Method for predicting pig calcium and phosphorus digestibility based on multi-scale additive snp weight rkhs model

Abstract

This invention discloses a method for predicting the digestibility of calcium and phosphorus in pigs based on a multi-scale additive SNP weighted RKHS model, comprising: S1, obtaining... n First pig m The genotype data and calcium and phosphorus digestibility phenotypic data corresponding to each SNP chip were preprocessed to obtain a training set; S2, single-label additive association analysis was performed on calcium and phosphorus digestibility in the training set to obtain additive association statistics; S3, SNP weights were obtained based on allele frequency regulation terms and multi-scale local additive association evidence; S4, after constructing weighted SNP distances based on SNP weights, the weighted genomic similarity of different pigs under the target trait was characterized by kernel functions, thereby obtaining a weighted RKHS kernel matrix; S5, based on... K t An RKHS prediction model was constructed to output predicted values ​​for calcium and phosphorus digestibility in individual pigs. This invention can predict calcium and phosphorus digestibility separately, providing technical support for screening pigs with high mineral utilization efficiency and potentially low phosphorus emissions.

Description

Technical Field

[0001] This invention relates to the field of predicting calcium and phosphorus digestibility in pigs. More specifically, this invention relates to a method for predicting calcium and phosphorus digestibility in pigs based on a multi-scale additive SNP weighted RKHS model. Background Technology

[0002] Calcium and phosphorus are essential mineral elements for pig growth, skeletal development, energy metabolism, and feed utilization efficiency. The digestibility of calcium and phosphorus in pigs is influenced by multiple factors, including dietary calcium and phosphorus levels, phytase supplementation, body weight, growth stage, breed, and individual genetic background. Existing methods for predicting calcium and phosphorus digestibility in pigs mainly rely on nutritional variables or conventional statistical models, with limited direct use of host genomic information.

[0003] With the development of pig SNP microarrays and genomic selection technologies, predicting complex quantitative traits using individual SNP genotypes has become an important technical approach in animal breeding. However, traditional GBLUP or P-BLUP typically assumes that all SNPs contribute equally to the genetic variance, failing to adequately consider the differences in the contribution of different SNPs and their respective LD regions to calcium and phosphorus digestibility.

[0004] The existing technology has the following main shortcomings: 1. Standard GBLUP or unweighted RKHS models assume that all SNPs contribute equally, making it difficult to highlight key genomic regions related to calcium and phosphorus digestibility.

[0005] 2. The single-point GWAS weighting method is easily affected by the random significance of a single SNP and has insufficient noise resistance.

[0006] 3. Single-window smoothing methods have difficulty simultaneously capturing sharp QTL signals, moderate LD block signals, and multigene signals over a wide region.

[0007] 4. Weighting methods that do not consider the stability of allele frequencies may cause low-frequency SNPs to receive excessive weights due to unstable estimation.

[0008] 5. Existing nutritional phenotype prediction models focus primarily on dietary and environmental variables, and do not adequately utilize the differences in Ca / P digestibility caused by the individual genetic background of pigs. Summary of the Invention

[0009] One object of the present invention is to solve at least the above-mentioned problems and / or defects, and to provide at least the advantages described below.

[0010] To achieve these objectives and other advantages of the present invention, a method for predicting the digestibility of calcium and phosphorus in pigs based on a multi-scale additive SNP weighted RKHS model is provided, comprising: S1, Obtain n First pig m The genotype data and calcium and phosphorus digestibility phenotypic data corresponding to each SNP chip were obtained by preprocessing to obtain the training set. S2. Perform single-label additive correlation analysis on calcium digestibility and phosphorus digestibility separately in the training set to obtain the additive correlation statistic. ,in, ; S3, the first equation is obtained through the following formula. j Each SNP corresponds to the target trait. t SNP original additive weights Then, normalization is performed to obtain the SNP weights: in, and Here, R represents the allele frequency regulation parameter, and R is the number of window sizes. For the first r The number of SNPs contained in each window. For frequency adjustment, For the first j Additive association statistic of SNPs, For the first r The first scale j The effective window of a SNP, and S4, SNP-based weights Constructing weighted SNP distance Subsequently, the Gaussian RKHS kernel function was used to characterize the target trait in different pig individuals. t Weighted genomic similarity This leads to the weighted RKHS kernel matrix. ; S5, based on An RKHS prediction model was constructed to output predicted values ​​for calcium and phosphorus digestibility for individual pigs.

[0011] Preferably, in S1, the preprocessing includes quality control, missing data filling, and additive coding of the genotype data; The quality control refers to setting the SNP detection rate to be no less than 0.90, the individual detection rate to be no less than 0.90, and the minimum allele frequency (MAF) to be no less than 0.01. The deletion imputation is performed by filling in the missing genotype using any one of the following: mean, population frequency, or haplotype. The additive coding is used to characterize the additive genetic differences of pig individuals on whole-genome SNPs by constructing an additively centered genotype coding matrix Z using the following formula: In the above formula, Indicates the first j Allele frequency of each SNP Indicates the first i The first pig in j Reference or effector allele doses on each SNP.

[0012] Preferably, in S2, the additive correlation statistic is obtained as follows: S20. In each training set, single-label additive association analysis is performed on calcium digestibility and phosphorus digestibility respectively using the following formula, then the... j The model corresponding to each SNP is: In the formula, Indicates the first i The first pig in the target trait t The digestibility phenotype is as follows: The intercept is... Indicates the first j Each SNP corresponds to the target trait. t The additive effect Indicates the first i The covariate vector of the first pig, For fixed effects parameters, For random residuals; S21. Fit the model from S20 to obtain an estimate of the additive effect of SNP. and standard error ; S22, the additive correlation statistic Obtained through the following formula: .

[0013] Preferably, in S3, the number of window scales is set. R =3, and take S 1=2, S 2 = 10 S If 3 = 40, then the corresponding window sizes are 5, 21, and 8 respectively; Multiple window sizes are simply averaged using the following formula: in, Indicates the first j A SNP in the target trait tEvidence of multi-scale local additive correlations Indicates the first j Centered on a SNP, with a window size of [missing information] s of Moving average; This represents the square of the additive correlation statistic.

[0014] Preferably, in S3, the normalization process refers to obtaining the original additive weights of the SNPs. Then, the SNP weights are normalized to the mean using the following formula: The normalized SNP weights satisfy: .

[0015] Preferably, in S4, the weighted SNP distance It is characterized by the following formula: In the above formula, Indicates the first i The first pig in j Genotype coding values ​​at each SNP locus Indicates the first The first pig in j Genotype coding values ​​at each SNP locus; The weighted genome similarity It is characterized by the following formula: In the above formula, The Gaussian kernel bandwidth parameter is selected from the candidate set through cross-validation in the training set. It is an exponential function.

[0016] Preferably, in S5, the RKHS prediction model is characterized by the following formula: In the above formula, This represents a phenotypic vector for calcium digestibility or phosphorus digestibility. For the fixed effects matrix, For fixed effects parameters, For weighted RKHS kernel matrix, Kernel regression coefficients This is the residual vector.

[0017] Preferably, in S5, the fixed effects and RKHS effects in the RKHS prediction model are jointly estimated using the following formula: In the above formula, These are estimates of the fixed effects. These are the estimated values ​​of the RKHS kernel regression coefficients. G is the regularization parameter selected through cross-validation within the training set, and G is the regularization matrix.

[0018] The present invention has at least the following beneficial effects: Firstly, this invention can simultaneously incorporate covariates such as body weight, dietary calcium, dietary phosphorus, phytase, breed, and batch, making it suitable for actual pig nutrition and breeding data. At the same time, this invention can separately predict calcium digestibility and phosphorus digestibility, providing technical support for screening pigs with high mineral utilization efficiency and potential low phosphorus emissions.

[0019] Secondly, compared to ordinary GBLUP or unweighted RKHS, this invention assigns different weights to different SNPs, which can highlight the genomic regions related to the digestibility of calcium and phosphorus in pigs. Thirdly, compared to single-point GWAS weighting, this invention uses in-window... The average value can reduce the noise caused by the random significance of a single SNP.

[0020] Fourth, compared to single-window weighting, this invention uses simple averaging of multiple window scales such as 5, 21, and 81, which can simultaneously capture peak QTLs, moderate LD blocks, and multi-gene signals over a wider region.

[0021] Fifth, this invention reduces the impact of low-frequency SNP estimation instability on the model by using the MAF frequency adjustment term, thereby improving the generalization ability of the prediction model.

[0022] Sixth, this invention directly embeds multi-scale SNP weights into the Gaussian RKHS kernel function, making high-weight SNPs have a greater impact on genetic similarity between individuals, thereby improving the ability to model complex nutritional phenotypes.

[0023] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0026] A method for predicting calcium and phosphorus digestibility in pigs based on a multi-scale additive SNP weighted RKHS model is proposed. This method employs an SNP weighting approach that integrates SNP frequency stability with evidence of multi-scale local additive associations, embedding these weights into the RKHS kernel function to achieve genomic prediction of calcium and phosphorus digestibility in pigs. Figure 1 As shown, its processing flow includes the following steps.

[0027] Step 1: SNP chip data input This step primarily involves obtaining genotypic data from porcine SNP microarrays and phenotypic data on calcium and phosphorus digestibility. Assume there are a total of... n First pig m Each SNP. Porcine SNP chips can be common 50K or 60K SNP chips for pigs, or other medium- or high-density porcine SNP chips. Genotype data are denoted as: in, Indicates the first i The first pig in j Reference allele or effector allele dose on each SNP; n For the number of individual pigs, m The number of SNP tags.

[0028] The phenotypes for calcium digestibility and phosphorus digestibility are denoted as follows: The model can include both fixed effects and covariates. Let the first... i The covariate vector of the first pig is: Fixed effects matrix From each individual Composed line by line. Among them, For weight, For dietary calcium levels, For dietary phosphorus levels, The amount of phytase added can also include gender, batch, variety, test site, and growth stage.

[0029] Step 2: SNP quality control, missing word filling, and additive coding SNP microarray data should be quality controlled, with SNP detection rates and individual detection rates set at no less than 0.90, and minimum allele frequency (MAF) at no less than 0.01. Deleted genotypes can be filled using mean imputation, population frequency imputation, or haplotype imputation methods.

[0030] No. j The allele frequencies of each SNP are calculated as follows: in, Indicates the first j Allele frequency of each SNP Indicates the first i The first pig in j Reference or effector allele doses on each SNP.

[0031] Constructing an additively centered genotype coding matrix Z : This additive coding is used to characterize the additive genetic differences in SNPs across the entire genome in individual pigs.

[0032] Step 3, Single-label additive association analysis This step primarily involves performing single-label additive association analysis on calcium and phosphorus digestibility separately within the training set, targeting the desired trait. t This indicates calcium digestibility or phosphorus digestibility, i.e.: For the first j The following model is established for each SNP: In the above formula, Indicates the first i The first pig in the target trait t The digestibility phenotype is as follows: The intercept is... Indicates the first j Each SNP corresponds to the target trait. t The additive effect Indicates the first i The covariate vector of the first pig, For fixed effects parameters, For random residuals; After fitting the model, the estimated values ​​and standard errors of the additive effects of SNPs were obtained: .

[0033] And calculate the standardized additive association statistic: .

[0034] Step 4: Calculation of Multi-Scale Additive SNP Weights This step mainly involves constructing SNP additive weights based on allele frequencies and additive association statistics within a multi-scale window, and then normalizing the SNP additive weights to make the average weight 1.

[0035] 1. Multiscale additive SNP weighting function To avoid window overflows at boundary SNPs, the first... r The first scale jThe effective window for each SNP is: Its effective window size is . No. j Each SNP corresponds to the target trait. The original additive weights are defined as follows: in, For the first j Each SNP has its original additive weights for the target trait t; and Here, R represents the allele frequency regulation parameter, and R is the number of window sizes. For the first r The number of SNPs contained in each window is determined by multiplying the "frequency stability term" and the "multi-scale local correlation term," which together determine the contribution of SNPs to the subsequent RKHS kernel.

[0036] The frequency adjustment term in the above formula is: when When the value is 1, it indicates that no MAF adjustment is performed; when In this case, medium-frequency SNPs receive relatively high weights, while low-frequency SNPs receive relatively low weights, thereby reducing the impact of the instability in low-frequency SNP estimation on the model. Preferably, it can be set that... or Alternatively, selection can be made in the training set through cross-validation.

[0037] The window term in the above formula is: This item indicates the first j The average of the squared additive association statistics of SNPs within a window containing each SNP is used to reflect the overall additive association strength of that region. Compared to single-point GWAS weighting, window averaging can reduce the noise caused by the accidental significance of individual SNPs.

[0038] 2. Recommended window settings and simple averaging fusion In a preferred embodiment, the following is provided: R =3, and take S 1 = 2, S 2 = 10 S 3=40, then the corresponding window sizes are 2 S 1+1=5, 2 S 2+1=21, 2 S 3 + 1 = 81.

[0039] A simple average is used directly for multiple window scales: in, Indicates the first j Centered on a SNP, with a window size of [missing information] s of Moving average. This simple averaging method avoids introducing additional window weight parameters while preserving complementary information from small, medium, and large windows.

[0040] 3. SNP weight mean normalization After calculating the original weights, the SNP weights are normalized to their mean: After normalization, the following conditions are met: This process ensures that the weights only change the relative contributions of different SNPs, without altering the overall weight scale across the entire genome.

[0041] Step 5: Construct the weighted Gaussian RKHS kernel function This step mainly involves constructing the Gaussian RKHS kernel function based on the weighted SNP distance, and estimating the RKHS model parameters through MAP or regularized least squares.

[0042] 1. Weighted Gaussian RKHS kernel function For target traits t We construct the weighted SNP distance using the final normalized additive SNP weights: Then construct the weighted Gaussian RKHS kernel function: in, Indicates the first i The first pig and the second The first pig in the target trait t Weighted genomic similarity; This refers to the Gaussian kernel bandwidth parameter. If the two pigs differ significantly on high-weight SNPs, then... As the number of cells increases, the RKHS similarity decreases.

[0043] 2. Constructing the RKHS prediction model For target traits The following RKHS prediction model is established: In the above formula, This represents a phenotypic vector for calcium digestibility or phosphorus digestibility. For the fixed effects matrix, For fixed effects parameters, For weighted RKHS kernel matrix, Kernel regression coefficients This is the residual vector.

[0044] If the phenotype is first corrected for by fixed effects, it can also be written as: 3. Estimate the RKHS parameters This invention estimates RKHS model parameters using MAP or regularized least squares. For corrected phenotypes... The objective function is: in, Let G be the regularization parameter, and G be the regularization matrix. G can be either the identity matrix I or the additive genomic relation matrix.

[0045] The corresponding closed-form solution of MAP is: When G=Ⅰ, the expression degenerates into the standard kernel ridge regression form: If the fixed effects and RKHS effects are estimated jointly, then the following normal equations must be solved: in, These are estimates of the fixed effects. The values ​​are estimates of the RKHS kernel regression coefficients; this joint equation allows covariate effects and genomic kernel effects to be estimated within the same training set.

[0046] 4. Selection of Gaussian kernel bandwidth and regularization parameters Gaussian kernel bandwidth parameters Candidates can be selected from the candidate set using cross-validation within the training set, for example: Regularization parameters Selection can be achieved through cross-validation within the training set, for example: Preferably, the selection criterion is the highest predicted relevance or the lowest RMSE.

[0047] Step 6: Output the predicted calcium and phosphorus digestibility values ​​for the individual pig to be predicted. For the pig individual to be predicted, additive coding is constructed using the allele frequencies estimated from the training set: In the above formula, For the additive coding of the individual pig to be predicted, For the individual pig to be predicted in the first j Reference or effector allele doses on each SNP.

[0048] Calculate the relationship between the individual pig to be predicted and the number of pigs in the training set. i Weighted distance of the first pig : Obtain the kernel vector elements: In the above formula, For the training set i The target trait and the individual pigs to be predicted t Weighted genomic similarity.

[0049] Final predicted value of the individual pig to be predicted for: If the phenotype has been corrected for fixed effects in advance, it can be written as: when t=Ca When, output the predicted value of calcium digestibility in pigs; when t=P At that time, the predicted value of swine phosphorus digestibility is output.

[0050] It should be noted that, to avoid data leakage, all SNP weights, kernel function parameters, and RKHS model parameters should be estimated within the training set. The validation or test set is only used for final prediction performance evaluation. For each cross-validation or training fold, the following steps should be followed: 1. Calculate the allele frequency of each SNP based on the genotypes in the training set: In the above formula, Indicates the first training set j Allele frequencies at each SNP locus This represents the number of individuals in the training set used to calculate allele frequencies.

[0051] 2. Construct an additively centered genotype coding matrix based on allele frequencies in the training set: In the above formula, Indicates the first training set i The individual in the first j Genotype coding values ​​at each SNP locus after allele frequency centering.

[0052] 3. Target traits within the training set Perform single-label additive association analysis to obtain the estimated additive effect value for each SNP. and its standard error .

[0053] 4. Calculate the standardized additive correlation statistic within the training set: 5. Calculate the weighting function of the multi-scale additive SNP. j The original additive weights of each SNP Then, mean normalization is performed to obtain the SNP weights. : 6. Construct a weighted RKHS kernel matrix for the training set based on the additive coding matrix and SNP weights: In the above formula, Indicates the first training set i The first pig and the second The first pig in the target trait t Weighted genomic similarity; m This represents the total number of SNP tags involved in the calculation; Indicates the first training set j A SNP in the target trait t Additive weights; Indicates target traits t The corresponding kernel bandwidth parameters.

[0054] 7. Select the RKHS kernel bandwidth parameter within the training set. and regularization parameters .

[0055] 8. Estimating fixed effects parameters based on the training set and RKHS kernel regression coefficients .

[0056] For individuals in the test set, use the estimates obtained from the training set. , and Construct the kernel vector between the test set and the training set: in, .

[0057] Then, based on the model parameters obtained from training, output the predicted values ​​of calcium digestibility and phosphorus digestibility for the test set: in, and They represent the first in the test set. s Predicted values ​​for calcium and phosphorus digestibility per pig.

[0058] Finally, the model's prediction performance was calculated using the true phenotypes and predicted values ​​on the test set, including the prediction correlation coefficient, root mean square error, and mean absolute error: Through the above training and validation process, it can be ensured that the construction of SNP weights, kernel matrix construction, parameter selection and model training are all strictly limited to the training set, thereby ensuring the independence and reliability of the calcium and phosphorus digestibility prediction results.

[0059] Example: Several pigs were selected as research subjects, and genotypic data for each pig were obtained using a common 50K SNP chip. Simultaneously, calcium digestibility for each pig was measured using the total fecal collection method, indicator method, or standardized total digestible tract digestibility measurement procedure. and phosphorus digestibility It also records information on covariates related to digestibility, including body weight. Dietary calcium levels Dietary phosphorus levels And phytase addition amount .

[0060] The aforementioned covariates are not recorded and discarded individually, but are used continuously in single-label association analysis, fixed-effects correction, and the final prediction model. Specifically, for the... i Construct a covariate vector using a single pig: If sex, batch, variety, or growth stage is also recorded, it should be included as a dummy variable or a grouped fixed effect. All trained individuals Composition of fixed effects matrix .

[0061] In single-label additive association analysis Used to deduct the systemic effects of body weight, dietary mineral levels, and phytase dosage on calcium and phosphorus digestibility, Closer to the first j The additive genetic contribution of each SNP to the target digestibility trait; in the RKHS prediction model. and They appear simultaneously, corresponding to the covariate explanation part and the weighted genome similarity explanation part, respectively.

[0062] For the individual pigs to be predicted, their body weight, dietary calcium level, dietary phosphorus level, and phytase supplementation were also recorded to form... If a joint model is used, the predicted values ​​are output as follows: If fixed-effects correction is performed first, the corrected genome predictions are obtained first, and then the corresponding fixed-effects predictions are added back.

[0063] SNP microarray data underwent quality control to remove SNPs with low detection rates, high deletion rates, or excessively low minimum allele frequencies. After quality control, the remaining SNPs were... m One valid SNP marker was identified, and the genotype was encoded as: in, Indicates the first i The first pig in j Reference allele or effector allele dose on each SNP; m The number of valid SNP markers retained after quality control.

[0064] For each training set, first calculate the... j Allele frequencies of each SNP: And construct an additively centered genotype matrix: Then, using calcium digestibility and phosphorus digestibility as target traits, single-label additive association analysis was performed in the training set to obtain the estimated additive effect value for each SNP. and its standard error ,and ,in, t=Ca Indicates calcium digestibility, when t=P This represents the phosphorus digestibility; the above estimates are then used in calculations. And multi-scale SNP weights.

[0065] In this embodiment, the number of multi-scale windows is set to... R =3, and take S 1=2, S 2 = 10 S If 3 = 40, then the corresponding window sizes are 2. S 1+1=5, 2 S 2+1=21、2 S 3 + 1 = 81.

[0066] Multiple window scales are merged using a simple averaging method, namely: MAF adjustment parameters are set as follows: or .

[0067] For target traits t , No. j The original additive weights of each SNP are calculated as follows: Then, the original weights are normalized to the mean: Make: Regarding calcium digestibility, utilizing Construct a weighted RKHS kernel matrix: For phosphorus digestibility, utilize Construct a weighted RKHS kernel matrix: Subsequently, prediction models for calcium digestibility and phosphorus digestibility were established respectively: in, The fixed-effects matrix consists of the weight of each pig, dietary calcium level, dietary phosphorus level, phytase addition, and optional covariates such as sex, batch, and breed. and These are the fixed-effect parameters in the calcium digestibility and phosphorus digestibility models, respectively. and These are the kernel regression coefficients in the two RKHS models, respectively; and This is the residual vector. Thus, the covariates measured and recorded at the beginning of the example are used in subsequent formulas... , and It has been explicitly used.

[0068] During training, the kernel bandwidth parameter is selected through in-training set cross-validation. , and regularization parameters , For phenotypes that have been corrected for fixed effects, the RKHS kernel regression coefficients can be estimated using the MAP closed-form solution: in, Represents the weighted RKHS kernel matrix of the corresponding traits. Here, G is the regularization parameter, and G is the regularization matrix, which can be either an identity matrix or an additive genomic relation matrix. This represents the phenotype after correction for fixed effects; when the phenotype is not pre-corrected, the aforementioned joint normal equation can be used to estimate both the fixed effects and the kernel regression coefficients simultaneously.

[0069] For the individual pig to be predicted, the allele frequencies estimated from the training set are used. SNP weights and kernel bandwidth parameters Construct kernel vectors between it and individuals in the training set. And combined with the covariate vector of the individual pig. Output predicted values: in: Finally, the predicted calcium digestibility values ​​for each individual pig were obtained. and predicted phosphorus digestibility .

[0070] This embodiment demonstrates that the method can unify pig SNP chip data, measured phenotypes of calcium and phosphorus digestibility, and covariates such as body weight, dietary calcium level, dietary phosphorus level, and phytase addition into the same modeling process: covariates are used for fixed-effect correction and new individual fixed-effect prediction, and SNP data are used to construct trait-specific multi-scale additive weights and weighted RKHS kernels, thereby achieving individualized prediction of calcium and phosphorus digestibility in pigs.

[0071] The above solution is merely an illustration of a preferred example and is not limited thereto. When implementing this invention, appropriate substitutions and / or modifications can be made according to the user's needs.

[0072] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Other modifications can be readily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A method for predicting the digestibility of calcium and phosphorus in pigs based on a multi-scale additive SNP weighted RKHS model, characterized in that, include: S1, Obtain n First pig m The genotype data and calcium and phosphorus digestibility phenotypic data corresponding to each SNP chip were obtained by preprocessing to obtain the training set. S2. Perform single-label additive correlation analysis on calcium digestibility and phosphorus digestibility separately in the training set to obtain the additive correlation statistic. ,in, ; S3, the first equation is obtained through the following formula. j Each SNP corresponds to the target trait. t SNP original additive weights Then, normalization is performed to obtain the SNP weights: in, and Here, R represents the allele frequency regulation parameter, and R is the number of window sizes. For the first r The number of SNPs contained in each window. For frequency adjustment, For the first j Additive association statistic of SNPs, For the first r The first scale j The effective window of a SNP, and S4, SNP-based weights Constructing weighted SNP distance Subsequently, the Gaussian RKHS kernel function was used to characterize the target trait in different pig individuals. t Weighted genomic similarity This leads to the weighted RKHS kernel matrix. ; S5, based on An RKHS prediction model was constructed to output predicted values ​​for calcium and phosphorus digestibility for individual pigs.

2. The method for predicting porcine calcium and phosphorus digestibility based on a multi-scale additive SNP weighted RKHS model as described in claim 1, characterized in that, In S1, the preprocessing includes quality control, missing data imputation, and additive coding of genotype data; The quality control refers to setting the SNP detection rate to be no less than 0.90, the individual detection rate to be no less than 0.90, and the minimum allele frequency (MAF) to be no less than 0.

01. The deletion imputation is performed by filling in the missing genotype using any one of the following: mean, population frequency, or haplotype. The additive coding is used to characterize the additive genetic differences of pig individuals on genome-wide SNPs by constructing an additively centered genotype coding matrix Z using the following formula: In the above formula, Indicates the first j Allele frequency of each SNP Indicates the first i The first pig in j Reference or effector allele doses on each SNP.

3. The method for predicting swine calcium and phosphorus digestibility based on a multi-scale additive SNP weighted RKHS model as described in claim 1, characterized in that... In S2, the additive correlation statistic is obtained as follows: S20. In each training set, single-label additive association analysis is performed on calcium digestibility and phosphorus digestibility respectively using the following formula, then the... j The model corresponding to each SNP is: In the above formula, Indicates the first i The first pig in the target trait t The digestibility phenotype is as follows: The intercept is... Indicates the first j Each SNP corresponds to the target trait. t The additive effect Indicates the first i The covariate vector of the first pig, For fixed effects parameters, For random residuals; S21. Fit the model from S20 to obtain an estimate of the additive effect of SNP. and standard error ; S22, the additive correlation statistic Obtained through the following formula: 。 4. The method for predicting porcine calcium and phosphorus digestibility based on a multi-scale additive SNP weighted RKHS model as described in claim 1, characterized in that... In S3, set the number of window sizes. R =3, and take S 1 = 2, S 2 = 10 S If 3 = 40, then the corresponding window sizes are 5, 21, and 81 respectively; Multiple window sizes are simply averaged using the following formula: in, Indicates the first j A SNP in the target trait t Evidence of multi-scale local additive correlations Indicates the first j Centered on a SNP, with a window size of [missing information] s of Moving average; This represents the square of the additive correlation statistic.

5. The method for predicting porcine calcium and phosphorus digestibility based on a multi-scale additive SNP weighted RKHS model as described in claim 1, characterized in that... In S3, the normalization process refers to obtaining the original additive weights of the SNPs. Then, the SNP weights are normalized to the mean using the following formula: The normalized SNP weights satisfy: 。 6. The method for predicting porcine calcium and phosphorus digestibility based on a multi-scale additive SNP weighted RKHS model as described in claim 1, characterized in that, In S4, the weighted SNP distance It is characterized by the following formula: In the above formula, Indicates the first i The first pig in j Genotype coding values ​​at each SNP locus Indicates the first The first pig in j Genotype coding values ​​at each SNP locus; The weighted genome similarity It is characterized by the following formula: In the above formula, The Gaussian kernel bandwidth parameter is selected from the candidate set through cross-validation in the training set. It is an exponential function.

7. The method for predicting swine calcium and phosphorus digestibility based on a multi-scale additive SNP weighted RKHS model as described in claim 1, characterized in that... In S5, the RKHS prediction model is characterized by the following equation: In the above formula, This represents a phenotypic vector for calcium digestibility or phosphorus digestibility. For the fixed effects matrix, For fixed effects parameters, For weighted RKHS kernel matrix, Kernel regression coefficients This is the residual vector.

8. The method for predicting porcine calcium and phosphorus digestibility based on a multi-scale additive SNP weighted RKHS model as described in claim 7, characterized in that... S5 also includes a joint estimation of the fixed effects and RKHS effects in the RKHS prediction model using the following formula: In the above formula, These are estimates of the fixed effects. These are the estimated values ​​of the RKHS kernel regression coefficients. G is the regularization parameter selected through cross-validation within the training set, and G is the regularization matrix.

Images