A breeding method for corn husk tightness based on whole genome selection

Abstract

The present invention discloses a breeding method based on whole genome selection of corn husk tightness, belonging to the technical field of plant breeding. In order to provide a method for obtaining corn husk tightness by whole genome selection, the present invention selects the whole genome data of corn inbred lines in an associated population as original data, and after pre-processing for quality control, supplements the missing genotypes, and then screens out the gene data of different density markers, and then obtains covariance data through principal component analysis and kinship analysis, and brings the corn husk tightness data of different environments into a mixed model to obtain a BLUP value, and substitutes the genotype data, covariance data and BLUP values ​​obtained by the above analysis into different statistical models for GS analysis, selects the optimal prediction model, and further predicts the preferred husk tightness according to the genotype of the material to be tested. The method of the present invention provides a basis for GS breeding and molecular breeding of other agronomic traits of corn.

Description

Technical Field

[0001] The invention belongs to the technical field of plant breeding, and in particular relates to a breeding method for selecting corn husk tightness based on the whole genome. Background Art

[0002] Corn husks, as modified leaves that wrap the ears, can protect the kernels from pests and diseases and dehydration during development. With the continuous expansion of corn planting area, ordinary manual harvesting has been difficult to keep up, so mechanized harvesting will become an inevitable trend. Many ear traits, including husk length, number of husk layers and tightness, directly or indirectly affect the mechanized harvesting of corn. Husk tightness is a comprehensive trait based on a variety of husk measurements including husk length, husk width, and husk thickness. In the study of Jiang et al. (2020), it was found that the tightness of the husk was significantly negatively correlated with its thickness and width, indicating that the thicker and wider the husk, the tighter it will be wrapped. Gao Shuren et al. (1999) conducted research and analysis on domestic and foreign literature and believed that the tightness of the husk, as a negatively correlated trait, affects the moisture content of corn kernels. Appropriate husk tightness is conducive to the growth and development of corn and mechanical harvesting.

[0003] With the continuous development of sequencing technology and the development of molecular markers in the field of molecular breeding, marker-assisted selection (MAS) has gradually become the main force in crop improvement in recent years (Stuber et al. 1999). However, many agronomic traits of corn are quantitative traits. Some limiting factors such as marker quality, accuracy of small-effect QTL detection and environmental variation have severely limited the effective use of MAS in complex quantitative trait breeding (Hasan et al. 2021, Platten et al. 2019, Hospital et al. 2009). GS breeding is to construct a training model by analyzing the genotype and phenotype of the population, and then substitute the genotype data of the breeding population into the model to obtain GEBVs, and select the test population according to GEBVs (Cui et al. 2020).

[0004] The analysis of markers covering the entire genome enables GS to explain total genetic variation, which is superior to MAS. This feature of GS introduces the role of phenotype from screening lines to the level of model construction. How to improve the accuracy of the model constructed by GS is the key to current GS research. At present, there is no good screening method when using multi-platform sequencing genotypes for GS, which leads to low prediction accuracy of directly integrated genotypes. In addition, no research has yet conducted GS analysis on corn husk tightness. Therefore, it is necessary to provide a whole genome selection (GS) husk tightness analysis breeding method. Summary of the invention

[0005] The purpose of the present invention is to provide a method for obtaining the tightness of corn husks by whole genome selection, and then a method for breeding according to the tightness of corn husks.

[0006] The present invention provides a breeding method for selecting corn husk tightness based on the whole genome, and the steps of the breeding method are as follows:

[0007] Step 1: Select the whole genome data of corn from four gene sequencing platforms as the original data, and perform preprocessing screening for quality control on the original data;

[0008] Step 2: Supplement the missing genotypes of the data pre-processed and screened in step 1 by calculating the r of the linkage disequilibrium degree of different markers in the genome. 2 Filter out genetic data with different density markers;

[0009] Step 3: Perform principal component analysis on the gene data marked with different densities to obtain principal component analysis gene data; perform kinship analysis on the gene data marked with different densities to obtain kinship analysis gene data;

[0010] Step 4: Count the corn husk tightness data in advance and bring it into the mixed model to obtain the BLUP value;

[0011] Step 5: Perform GS analysis on the principal component analysis gene data and kinship analysis gene data obtained in step 3 and the BLUP values ​​obtained in step 4 to obtain the optimal statistical model;

[0012] Step 6: Determine the genotype of the maize inbred line to be tested, bring it into the optimal model obtained in step 5, obtain the genome estimated breeding value, and select some of the lines with the highest GEBV values ​​as the preferred breeding materials according to the needs of breeders.

[0013] It is further defined that the four sequencing platform data in step 1 are: SNP data from two gene chips of corn inbred lines, genotype data from transcriptome sequencing of corn inbred lines, and genotyping data from a simplified genome sequencing platform of corn inbred lines; the SNP data of the two gene chips are 50K and 600K.

[0014] It is further defined that the pre-processing method in step 1 is to screen based on the criteria of missing rate less than 20% and MAF greater than 0.05.

[0015] It is further defined that the genotype deletion in step 2 is determined using Beagle 4.0 software; the r2 parameters in step 2 are 0.8, 0.5, 0.2, 0.1 and 0.01, respectively.

[0016] It is further defined that in step 3, principal component analysis is performed using the function prcomp in the R language; and kinship analysis is performed using the GAPIT software package.

[0017] It is further defined that the mixed linear model described in step 4 is: , represents the phenotypic value of the “i”th family in the model, represents the mean of the phenotypes in multiple environments, is the variety effect, is the environmental effect, and is the residual.

[0018] Further defined, the specific steps of GS analysis in step 5 are as follows:

[0019] (1) Genome prediction: BayesA model, BayesB model, BayesC model, BL model, BRR model and gBLUP model were used to perform genome prediction using the R packages “rrBLUP v4.5” and “BGLR”;

[0020] (2) Prediction accuracy: 80% of the inbred lines with known genotypes of corn husk tightness data were randomly selected as the training set, and the remaining 20% ​​were used as the test set. The genotype data of the test set were imported into the prediction model, and the genome-based estimated breeding value of the test set was calculated. This was repeated hundreds of times, and the phenotypic data of the test set were used for correlation analysis between the actual breeding value and the genome-based estimated breeding value. The statistical model with the highest accuracy was the optimal one.

[0021] It is further defined that the prediction models are: BayesA model, BayesB model, BayesC model, BL model, BRR model and gBLUP model.

[0022] The present invention provides a computer device for a breeding method for selecting the tightness of corn husks, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the breeding method for selecting the tightness of corn husks based on the whole genome as described above is implemented.

[0023] The present invention provides a non-temporary computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the breeding method for selecting corn husk tightness based on the whole genome is implemented as described above.

[0024] Beneficial effects: This patent uses corn-related populations as materials, and corn husk tightness traits as phenotypic indicators. GS analysis is performed on corn husk tightness traits under different statistical models, different marker densities, different sequencing platforms, different sampling strategies, and different population structures. Multiple factors such as sequencing price, calculation time, and prediction accuracy are comprehensively considered to obtain the optimal prediction method, while providing a basis for GS breeding of other agronomic traits of corn.

[0025] The present invention performs GS analysis on different gene sequencing platforms and different marker densities of husk tightness, and selects the best platform and marker density for corn husk tightness without affecting the GS prediction accuracy, thereby greatly saving sequencing costs; in addition, the present invention selects different population structures for GS analysis of husk tightness, and selects the subpopulation with the highest GS prediction accuracy in the trait of husk tightness. If it is applied to the breeding of husk tightness, it can greatly improve the breeding efficiency. According to the GS results of corn husk tightness in this patent, the selection method can be extended to more corn populations and traits.

[0026] By comparing the prediction accuracy of sequencing platforms, molecular marker density, statistical models, sampling strategies and population structure, we attempted to explore the optimal strategy for GS breeding of important agronomic traits of maize, including bract tightness, to achieve the optimal prediction accuracy and research cost, and provide a theoretical basis for further molecular breeding. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Flow chart for screening the optimal statistical model for bract tightness GS;

[0028] Figure 2 Flow chart for screening the optimal marker density and sequencing platform for bract tightness GS;

[0029] Figure 3 Flowchart for optimal sampling strategy and population structure screening for bract tightness GS;

[0030] Figure 4 The GS prediction accuracy of bract tightness of associated populations in different genotyping platforms under different statistical models;

[0031] Figure 5 The GS prediction accuracy of bract tightness of associated populations at different marker densities in different genotyping platforms;

[0032] Figure 6 The heritability of bract tightness is estimated. The horizontal axis is the sequencing platform, and the vertical axis is the heritability;

[0033] Figure 7 is the GS prediction accuracy of bract tightness under different sampling strategies;

[0034] Figure 8 The GS prediction accuracy of bract tightness under different population structures; DETAILED DESCRIPTION

[0035] 1. Whole genome selection (GS): Based on Genomic Estimated Breeding Value (GEBV), it detects molecular markers covering the entire genome and uses genetic information at the genome level to conduct genetic evaluation of individuals to obtain higher accuracy of breeding value estimation.

[0036] 2. BLUP value: Best Linear Unbiased Prediction is a statistical method that is suitable for predicting individual breeding values, i.e. genetic assessment, in breeding. It can improve the accuracy of prediction.

[0037] 3. Correlation analysis: Correlation analysis refers to the analysis of two or more correlated variable elements to measure the degree of correlation between the two variable factors. Correlation analysis can only be performed if there is a certain connection or probability between the correlated elements.

[0038] 4. Heritability: Also known as heritability, it refers to the ratio of genetic variance to total variance (phenotypic variance), which can be used as an indicator for selecting hybrid offspring. Heritability is divided into broad heritability and narrow heritability. Quantitative traits are greatly affected by environmental factors, so phenotypic variation may have genetic factors, environmental factors, and even factors of environmental and genetic interactions.

[0039] Example 1. A breeding method based on predicting the tightness of corn husks

[0040] A breeding method based on predicting the tightness of corn husks, characterized in that the breeding method comprises the following steps:

[0041] (1) The 50K genotype was obtained by Yang et al. (2011) using DNA extracted from leaves of an associated population (513 maize inbred lines) and then using the MaizeSNP50 BeadChip gene chip to obtain genotyping, which contains 56,110 SNPs; the transcriptome genotype was obtained by Fu et al. (2013) using RNA extracted from grains of 368 inbred lines in the above-mentioned associated population 15 days after pollination and sequencing on the Illumina platform; the GBS and 600K genotypes were obtained by Liu et al. (2017) using DNA extracted from leaves of 469 and 153 inbred lines and using GBS (reduced genome sequencing) and Affymetrix AxiomMaize 600K array sequencing platforms for genotyping. The raw data was processed through the following steps to obtain five genotype files.

[0042] ① Use Tassel 5 software to open the raw data and screen with the missing rate less than 20% and MAF (minimum allele frequency) greater than 0.05 as the standard;

[0043] ② Use Beagle4.0 software to supplement missing genotypes;

[0044] ③ Use plink software to calculate the r of linkage disequilibrium (LD) degree of different markers in the genome 2 , set r 2 The parameters are 0.8, 0.5, 0.2, 0.1 and 0.01 respectively, and the marker names retained at different densities can be obtained. The larger the value, the higher the marker density. According to the retained marker names, the genotype files at different marker densities can be extracted. The genotype files are converted from base types such as ATGC to numerical types 1, 0 and 2 in the R language software;

[0045] ④Use the prcomp function in R language to perform principal component analysis to obtain the PCA file.

[0046] ⑤Use GAPIT software package to perform kinship analysis to obtain Kinship (K) file.

[0047] (2) This paper measured the bract tightness data of Sanya City, Hainan Province in 2015 and 2016 and Fushun City, Liaoning Province in 2016, and used a mixed linear model to calculate the BLUP (Best Linear Unbiased Prediction Value). The mixed linear model is: i =μ+f i +e i +ε i , in the model i represents the phenotypic value of the “i” family, μ represents the mean of the phenotype in multiple environments, and f i is the variety effect, e i is the environmental effect, ε i is the residual. For each environment, the mean is treated as a fixed effect, while the variety and environment effects are treated as random effects. Finally, the estimated values ​​of the fixed effect and the variety effect are summed, which is the BLUP value. The inbred coefficients of bract tightness determined in this paper are 438, but different genotype files contain different numbers of inbred lines of associated populations. Therefore, based on the 438 inbred lines in the phenotype file, this paper compares the names of inbred lines shared by different sequencing platforms and phenotype files, and finally generates 5 different phenotype files containing the corresponding inbred lines (the genotype integration files of the four sequencing platforms, 50K, 600K, GBS and RNA-seq have 438, 438, 133, 380 and 315 inbred lines, respectively);

[0048] (3) Perform GS analysis using the genotype file of step (1) and the phenotype file of step (2).

[0049] ① Genomic prediction model: The environment (environmental factors are the impact of different environments on prediction) and the first three principal components (PCA) are taken as fixed effects, and all additive genetic effects (genotypes) and residuals are random effects, and a mixed linear statistical model is obtained: y = μ + Xβ + Zu + ε, where y is the vector of bract tightness observations (n×1), n ​​is the family coefficient; μ is the total phenotypic mean; X is the design matrix of fixed effects (n×p), β is the vector of fixed effects (p×1), where, when p is equal to 3, the first three principal components of PCA obtained in (1) ④ are regarded as fixed effects; Z is the design matrix of random effects (n×q); u is the vector of random effects of individual total additive genetic effects (genotypes) (q×1); ε (n×1) is the residual. Random effects follow a normal distribution: u~N(0, ),ε~N(0, ), where I is the homology matrix, K is the kinship matrix from (1) ⑤, is the variance of individual additive genetic effects, is the variance of the residual. u is the genomic estimated breeding value (GEBV) that needs to be calculated in the end.

[0050] The gBLUP was used for genome prediction using the above mixed linear model through the R package “rrBLUP V4.5”, and the BayesA, BayesB, BayesC, BL, and BRR were used for genome prediction through the R package “BGLR”.

[0051] ② Calculation of prediction accuracy: 80% of all inbred lines were randomly selected as the training set, and the remaining 20% ​​were used as the test set, and repeated 100 times. The phenotypic data of the test set are the actual breeding values ​​(TBV), and the genotype and phenotypic data of the training set are used to "train" the prediction model. After the genotype data of the test set is imported into the prediction model of (3)①, the genomic estimated breeding value (GEBV) of the test set can be calculated. Pearson correlation analysis is performed on TBV and GEBV to obtain the prediction accuracy of genomic selection.

[0052] The five genotype files obtained were analyzed by GS using BayesA, BayesB, BayesC, BL, BRR and gBLUP models respectively. The prediction results were tested for significance using SPSS software, and the prediction accuracy of each model was compared. The results showed that the gBLUP model was the optimal model.

[0053] The above model is as follows:

[0054] The Bayes model is Where y is the vector of n phenotype observations; b is the fixed effect vector, X is the association matrix of fixed effects; q is the number of SNPs, z kis the genotype vector of the kth SNP, g k is the effect value of the kth SNP; e is the remaining random residual vector.

[0055] BayseA: Assume that all SNPs have an effect and the variance of all SNP effects follows a normal distribution with a scaled inverse chi-square distribution.

[0056] BayesB: Only a small number of marker sites have an effect, and most other chromosome fragments have no effect. The proportion of sites with an effect needs to be set in advance.

[0057] BayesC: Same as BayesB, but the proportion of effective sites is unknown and does not need to be preset.

[0058] BL: The variance follows a normal distribution with an exponential distribution.

[0059] BRR: It obeys Gaussian distribution, and the hyperparameter mean α and standard deviation λ obey gamma distribution.

[0060] gBLUP: y = Xb + Zg + e, y is the phenotypic vector, X is the design matrix associated with fixed effects, b is the vector of fixed effects, Z is a design matrix of genetic effects, g is the vector of additive genetic effects, and e is the vector of random normal deviations of variance. The kinship matrix constructed based on pedigree information is replaced by constructing a genomic relationship matrix.

[0061] (4) In actual breeding, the genotype of the maize inbred line to be tested is first determined, and the gBLUP statistical model obtained from the above training population is used to calculate the GEBV. According to the needs of the breeder, some breeding materials with high GEBV values ​​are selected or some breeding materials with low GEBV values ​​are eliminated. This can decide whether to keep or discard the breeding materials without the time-consuming and laborious measurement of phenotypic data.

[0062] The technical solution of the present invention is mainly divided into the following steps:

[0063] 1. Screening of the optimal statistical model

[0064] like Figure 1 As shown in the figure, the data of the four sequencing platforms of 50K, 600K, GBS and RNA-seq and the integrated genotype were processed into five genotype files in the required format, and the BLUP value of bract tightness was used as the phenotype file. 80% of the entire association population was randomly selected as the training population, and the remaining 20% ​​was used as the test population. The BayesA, BayesB, BayesC, BL, BRR models in the BGLR package in the R language software and the gBLUP model in the rrBLUP package were selected for GS analysis. The GS prediction results of bract tightness were analyzed by SPSS software for correlation analysis to obtain the prediction accuracy and screen the optimal statistical model.

[0065] 2. Screening of optimal labeling density and sequencing platform

[0066] like Figure 2 As shown, the plink software was used to adjust the r2 parameter to obtain the genotype files of different marker densities and sequencing platforms. The five sequencing platforms had six marker densities, totaling 30 genotype data. 80% of the entire association population was randomly selected as the training population, and the remaining 20% ​​was used as the test population. The optimal statistical model was selected for GS analysis. The GS prediction results of bract tightness were analyzed using SPSS software for correlation analysis to obtain the prediction accuracy, so as to screen the optimal marker density and the optimal sequencing platform.

[0067] 3. Screening of optimal sampling strategy and population structure:

[0068] like Figure 3 As shown in the figure, the sampling strategy is set as follows: 10-90% of the associated population is randomly selected as the prediction population, and the rest is the training population. The population structure is set as two treatments: within the subgroup (Within) and between subgroups (Across). Within the subgroup is to sample from the four subgroups (MIXED, NSS, SS, TST) in proportion as the training population, and other families in the same subgroup are used as the prediction population; between-subgroup GS prediction is to sample from one of the four subgroups in proportion as the training population, and all other families (including the four subgroups) are used as the prediction population. According to 1 and 2, we have selected the optimal statistical model, marker density and sequencing platform for bract tightness GS, compared the prediction accuracy of different sampling strategies and population structures, and finally screened out the optimal sampling strategy and population structure.

[0069] Example 2. Computer device for a breeding method for selecting corn husk tightness

[0070] A computer device for a breeding method for selecting corn husk tightness, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the breeding method for selecting corn husk tightness based on the whole genome as described in Example 1 is implemented.

[0071] Embodiment 3. A non-transitory computer-readable storage medium

[0072] A non-transitory computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the breeding method for corn husk tightness based on whole genome selection as described in Example 1 is implemented. The storage medium may be a read-only memory, a disk or an optical disk, etc.

[0073] The following tests are used to verify the test results:

[0074] 1. Effect of GS models on prediction accuracy under different genotyping platforms

[0075] In order to screen the most optimal statistical model for GS of husk tightness, the BayesA, BayesB, BayesC, BL, BRR models of the BGLR package in R language and the gBLUP model of the rrBLUP package were selected for comparison. In addition, a multi-genome sequencing platform (integrated, 50K, 600K, GBS, RNA-seq) was used for comprehensive consideration to compare the effects of genotyping platforms and statistical models on the prediction of GS of corn husk tightness. One of the sampling ratios was selected for illustration. 80% of the entire association population was randomly selected as the training population, and the remaining 20% ​​was used as the test population. Figure 4 As shown in the figure, the significance test found that the GS prediction accuracy of the gBLUP model was significantly higher than that of other models for the five genotyping platforms. After the gBLUP model was selected, the 50K gene chip had the highest prediction accuracy among the five sequencing platforms, with a prediction accuracy of 36.74%, followed by the GBS sequencing platform, with a prediction accuracy of 36.18%, followed by RNA-seq, integration, and 600K gene chips. The results showed that the optimal statistical model for GS prediction of bract tightness was gBLUP, and the prediction accuracy of the 50K and GBS sequencing platforms under this model was slightly higher than that of other platforms.

[0076] Figure 4 The legend in the figure represents different GS statistical models, the horizontal axis represents different genotyping platforms, and the vertical axis represents the GS prediction accuracy. The sampling strategy is random sampling in all associated populations, and the number of samplings is 100. The Duncan statistical method is selected for the difference significance test, and the results are shown in the letters a to h in the figure.

[0077] 2. Effect of marker density on GS prediction accuracy under different genotyping platforms

[0078] The original number of markers for different genotyping platforms (integrated, 50K, 600K, GBS, RNA-seq) were 196758, 47368, 436972, 887, and 93081 SNPs, respectively. In order to screen the optimal marker density for bract tightness GS, different marker density genotypes were set by changing the linkage disequilibrium (LD) r2 parameter in the plink software. The r2 is 0.01, 0.1, 0.2, 0.5, and 0.8, respectively, that is, the marker density is: the integrated genotypes have 167, 8417, 20905, 52826, and 90292 SNPs, the 50K genotypes have 113, 8485, 18292, 29537, and 35964 SNPs, the 600K genotypes have 428, 4524, 33568, 111713, and 172319 SNPs, the GBS genotypes have 64, 676, 743, 801, and 817, and the RNA-seq genotypes have 80, 4395, 14465, 33978, and 49888 SNPs. One of the sampling ratios is selected for illustration, and 80% of the entire association population is randomly selected as the training population, and the remaining 20% ​​is used as the test population. Figure 5 As shown in the figure, after significance test, it was found that the optimal marker density of different genotyping platforms was different. The 50K and GBS sequencing platforms both had the highest prediction accuracy when r2 was 1.0, that is, no original markers were deleted; the 600K, RNA-seq and integrated sequencing platforms had the highest prediction accuracy when r2 was 0.5, 0.2 and 0.8, respectively. However, except for 0.01, there was no significant difference in the prediction accuracy of all marker densities under all sequencing platforms. Therefore, considering the sequencing price and prediction accuracy, the selection of a marker density of 0.1 is low in cost and will not significantly affect the GS results. In addition, the GBS sequencing platform is cheap, and the prediction accuracy reaches 34.92% when r2 is 0.1. In summary, the use of a marker density of 0.1 (only 676 markers) under the GBS platform is the best choice for GS breeding of bract tightness.

[0079] Figure 5 The legend in the figure represents different marker densities, the horizontal axis represents different genotyping platforms, and the vertical axis represents the GS prediction accuracy. The sampling strategy is random sampling in all associated populations, and the number of samplings is 100. The Duncan statistical method is selected for the difference significance test, and the results are shown in the letters a to g in the figure.

[0080] 3. Heritability of bract tightness under different genotyping platforms

[0081] As mentioned above, the marker density with r2 of 0.1 was taken as the optimal marker density, under which the integration, 50K, 600K, GBS and RNA-seq platforms had 8417, 8485, 4524, 676 and 4395 SNP markers, respectively. Figure 6 As shown in the figure, the narrow heritability of the phenotypic values ​​of bract tightness and their BLUP values ​​in different environments were evaluated on different sequencing platforms. The heritability of bract tightness ranged from 0.02 to 0.71. The 600K gene chip sequencing platform had the highest heritability, reaching 0.71. The heritability of the GBS sequencing platform was not significantly different from that of 50K, integration, and RNA-seq, but was significantly different from that of 600K. In theory, higher heritability should bring higher GS prediction accuracy, but the prediction accuracy under the 600K sequencing platform is not high, which may be due to the large number of missing values ​​in the phenotypic data of 600K. In summary, the 600K sequencing platform has the highest heritability.

[0082] Figure 6 All bract tightness phenotypic data and their BLUP values ​​were derived from the field surveys of 15SY, 16SY and 16FS.

[0083] 4. Effect of different sampling strategies on GS prediction accuracy

[0084] So far, we have selected the optimal model gBLUP, the optimal marker density of 0.1, and the optimal sequencing platform GBS for the GS analysis of bract tightness. In order to further analyze the impact of different sampling strategies on the GS prediction accuracy, the gBLUP model was used to randomly sample the GBS sequencing platform with a marker density of 0.1. 10-90% of the entire association population was randomly selected as the prediction population, and the rest was the modeling population. As the sampling strategy of the prediction population increased, the GS prediction accuracy showed a downward trend ( Figure 7 ). Among them, when the proportion of the prediction group to the associated group is 10%, the maximum prediction accuracy is 35.24%. The results show that the maximum GS prediction accuracy is achieved when the prediction group accounts for 10%, that is, the training group accounts for 90%. Figure 7 The horizontal axis is the sampling ratio of the prediction group, and the vertical axis is the prediction accuracy. The number of random sampling is 100 times.

[0085] 5. The impact of different population structures on GS prediction accuracy

[0086] In order to analyze the effect of different population structures on the prediction accuracy of bract tightness by GS, we still used gBLUP as the optimal statistical model, the optimal marker density with r2 of 0.1 and the optimal sequencing platform of GBS. At the same time, the associated population was divided into four subgroups: MIXED, NSS, SS, and TST. Then, two population structures, within subgroup (Within) and across subgroup (Across), were used for random sampling ( Figure 8 ). Intra-subpopulation sampling means that both the prediction population and the modeling population are confined to a specific subpopulation, and random sampling is performed only within the subpopulation. For example, 10-90% of the SS subpopulation is randomly selected as the prediction population, and the remaining SS subpopulation inbred lines are the modeling population; inter-subpopulation sampling only limits the prediction population to a certain subpopulation, while the modeling population can come from any subpopulation. For example, 10-90% of the SS subpopulation is randomly selected as the prediction population, and all other inbred lines are regarded as the modeling population. In general, with the increase in the sampling proportion of the prediction population, the GS prediction accuracy shows a downward trend. In the MIXED, NSS, and SS subpopulations, the GS prediction accuracy obtained by random sampling between subpopulations is higher than that by random sampling within subpopulations, while the opposite is true in the TST subpopulation. The results show that in all subpopulations, no matter which sampling strategy is adopted, the highest prediction accuracy is shown in the SS subpopulation, indicating that the SS subpopulation is more suitable for studying the bract tightness trait of this association population.

[0087] Figure 8 The legend in the middle figure uses different colors to represent the sampling ratio of the test population. The horizontal axis represents different population structures, that is, sampling within and between subgroups, which are randomly selected from four subgroups (MIXED, NSS, SS, TST), and the vertical axis represents the prediction accuracy. The number of random sampling is 100 times.

[0088] In summary, we concluded that the most suitable model for GS analysis of bract tightness is gBLUP, the optimal marker density and sequencing platform is the GBS platform (676 markers) with a marker density LD of 0.1, the sampling strategy is to extract 10% as the prediction population, and the population structure selects the SS subpopulation. GS breeding for bract tightness in this associated population is the best.

Claims

1. A breeding method for corn husk tightness based on whole genome selection, characterized in that: The steps of the breeding method are as follows: Step 1: Select the whole genome data of corn from four gene sequencing platforms as the original data, and perform preprocessing screening for quality control on the original data; Step 2: Supplement the missing genotypes of the data pre-processed and screened in step 1 by calculating the r of the linkage disequilibrium degree of different markers in the genome. 2 Filter out genetic data with different density markers; Step 3: Perform principal component analysis on the gene data with different density markers to obtain principal component analysis gene data; Performing kinship analysis on gene data marked with different densities to obtain kinship analysis gene data; Step 4: Count the corn husk tightness data in advance and bring it into the mixed model to obtain the BLUP value; the mixed model is: , in the model represents the phenotypic value of the "i" pedigree, represents the mean of the phenotype in multiple environments, For the variety effect, It is the environmental effect. is the residual; Step 5: Perform GS analysis on the principal component analysis gene data and kinship analysis gene data obtained in step 3 and the BLUP values ​​obtained in step 4 to obtain the optimal statistical model; Step 6: Determine the genotype of the maize inbred line to be tested, and bring it into the optimal model obtained in step 4 to obtain the genome estimated breeding value. According to the needs of breeders, select some of the GEBV values ​​with the highest ranking as the preferred breeding materials.

2. The method according to claim 1, characterized in that The whole genome data of corn of the four gene sequencing platforms in step 1 are: SNP data from two gene chips of corn inbred lines, genotype data from transcriptome sequencing of corn inbred lines, and genotyping data from a simplified genome sequencing platform of corn inbred lines; the SNP data of the two gene chips are 50K and 600K.

3. The method according to claim 1, characterized in that The pre-processing screening method described in step 1 is: the missing rate is less than 20% and the MAF is greater than 0.05 as the standard for screening.

4. The method according to claim 1, characterized in that: The genotype missing complement in step 2 was performed using Beagle 4.0 software; 2 The parameters are 0.8, 0.5, 0.2, 0.1 and 0.01 respectively; calculate r 2 Using plink software.

5. The method according to claim 1, characterized in that In step 3, the principal component analysis was performed using the prcomp function in the R language, and the kinship analysis was performed using the GAPIT software package.

6. The method according to claim 1, characterized in that The specific steps of GS analysis in step 5 are as follows: (1) Genome prediction: BayesA model, BayesB model, BayesC model, BL model, BRR model and gBLUP model were used, and R packages "rrBLUP v4.5" and "BGLR" were used for genome prediction; (2) Prediction accuracy: 80% of the inbred lines with known genotypes of corn husk tightness data were randomly selected as the training set, and the remaining 20% ​​were used as the test set. After the genotype data of the test set were imported into the prediction model, the genomic estimated breeding value of the test set was calculated. This was repeated hundreds of times, and the phenotypic data of the test set were used for correlation analysis between the actual breeding value and the genomic estimated breeding value. The statistical model with the highest accuracy was the optimal one.

7. The method according to claim 6, characterized in that The prediction models are: BayesA model, BayesB model, BayesC model, BL model, BRR model and gBLUP model.

8. A computer device for a breeding method for selecting the tightness of corn husks, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the breeding method for corn husk tightness based on whole genome selection 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, the breeding method for corn husk tightness based on whole genome selection is implemented as described in any one of claims 1 to 7.

Images