genome sequencing technology. Compared with phenotype breeding method and single gene marker method, this technology can screen individuals after birth without waiting After the individual grows up and matures, it is observed, and through the analysis of the whole genome data, it is possible to comprehensively screen and analyze the SNP sites of the whole genome of the individual to be tested. Compared with the previous technology, it is possible to select individuals with excellent traits at an early stage. Reduce breeding costs and shorten breeding time. In addition, through the selection strategy of genome-wide SNP screening, all genes that affect the trait can be found at the same time, so as to reduce breeding variation and improve the accuracy and stability of breeding. That is to say, the methods and markers disclosed in this disclosure can be used for early detection of important economic traits such as growth, meat quality or reproduction of the individual to be tested, early detection and elimination of individuals with poor traits, while retaining individuals with excellent traits to reduce selection and Expenditure of breeding costs, improve the overall economic benefits.

The following non-limiting examples assist those skilled in the art to practice the invention. These examples should not be considered to unduly limit the invention. Those skilled in the art to which the present invention pertains can make modifications and changes to the embodiments discussed herein without departing from the spirit or scope of the present invention, and still belong to the scope of the present invention. example

Equation (1) of the present disclosure can be established through the process shown in FIG. 1 . A total of 2,531 pig samples were sampled for 21 traits in growth, meat quality and reproduction, of which 251 were whole-genome sequencing, 1,239 were Illumina Porcine SNP60 v2 BeadChip genome-wide SNP arrays, and 1,041 were Affymetrix Axiom Porcine Breeders Array genome-wide SNP chips, after detection and analysis by these three platforms, after excluding SNPs with poor quality, a total of 47,672 SNP markers can be used for subsequent genome-wide association analysis (GWAS), including 4 kinds of growth Traits, 9 kinds of meat quality traits and 8 kinds of reproductive traits, the analysis steps and methods are described in detail below. Analysis steps and methods (1) Experimental sample collection

The present invention collects 21 other phenotypic data of important economic traits such as pig growth, meat quality and reproduction, and samples 2531 pieces of whole-genome genotype identification data from the above-mentioned recorded pig samples, wherein the samples include Duroc, blue Reese and Yorkshire. (2) Pig genomic DNA extraction a. Tissue samples

Place the collected fresh tissue in a microcentrifuge tube containing 250 μl of lysis buffer, cut the tissue into pieces with scissors, and then add 400 μl of lysis buffer and 35 μl of proteinase K (10 mg/ml ), mix the sample with lysis buffer evenly, place in a 55°C water bath overnight, and then centrifuge at 320 xg for 5 minutes at room temperature. Take the supernatant to a new centrifuge tube and add an equal volume of phenol, mix thoroughly for 10 minutes and then centrifuge at 10,000 xg for 10 minutes. Collect the supernatant, add an equal volume of phenol/chloroform-isoamyl alcohol (Phenol/chloroform-isoamyl alcohol), mix well and centrifuge at 10,000 xg for 10 minutes. Collect the supernatant, add an equal volume of chloroform-isoamyl alcohol (Chloroform-isoamyl alcohol, 24:1), mix well and centrifuge at 10,000 xg for 10 minutes. Take the supernatant to a new centrifuge tube, add 1/10 of the total volume of 3 M sodium acetate (Sodium acetate, NaOAc), then add an equal volume of isopropanol (Isopropanol), mix well and place in a -20°C refrigerator to accelerate precipitation 30 minutes, followed by centrifugation at 12,000 xg for 15 minutes. After removing the supernatant, centrifuge with 1 ml 70% ethanol and 1 ml 100% ethanol at 12,000 xg for 5 minutes at 4°C to remove salt and water, and finally add 50 μl TE buffer to dissolve the DNA and store Store at -80°C for later use. b. Blood samples

Blood DNA was extracted using the Geneaid DNA extraction kit. Take 250 μl of whole blood, add 50 μl of proteinase K (10 mg/ml), mix well and then act at 60 °C for 30 minutes; add 300 μl GSB buffer, mix well and then in Incubate at 60°C for 30 minutes; then add 300 μl of 100% ethanol and mix well; transfer the above mixture to a filter column, centrifuge at room temperature at 16,000 xg for 1 minute, discard the filtered waste solution; then repeat this step twice , add 300 μl W1 buffer, centrifuge at 16,000 xg for 1 minute at room temperature, discard the filtered waste liquid; then repeat this step twice, add 300 μl washing buffer, and centrifuge at 16,000 xg for 1 minute at room temperature, pour Discard the filtered waste liquid; then centrifuge at 16,000 xg at room temperature for 3 minutes to remove the residual liquid; finally add 100 μl of elution buffer, and centrifuge at 16,000 xg at room temperature for 3 minutes. The sample DNA is in this solution and stored Store at -80°C for later use. c. Semen sample

Semen DNA was also extracted using the Geneaid DNA extraction kit. Take 100 μl of the concentrated semen, mix it with 1xPBS at a ratio of 1:9, centrifuge at 16,000 xg for 1 minute at room temperature, remove the supernatant; add 100 μl of lysis buffer ( 850 μl Geneaid sperm lysis buffer + 160 μl 100 mM DTT + 20 μl 10 mg/ml proteinase K), mix evenly and place at 60°C for 16 hours; add 100 μl GSB buffer, mix well; then add 100 μl 100 % alcohol, mix well; transfer the above mixture to a filter column, centrifuge at 16,000 xg at room temperature for 1 minute, pour off the filtered waste liquid; then repeat this step twice, add 300 μl W1 buffer, and store in the chamber Centrifuge at 16,000 xg for 1 minute, discard the filtered waste; then repeat this step twice, add 300 μl of washing buffer, and centrifuge at 16,000 xg for 1 minute at room temperature, discard the filtered waste; then wash at room temperature 16,000 Centrifuge at xg for 3 minutes to remove the remaining liquid; finally add 100 μl of elution buffer and centrifuge at room temperature for 3 minutes at 16,000 xg. The solution is the sample DNA, which is stored at -80°C for future use. (3) Sample DNA quality analysis

The pig DNA used in this disclosure must pass the following two quality inspection methods at the same time after the extraction is completed. Only when they meet the standards can they be used for subsequent sequencing and analysis. If the quality inspection fails, the sample must return to step (2) Re-extract. a. Agarose gel electrophoresis

Take 0.8 g of agarose (Agarose I, purchased from Amersco) and add 100 ml of 0.5 X TBE buffer solution (purchased from Bioman scientific), mix well and heat in a microwave oven until the agarose is completely dissolved and appears as a transparent solution. Cool down to about 50°C, add 5 μl of ethidium bromide (EtBr, 10 mg/ml, purchased from Sigma), mix well, pour into the gel-making model, and remove the tooth comb after the gel is solidified. Take 2 μl of the genomic DNA extracted in step (2) and mix it with an appropriate amount of indicator for 0.8% agarose gel electrophoresis. The electrophoresis condition is 100 volts for about 30 minutes. Confirm the quality and integrity of genomic DNA. Observed under ultraviolet light, if the DNA has no small stain bands (Smear) and the main bright band is greater than 20 kb, it means that the extracted genomic DNA is of high integrity and can be used for subsequent sequencing. b. Micro spectrophotometer analysis

Take 1 μl of genomic DNA sample, drop it on a NanoDrop 2000 ultra-micro spectrophotometer (Thermo Fisher Scientific), measure the OD230, OD260, and OD280 reading values of the DNA sample, and calculate the concentration of the sample and the ratio of OD260/OD230, OD260/OD280, OD260 The /OD230 value should be greater than 2.0, and the OD260/OD280 value should be between 1.8 and 2.0, indicating that the extracted DNA has high purity and less protein and polysaccharide residues, and can be used for subsequent sequencing. (4) Analysis of SNP loci in the whole pig gene

Using the Porcine SNP60 v2 BeadChip whole-genome chip developed by Illumina and the Axiom Porcine Breeders Array chip developed by Thermo Fisher, which contain 64,232 and 55,150 porcine SNP sites respectively, which can provide a whole-genome with an average of one detection point per 40kb Coverage, providing detailed information on the genetic variation of pigs with dense SNP loci. After image analysis, the obtained SNP information was compared with the Sus Scrofa 11.1 database. (5) Quality control of SNP genotype data

GAPIT 3 (Genome Association and Prediction Integrated Tool 3) software was used for quality control of SNP genotypes. The screening criteria were: removal of SNP call rate (SNP call rate) ≦99%, minimum allele frequency (Minor Allele Frequency) , MAF) ≦0.01 SNP sites, after the above screening process, the SNP sites that finally pass the screening will be used for subsequent GWAS analysis. (6) Genome-wide association analysis

GWAS analysis was performed using the FarmCPU (Fixed and random model circulating probability unification) model in the GAPIT 3 software package.

After the collected sample genome-wide SNP data and external phenotype trait data are calculated through the FarmCPU model, the model will give each SNP site a weight value (e value) of influence degree for each trait, and establish its genome-wide association analysis The Manhattan map, as shown in Figure 2, takes the average tire distance as an example; then use the quantile map (QQ plot) to view these data, as shown in Figure 3, the average tire distance. From the results in Figure 3, it can be found that the SNP sites after the X-axis-log ₁₀ 3 of the quantile map deviate significantly from the uniform distribution line of the null hypothesis, thus confirming that these sites do have a significant correlation with the target trait. (7) Machine learning

This training data is characterized by the whole genome SNP data of pigs, and the phenotype data of subsequent growth conditions are used as markers. By properly selecting key gene combinations that affect pig growth, machine learning methods are used to establish a predictive classification model. The process of building the model is shown in Figure 4. This disclosure uses the following three types of machine learning techniques respectively: a. Supervised Learning

Supervised learning means that during the training phase, every piece of data used to train the model contains the correct answer. Taking the evaluation of pig traits as an example, the database collects the genotype traits of all pigs to be analyzed and the data of various external traits of pigs after they grow up, and then the inference model can be established according to the machine learning method of supervised learning , to find out the correlation mode between genotype characteristic data and phenotype traits. Afterwards, as long as the genotype characteristic data is input into the inference model, the external phenotype traits of pigs can be predicted after they grow up. That is to say, the supervised learning method will provide feature data and corresponding answers during the training phase, and find out the correlation mode between the two. b. Unsupervised Learning

Compared with providing data and corresponding answers in the training phase to find relevant patterns, unsupervised learning is only given data and features, so that machine learning methods can automatically find possible answers. Taking the evaluation of pig traits as an example, the database only has genotype data or only phenotype data, but there is no corresponding information about the complete genotype and phenotype traits. This type of problem is very common in reality. Due to the lack of a correlation model between genotype data and phenotype traits, supervised learning cannot be used, and only unsupervised learning methods can be used. Only the trend of genotype data or phenotype traits can be observed, and the similarity between data and phenotype traits can only be observed. Change trends to find possible association patterns. c. Semi-supervised Learning

Since the supervised method requires historical data that has already marked the answer, the burden of manual labeling is heavy. Unsupervised learning does not have labeled answers and can only infer based on the similarity of the data. The semi-supervised learning method extracts the characteristics of supervised and unsupervised. On the one hand, it finds a model from the information that has already marked the answer, and then uses the model that has been found to infer the data that has not marked the answer. Then find out more reliable data from these inference results, and incorporate it into the training data set that has marked the answer. As a result of this iterative operation, an inference model for the entire data set can be found. Since it is not easy to find complete and marked answers in real data, semi-supervised learning methods are of considerable importance in practice.

The SNP sites found in the GWAS analysis that have a significant correlation with the average live litter size of the 3rd to 5th litters were compared with the Sus scrofa 11.1 database on the Ensemble website to find out the gene sequence position of the SNP.

After repeated calculations of the FarmCPU model, the weights of these 47,672 SNP markers affecting the average number of live piglets in the third to fifth litters were obtained. By counting the weights obtained by the genotype of each SNP site, you can get the The estimated breeding value of the genome. In addition, the results from the Manhattan plot show that there is a SNP site that has a significant correlation with the average number of living pigs in the 3rd to 5th litters, as shown in Figure 5 and Table 1. Use the quantile map to test these data. From the results in Figure 6, it can be found that the SNP site of the quantile map after about -log ₁₀ 3.5 on the X axis obviously deviates from the uniform distribution line of the null hypothesis, thus confirming that this site is indeed related to the first There is a correlation between the average number of piglets alive from 3 to 5 litters. Table 1: SNP loci highly correlated with the average live litter size from the 3rd to the 5th parity chromosome SNP markers Location P value Weights SNPs nearest gene 15 ALGA0085579 56317018 2.97E-11 -0.800 T/C mfhas1

The above-mentioned embodiments are only to illustrate the principles and effects of the present invention, but not to limit the present invention. Modifications and changes made to the above-mentioned embodiments by those skilled in the technical field of the present invention still do not violate the spirit of the present invention. The scope of rights of the present invention should be listed in the scope of patent application described later.

FIG. 1 shows a flowchart of an embodiment of the present disclosure establishing equation (1).

Figure 2 shows the Manhattan plot of the genome-wide association analysis of the average fetal distance.

Figure 3 shows the quantile map of the genome-wide association analysis of the mean fetal distance.

Figure 4 uses machine learning to establish a classification prediction model and the actual operation process.

Figure 5 shows the Manhattan plot of the genome-wide association analysis for the average number of litters alive from the 3rd to the 5th litter.

Figure 6 shows the quantile map of the genome-wide association analysis of the average live litter size from the 3rd to the 5th parity.

Claims (11)

A method for establishing a calculation formula for predicting a trait breeding value (genomic estimated breeding value, GEBV) in a population, which includes: establishing a genome-wide single nucleotide polymorphism (SNP) database of the population, which Including the whole genome SNP of each individual in the group; establishing a trait quantification database of the trait in the population, which includes the trait quantification value and the average trait quantification value of the trait in each individual in the population; the whole genome single The nucleotide polymorphism database and the trait quantification database were used for genome wide association study (GWAS), and each polymorphism of each SNP site was given a weight value representing the degree of influence (e value); and select n SNPs that affect the trait by machine learning methods, and establish the calculation formula of formula (1): GEBV _x =Mean _x +e ₁ ×SNP ₁ +e ₂ ×SNP ₂ +.... ..+e _n ×SNP _n Formula (1) In Formula (1): GEBV _x is the breeding value of the trait; Mean _x is the quantified value of the average trait; e ₁ to e _n are the weight values of each SNP locus; and SNP ₁ to SNP _n are numbers respectively, which are given according to the genotype of each SNP locus, where if the genotype is heterozygous, a reference value is given; if the genotype is the same type that promotes the quantitative value of the trait If the genotype is a zygote, a number greater than the reference value is given; if the genotype is an isozygote that reduces the quantitative value of the trait, a number smaller than the reference value is given. The method of claim 1, wherein the genome-wide association analysis is performed using a fixed and random model circulating probability unification model. The method of claim 1, wherein the machine learning includes supervised learning, unsupervised learning and/or semi-supervised learning. The method according to claim 1, wherein the numbers converted from SNP ₁ to SNP _n are 0, 1 or 2. The method according to claim 1, wherein the herd is pigs. The method as claimed in item 5, wherein the pig is Duroc, Yorkshire, Lambrus or a hybrid thereof. The method according to any one of claims 1 to 6, wherein the traits are growth traits, meat quality traits or reproductive traits of pigs. The method according to claim 7, wherein the pig growth traits are average daily gain, backfat thickness, age at 100 kg or feed efficiency. The method according to claim 7, wherein the meat quality traits are meat color, marbling, loin eye area, loin eye depth, backfat depth, crude fat, live weight, carcass weight or cooking weight loss rate. According to the method of claim 7, wherein the reproductive traits of pigs are average total litter size, average live piglet size, average number of deformed/weak/stillborn fetuses, average fetal distance, average number of days difference between expected delivery dates, average birth litter weight, and average three-week Litter size alive or average three-week litter weight. A method for predicting the breeding value of a trait of an individual to be tested, comprising detecting n SNP polymorphisms affecting the trait in the genome of the individual to be tested, and calculating the breeding value of the individual to be tested by formula (1); GEBV _x =Mean _x +e ₁ ×SNP ₁ +e ₂ ×SNP ₂ +......+e _n ×SNP _n Formula (1) In formula (1): GEBV _x is the breeding value of the trait; Mean _x is the quantified value of the average trait; e ₁ to e _n are the weight values of each SNP site; and SNP ₁ to SNP _n are numbers respectively, which are given according to the genotype of each SNP site, wherein if the genotype is If the genotype is a homozygote that promotes the quantitative value of the trait, a number greater than the benchmark value is given; if the genotype is a homozygous that reduces the quantitative value of the trait, Then it is given as a number smaller than the reference value; and formula (1) is established by the method as in any one of claims 1 to 10.

Images