A block-adaptive graph regularization method and apparatus for whole-genome selection

Abstract

This invention provides a block-adaptive graph regularization method and apparatus for whole-genome selection, relating to the field of genome prediction technology. The method includes: obtaining the encoding of gene information from the genotype data of an individual to determine the corresponding genotype matrix; constructing a linkage disequilibrium graph based on each single nucleotide polymorphism (SNP) site in the genotype data, and dividing the SNP sites in the linkage disequilibrium graph into blocks to obtain linkage disequilibrium blocks; determining regularization constraint weights based on the block size of the linkage disequilibrium blocks, and constructing a whole-genome selection model by combining the linkage disequilibrium blocks and the genotype matrix to predict phenotypic data and obtain the genome selection result of the individual. This invention, in whole-genome selection modeling, effectively incorporates LD structure information and achieves adaptive regularization adjustment for differences in the scale of different LD blocks, thus balancing prediction accuracy and overcoming the shortcomings of LD structure modeling.

Description

Technical Field

[0001] This invention relates to the field of genome prediction technology, and more specifically to a block-adaptive graph regularization method and apparatus for whole-genome selection. Background Technology

[0002] Genomic selection (GS), a technique that uses high-density molecular markers to predict complex traits, has been widely applied in plant and animal breeding. Existing GS methods are typically based on a linear regression framework, modeling a large number of single nucleotide polymorphism (SNP) markers to estimate their contribution to phenotypic traits. Commonly used GS models include the GBLUP method based on optimal linear unbiased prediction and various Bayesian regression models. These methods often make simplified assumptions about the correlation between SNPs, either treating SNPs as independent or indirectly characterizing linkage disequilibrium (LD) relationships only through global shrinkage parameters, thus ignoring to some extent the more complex dependency structures between SNPs in the genome. Furthermore, LD relationships are not simple pairwise correlations at the genomic scale, but rather exhibit structured features with a block-like organizational form. SNPs within the same LD block are often highly correlated, forming locally densely connected substructures. Different blocks differ significantly in size, association strength, and genetic background. Ignoring this structured dependency may lead to insufficient estimation of SNP effects, thereby affecting the accuracy of genome prediction.

[0003] In existing technologies, graph regularization methods have been introduced into the field of genome prediction. By modeling the dependencies between SNPs as a graph structure, and then using the graph structure to construct a GS model to achieve genome selection, this improves the model's ability to characterize structural information to some extent. However, existing graph-based GS methods have shortcomings in modeling LD structure, affecting the model's prediction accuracy. Specifically, these methods typically apply a uniform regularization strength to the regularization parameters of different LD blocks in the GS model prediction. This fails to distinguish the structural differences of LD blocks at different scales during GS model construction and parameter solving, easily leading to over-smoothing in large LD blocks, weakening the GS model's ability to distinguish local genetic signals, while insufficient regularization in small LD blocks may result in unstable effect estimation. Summary of the Invention

[0004] In view of this, it is necessary to provide a genome-wide selection method and device with block adaptive graph regularization. In the process of genome-wide selection modeling, on the basis of effectively introducing LD structure information, adaptive regularization adjustment is achieved for the differences in scale of different LD blocks, so as to take into account the prediction accuracy and overcome the defects in LD structure modeling.

[0005] To address the above problems, this invention provides a block-adaptive graph regularized whole-genome selection method, comprising: Acquire genotype data and corresponding phenotypic data of biological individuals, and determine the corresponding genotype matrix based on the encoding of gene information in the genotype data; Based on each single nucleotide polymorphism site in the genotype data, a linkage disequilibrium graph is constructed, and the single nucleotide polymorphism sites in the linkage disequilibrium graph are divided into blocks to obtain linkage disequilibrium blocks. Regularization constraint weights are determined based on the block size of the linkage imbalance block, and a whole-genome selection model is constructed based on the regularization constraint weights, the linkage imbalance block, and the genotype matrix. The regularization constraint weights are negatively correlated with the block size, and the regularization constraint weights are used to adjust the regularization parameters of the linkage imbalance block during whole-genome selection model prediction. The genome-wide selection model is invoked to predict the phenotypic data, thereby obtaining the genome selection results of the biological individual.

[0006] In one possible implementation, the chain imbalance graph is constructed as follows: The single nucleotide polymorphism sites in the genotype data are used as nodes in the linkage disequilibrium graph; Determine the imbalance coefficient between the nodes; When the imbalance coefficient is greater than a preset threshold, an edge is established between the nodes; Based on the nodes and edges, a chain imbalance graph is constructed.

[0007] In one possible implementation, the step of dividing the single nucleotide polymorphism sites in the linkage disequilibrium graph into blocks to obtain linkage disequilibrium blocks includes: Based on the physical location of the single nucleotide polymorphism sites on the chromosome in the linkage disequilibrium graph, the nodes in the linkage disequilibrium graph are divided to obtain linkage disequilibrium blocks; Alternatively, based on the linkage disequilibrium block information pre-labeled in the bioinformatics database of the biological individual, the single nucleotide polymorphism sites in the linkage disequilibrium graph are divided to obtain linkage disequilibrium blocks; Alternatively, a clustering model can be used to perform cluster analysis on the single nucleotide polymorphism sites in the linkage disequilibrium diagram to obtain linkage disequilibrium blocks.

[0008] In one possible implementation, the adjustment process based on the regularization constraint weights includes: The regularization parameters of the linkage imbalance block in the whole-genome selection model prediction are weighted based on the regularization constraint weights to obtain the regularization parameters of the linkage imbalance block after adjustment. The weighting process of the regularization parameters is expressed by the following formula:

[0009] in, This represents the regularization parameter used in genome-wide selection model prediction of the linkage disequilibrium blocks. This represents the regularization parameter after adjustment for the k-th chain imbalance block. This represents the weight of the regularization constraint. This represents the number of nodes in the k-th chain of unbalanced blocks. This is a preset positive real number, representing the scaling adjustment coefficient.

[0010] In one possible implementation, constructing a genome-wide selection model based on the regularization constraint weights, the linkage imbalance blocks, and the genotype matrix includes: Based on the regularization parameter of linkage disequilibrium blocks in the whole-genome selection model prediction, the effect parameter of single nucleotide polymorphism sites, the genotype matrix, and the phenotypic vectors corresponding to the phenotypic data, a target optimization function is constructed. A preset graph regularized regression model is obtained, and the objective optimization function is determined as the optimization function of the graph regularized regression model to obtain a genome-wide selection model.

[0011] In one possible implementation, the objective optimization function is expressed as follows:

[0012] in, This represents the genotype matrix. This represents the phenotypic vector corresponding to the phenotypic data. A vector representing the effector parameters of a single nucleotide polymorphism site. This represents the effect parameter vector of the single nucleotide polymorphism site in the k-th linkage disequilibrium block. This represents the regularization parameter after adjusting the k-th chain imbalance block. The graph Laplace matrix corresponding to the k-th chained unbalanced block. This indicates the calculation of the L2 norm. The hyperparameter represents the regularization strength, and K represents the total number of chained unbalanced blocks.

[0013] In one possible implementation, the step of calling the whole-genome selection model to predict the phenotypic data and obtaining the genomic selection result of the biological individual includes: The objective optimization function in the whole-genome selection model is solved iteratively to obtain the effect parameters of single nucleotide polymorphism sites; Based on the effect parameters, the phenotypic data are predicted to obtain the genomic selection results of the biological individual.

[0014] The present invention also provides a whole-genome selection device with block adaptive graph regularization, comprising: The data acquisition module is used to acquire the genotype data and corresponding phenotypic data of biological individuals, and determine the corresponding genotype matrix based on the encoding of gene information in the genotype data; The graph construction module is used to construct a linkage disequilibrium graph based on each single nucleotide polymorphism site in the genotype data, and to divide the single nucleotide polymorphism sites in the linkage disequilibrium graph into blocks to obtain linkage disequilibrium blocks. An adaptive regularization module is used to determine regularization constraint weights based on the block size of the linkage imbalance block, and to construct a genome-wide selection model based on the regularization constraint weights, the linkage imbalance block, and the genotype matrix. The regularization constraint weights are negatively correlated with the block size, and the regularization constraint weights are used to adjust the regularization parameters of the linkage imbalance block during genome-wide selection model prediction. The prediction output module is used to call the whole genome selection model to predict the phenotypic data and obtain the genome selection result of the biological individual.

[0015] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a program; the processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps of the above-described block adaptive graph regularization whole genome selection method.

[0016] The present invention also provides a whole-genome selection system, including a gene sequencer, a phenotype device, the aforementioned electronic device, and a display device, wherein the gene sequencer and the phenotype device are both communicatively connected to the electronic device, and the display device is installed in the electronic device; The gene sequencer is used to collect genotype data of biological individuals and store it in the memory of the electronic device; The phenotyping device is used to collect phenotypic data of biological individuals and store it in the memory of the electronic device; The display device is used to visualize the genomic selection results of the obtained biological individual after the processor of the electronic device executes the program stored in the memory.

[0017] The beneficial effects of the above implementation are as follows: The genome-wide selection method and apparatus for block-adaptive graph regularization provided by this invention divides single nucleotide polymorphism sites into blocks based on the construction of a linkage disequilibrium graph. During block modeling, instead of applying a uniform regularization strength to linkage disequilibrium blocks, the regularization parameters of the blocks are adaptively adjusted using the negative correlation between the regularization constraint weights and the block scale. Therefore, at large-scale blocks, since the graph Laplacian regularization term naturally has a stronger smoothing constraint capability, this invention compensates for the reduction of the regularization parameters of large-scale blocks by using relatively small regularization constraint weights to balance the constraint contributions between blocks of different scales, thereby avoiding the weakening of local genetic signal expression due to excessive smoothing in large-scale blocks. At small-scale blocks, the regularization parameters can be adjusted by relatively large regularization constraint weights to appropriately weaken the regularization constraint, overcoming the defect of unstable model effect estimation caused by insufficient regularization. By adaptively adjusting the regularization parameters of linkage disequilibrium blocks, an adaptive balance of constraint strength is achieved at different linkage disequilibrium block scales, enabling subsequent genome-wide selection models to more fully estimate the effects of single nucleotide polymorphism sites and improve the accuracy of genome prediction. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A schematic flowchart of the whole genome selection method with block adaptive graph regularization provided by the present invention; Figure 2 A schematic diagram illustrating the division of chained unbalanced blocks provided by the present invention; Figure 3 A schematic diagram of the whole genome selection device for block adaptive graph regularization provided by the present invention; Figure 4 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] In the description of the embodiments of this application, unless otherwise stated, "a plurality of" means two or more.

[0022] In the embodiments of the present invention, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, apparatus, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product or device.

[0023] The naming or numbering of steps in the embodiments of the present invention does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering. The execution order of the named or numbered process steps can be changed according to the technical purpose to be achieved, as long as the same or similar technical effect can be achieved.

[0024] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0025] The block-adaptive graph regularized whole-genome selection method provided in this invention can be applied to various genome selection scenarios for plant and animal organisms. The executing entity can be various servers, terminals, or remote cloud devices. During genome selection, the genotype data and corresponding phenotypic data of the organism are uploaded. Then, the block-adaptive graph regularized whole-genome selection method provided in this invention is called to process the uploaded data, sequentially constructing a linkage disequilibrium graph, partitioning blocks, adjusting regularization parameters, and constructing a whole-genome selection model. Finally, the whole-genome selection model is used to predict the phenotypic data to obtain the genome selection result of the organism. This selection result can be used for phenotypic evaluation, breeding decisions, or other genome prediction application scenarios.

[0026] The following describes in detail the whole genome selection method with block adaptive graph regularization provided by this invention.

[0027] Figure 1 The flowchart of the block adaptive graph regularization whole-genome selection method provided by the present invention is shown below. Figure 1 As shown, the block adaptive graph regularized whole genome selection method can be implemented through the following steps 101 to 104, which are explained in detail below.

[0028] Step 101: Obtain the genotype data and corresponding phenotypic data of the biological individual, and determine the corresponding genotype matrix based on the encoding of gene information in the genotype data.

[0029] Here, genotype data includes genotype information for multiple single nucleotide polymorphism (SNP) sites, while phenotypic data consists of measurements of the target trait corresponding to the individual organism. Genotype information is generally the encoding of gene information, i.e., the encoding of single nucleotide polymorphism sites, and requires preprocessing, including but not limited to missing value handling, outlier filtering, and standardization, to obtain a standardized genotype matrix. Preprocessing can reduce the impact of dimensional differences among different single nucleotide polymorphism sites on the subsequent modeling process and improve the stability of the model solution.

[0030] Step 102: Construct a linkage disequilibrium map based on each single nucleotide polymorphism site in the genotype data, and divide the single nucleotide polymorphism sites in the linkage disequilibrium map into blocks to obtain linkage disequilibrium blocks.

[0031] For genotype data, linkage disequilibrium graphs can be constructed using the linkage disequilibrium relationships between various single nucleotide polymorphism sites, which can then be used for subsequent linkage disequilibrium block modeling.

[0032] In one possible implementation, the chain imbalance graph is constructed as follows: Single nucleotide polymorphism sites in genotype data are used as nodes in the linkage disequilibrium graph; Determine the imbalance coefficient between nodes; When the imbalance coefficient is greater than a preset threshold, an edge is established between the nodes; Based on nodes and edges, a chain-disequilibrium graph is constructed.

[0033] Specifically, when constructing a linkage disequilibrium graph, single nucleotide polymorphism (SNP) sites in the genotype data are directly used as nodes in the graph. Then, the disequilibrium coefficient between nodes is calculated, which determines the linkage disequilibrium relationship between each SNP site. The disequilibrium coefficient can be a measure of the degree of uncorrelation or dissimilarity between nodes, i.e., the difference between 1 and a correlation measure or a similarity measure.

[0034] Next, edges are established based on the chain imbalance relationship. When the imbalance coefficient is greater than a preset threshold, it indicates that the chain imbalance relationship between nodes meets the requirements, and an edge is established between those nodes; otherwise, no edge is established. After determining the edges between all nodes, the nodes are connected by the edges to form a graph structure, which is the chain imbalance graph.

[0035] In this embodiment of the invention, a linkage disequilibrium graph is constructed by each single nucleotide polymorphism site, which not only improves the ability to characterize genomic structural information, but also facilitates subsequent block modeling.

[0036] Furthermore, the single nucleotide polymorphism sites in the linkage disequilibrium diagram were divided into blocks to obtain multiple linkage disequilibrium blocks.

[0037] In one possible implementation, the single nucleotide polymorphism sites in the linkage disequilibrium graph are divided into blocks to obtain linkage disequilibrium blocks. This can be achieved in the following ways, which are explained in detail below.

[0038] Based on the physical location of single nucleotide polymorphism (SNP) sites on chromosomes in the linkage disequilibrium graph, nodes in the linkage disequilibrium graph are divided to obtain linkage disequilibrium blocks; or based on pre-labeled linkage disequilibrium block information in a bioinformatics database, SNP sites in the linkage disequilibrium graph are divided to obtain linkage disequilibrium blocks; or clustering models are used to perform cluster analysis on SNP sites in the linkage disequilibrium graph to obtain linkage disequilibrium blocks.

[0039] Here, there are three ways to divide the blocks in the chain imbalance graph, which will be explained below.

[0040] The first method is based on chromosome location. This method uses the physical location of single nucleotide polymorphism (SNP) sites on chromosomes to divide them. Based on the continuity of the physical location of SNP sites on chromosomes, SNP sites that are physically adjacent and have a strong linkage disequilibrium relationship are divided into the same block. In other words, SNP sites corresponding to nodes that are physically close to each other on chromosomes and have edges connected in the linkage disequilibrium graph are divided into the same block.

[0041] The second method utilizes bioinformatics databases of individual organisms for block partitioning. These databases provide annotation information for linkage disequilibrium blocks. This annotation information is pre-annotated and can be directly retrieved from the bioinformatics database. Based on this annotation information, single nucleotide polymorphism sites are directly partitioned to determine the corresponding linkage disequilibrium blocks.

[0042] The third approach uses clustering algorithms for block partitioning. This method involves performing community detection or graph clustering analysis on the single nucleotide polymorphism sites within the linkage disequilibrium graph. The appropriate clustering algorithm automatically identifies tightly connected subgraphs as linkage disequilibrium blocks. Clustering algorithms can include, but are not limited to, spectral clustering algorithms, community detection algorithms based on modularity optimization, or other graph partitioning methods.

[0043] like Figure 2 As shown, after dividing the linkage disequilibrium graph constructed from the genotype data of a certain biological individual into blocks, three linkage disequilibrium blocks are obtained, namely block A, block B, and block C. It can be seen that the single nucleotide polymorphism sites in each linkage disequilibrium block are connected by edge relationships through linkage disequilibrium relations, but the number of single nucleotide polymorphism sites in each block is different, that is, the block size is different, and the block scale is different.

[0044] In this embodiment of the invention, by dividing the single nucleotide polymorphism sites in the linkage disequilibrium graph into blocks, multiple linkage disequilibrium blocks with structural differences are obtained. This allows for targeted application of regularization strength to the linkage disequilibrium blocks, achieving adaptive regularization adjustment.

[0045] Step 103: Determine the regularization constraint weights based on the block size of the linkage imbalance blocks, and construct a whole-genome selection model based on the regularization constraint weights, linkage imbalance blocks, and genotype matrix.

[0046] After identifying linkage disequilibrium blocks through block partitioning, this embodiment of the invention constructs block-adaptive graph regularization constraints based on block scale. Specifically, the regularization constraint weights are determined based on the block size information of the linkage disequilibrium blocks, and are used to apply differentiated regularization strengths to the regularization parameters of different linkage disequilibrium blocks, so that different linkage disequilibrium blocks are subject to different degrees of regularization constraints in the whole-genome selection model. As the scale of linkage disequilibrium blocks increases, large-scale blocks naturally have stronger smoothing constraint capabilities because the spectral range, trace, or eigenvalue distribution of the corresponding graph Laplacian matrix shows an expanding trend. Therefore, this invention compensates for the regularization parameters corresponding to large-scale blocks to balance the graph Laplacian regularization contribution at different block scales, thereby avoiding the weakening of local genetic signal expression capabilities due to excessive smoothing in large-scale blocks. Correspondingly, for small-scale blocks, an appropriate structural constraint strength is maintained to improve the stability of effect parameter estimation and model generalization ability.

[0047] Regularization constraint weights are used to adjust the regularization parameters of linkage disequilibrium blocks during genome-wide selection model prediction. Specifically, regularization constraint weights determined by different block scales are used to weight the regularization parameters of the corresponding blocks to achieve adaptive adjustment of the regularization parameters.

[0048] In one possible implementation, based on the adjustment process of the regularization constraint weights, it can be achieved in the following way, which is explained in detail below.

[0049] The regularization parameters of linkage imbalance blocks in whole-genome selection model prediction are weighted based on the regularization constraint weights to obtain the regularization parameters after linkage imbalance block adjustment. The weighting process for regularization parameters is expressed by the following formula: (1) in, This represents the regularization parameter used in genome-wide selection model predictions for linkage-disequilibrium blocks; specifically, it represents the block-level regularization parameter in the graph Laplacian regularization term corresponding to the linkage-disequilibrium block. This represents the regularization parameter after adjusting the k-th chain imbalance block. Indicates the weight of the regularization constraint. This represents the number of nodes in the k-th linkage disequilibrium block, and is a block size parameter, i.e., the total number of single nucleotide polymorphism sites contained in the linkage disequilibrium block, used to represent the block size of the linkage disequilibrium block. In specific calculations... It can also be defined as a combination function of the number of nodes and edge weights in a chained unbalanced block, the sum of node degrees, the weighted number of nodes, or other statistics that can reflect the structural scale characteristics of the block.

[0050] Regularization constraint weights It is a power function in form, where, Let be a preset positive real number, representing the scaling factor, used to control the rate at which the block size affects the regularization strength, thereby achieving differentiated constraints at different block sizes. This explains why, when the scaling factor... When it is a positive real number, it changes with the block size. Increase the weight of the regularization constraint. This is correspondingly reduced, thereby compensating for the graph regularization intensity of large-scale blocks and mitigating the over-smoothing effect caused by block size expansion. Accordingly, when the block size... When the regularization constraint weights are small, The relative size is increased, thereby maintaining appropriate structural constraints on small-scale blocks to improve the model's ability to stably express local genetic signals.

[0051] In this embodiment of the invention, regularization constraint weights are constructed using a power function of block size to adaptively adjust the regularization parameters. This theoretically balances the Laplace regularization contribution among chained imbalanced blocks of different scales, eliminating constraint inhomogeneity induced by differences in block size and maintaining a relative balance in constraint strength across the overall model. Furthermore, the adaptive adjustment only uses a power function to adjust the block-level regularization parameters in the graph Laplace regularization term at the scale level, without altering the topological relationships of the chained imbalanced graph structure itself. This achieves balanced constraints across different block scales while preserving the original structural dependency information.

[0052] Furthermore, in this embodiment of the invention, a whole-genome selection model is constructed based on regularization constraint weights, linkage imbalance blocks, and genotype matrix to perform genome selection.

[0053] In one possible implementation, a genome-wide selection model is constructed based on regularization constraint weights, linkage imbalance blocks, and the genotype matrix, including: Based on the regularization parameters, regularization constraint weights, effect parameters of single nucleotide polymorphism sites, genotype matrix, and phenotypic vectors corresponding to phenotypic data of linkage disequilibrium blocks in the whole-genome selection model prediction, an objective optimization function is constructed. Obtain a pre-defined graph regularized regression model, and determine the objective optimization function as the optimization objective of the graph regularized regression model to obtain a genome-wide selection model.

[0054] Here, the whole-genome selection model can be implemented using a graph-regularized regression model. The optimization objective of this model can be determined by the regularization parameters of linkage disequilibrium blocks in the whole-genome selection model prediction, the effect parameters of single nucleotide polymorphism sites, the genotype matrix, and the phenotypic vectors corresponding to the phenotypic data. A gene loss term can be constructed based on the genotype matrix, phenotypic vectors, and effect parameters. An L2 regularization term can be constructed based on the effect parameters. Furthermore, a graph regularization term can be constructed by combining the regularization parameters of linkage disequilibrium blocks in the whole-genome selection model prediction with the graph Laplacian matrix corresponding to the linkage disequilibrium blocks and the regularization constraint weights. Summing these three terms yields an objective optimization function.

[0055] By defining the objective optimization function as the optimization function of the graph regularized regression model, a genome-wide selection model can be obtained.

[0056] In one possible implementation, the objective function is expressed as follows: (2) in, Represents the genotype matrix. This represents the phenotypic vector corresponding to the phenotypic data. A vector representing the effector parameters of a single nucleotide polymorphism site. This represents the effect parameter vector of the single nucleotide polymorphism site in the k-th linkage disequilibrium block. This represents the regularization parameter after adjusting the k-th chain imbalance block. The graph Laplace matrix corresponding to the k-th chained unbalanced block. This indicates the calculation of the L2 norm. The hyperparameter represents the regularization strength, and K represents the total number of chained unbalanced blocks.

[0057] In the above formula (2), This is the gene loss term, representing the least squares error, used to measure the model predictions of the whole-genome selection model. Compared with the true value The difference ensures that the model has a good fit to the training data. By minimizing this least squares error, the whole genome selection model can fit the genotype-phenotype associations in the gene data as accurately as possible, thereby improving the accuracy of the model's predictions. The L2 regularization term, equivalent to the ridge regression part of a graph regularized regression model, functions by limiting the coefficients... The size of the L2 regularization reduces model complexity. When there is a high correlation between features, L2 regularization can make the solution of the whole genome selection model more stable. That is, the graph regularization term. Specifically, the regularization parameter is obtained by adjusting the regularization constraint weights of linkage disequilibrium blocks in the genome-wide selection model prediction. The purpose of the graph regularization term is to make a smoothness assumption, which is the graph regularization assumption. Specifically, if two nodes are connected in the graph (i.e., highly similar), then their effect values ​​on the phenotype should be similar, meaning the output of the genome-wide selection model for them should be similar. The graph regularization term penalizes the difference in coefficients between connected nodes. That is, if the effect difference between two adjacent single nucleotide polymorphism sites in a linkage disequilibrium block is large, it will be penalized, so that the genome-wide selection model can consider the connection relationships between nodes more when solving the problem.

[0058] In this embodiment of the invention, an objective optimization function is constructed by simultaneously calculating gene loss terms, L2 regularization terms, and graph regularization terms to complete the modeling of genome prediction. This can effectively improve the prediction accuracy, prediction stability, and model generalization ability of the whole genome selection model.

[0059] Step 104: Use the whole genome selection model to predict the phenotypic data and obtain the genome selection results of the organism.

[0060] The prediction process of the whole genome selection model involves iteratively solving its objective optimization function to determine the effect parameters of single nucleotide polymorphism sites, which are then used to predict the phenotypic data of an organism and complete the genome selection.

[0061] In one possible implementation, a genome-wide selection model is invoked to predict phenotypic data, yielding the genome selection results for the individual organism, including: The objective optimization function in the genome-wide selection model is solved iteratively to obtain the effect parameters of single nucleotide polymorphism sites; Based on effect parameters, phenotypic data are predicted to obtain the genomic selection results of biological individuals.

[0062] Here, the prediction process of the whole-genome selection model is achieved by iteratively solving the objective optimization function using an iterative algorithm. After inputting known quantities from the objective optimization function—such as the regularization parameter of the linkage disequilibrium blocks determined based on the biological individual, the genotype matrix, and the phenotypic vectors corresponding to the phenotypic data—into the whole-genome selection model, the model can then perform iterative solving. The main objective of the model's iterative solution is to solve for the effect parameter vector of the single nucleotide polymorphism sites in the above formula (2). The unknowns in the objective function are the effect parameter vectors for each single nucleotide polymorphism (SNP) site, used to determine the effect parameters for each SNP site. Since the objective function is convex and differentiable, iterative solutions can be obtained by taking derivatives to obtain closed-form solutions, or by using iterative optimization algorithms such as batch gradient descent and coordinate descent.

[0063] Furthermore, the effect parameters of single nucleotide polymorphism sites obtained through iterative solutions using a genome-wide selection model are used to predict phenotypic data, yielding the genomic selection results for the organism. Specifically, this prediction process combines the genotype information of the organism with effect parameters to calculate, for example, the genome-wide breeding value (GEBV) or predicted phenotypic value, as the genomic selection result.

[0064] Effect parameters represent the magnitude and direction (positive or negative) of the contribution of single nucleotide polymorphisms (SNPs) to target traits (such as plant yield, disease resistance, height, etc.). Since SNPs are typically encoded numerically, such as 0, 1, and 2, representing reference homozygote, heterozygote, and alternative homozygote respectively, in specific prediction calculations, the genotype vector corresponding to the SNP (i.e., the vector constructed from the encoded values) is linearly combined with the vector corresponding to the effect parameter—that is, a product operation. The sum of the product results for each SNP is then obtained, representing the genomic selection result for that individual organism, representing the estimated breeding value or predicted phenotypic value.

[0065] In this embodiment of the invention, the genomic selection results of an organism are predicted using effect parameters obtained through iterative optimization in a genome-wide selection model, which can more accurately capture the genetic differences among organisms. Simultaneously, the effect parameters determined by the genome-wide selection model achieve unbiased estimation of the effects of single nucleotide polymorphism sites, avoiding the omission of genetic information and ensuring the accuracy of genomic selection.

[0066] The following example illustrates the genome-wide selection method with block adaptive graph regularization provided by this invention. Taking ducks as an example, the genotype data of the duck population and three traits—body length, shank length, and keel length—are obtained as phenotypes. The genotype data includes multiple single nucleotide polymorphism sites on the duck chromosome, and the phenotypic data are continuous quantitative traits.

[0067] In genome selection, a linkage disequilibrium graph is constructed based on single nucleotide polymorphisms (SNPs) in the genotype data. These SNPs are then divided into blocks to obtain linkage disequilibrium blocks. Regularization constraint weights are determined based on the block size of these linkage disequilibrium blocks, and the regularization parameters for predicting linkage disequilibrium blocks in the whole-genome selection model are adjusted according to these weights. The block size parameter... Explicitly selected as the first The number of single nucleotide polymorphism sites contained within a linkage disequilibrium block.

[0068] Finally, a genome-wide selection model was constructed based on linkage disequilibrium blocks, regularization constraint weights, and the genotype matrix to predict phenotypic data and obtain the genome selection results for the duck population. To verify the effectiveness of the block-adaptive graph regularized genome-wide selection method of this invention, comparative experiments were conducted using traditional genome-wide selection methods such as GBLUP, rrBLUP, BayesC, and BayesianLasso. All methods were run under the same training and test set partitioning conditions to ensure fairness in the comparison. The prediction correlation coefficients (ten-fold average) for different shapes in the gene selection results are shown in Table 1 below.

[0069] Table 1

[0070] As shown in Table 1, the prediction correlation coefficients of the method of this invention are superior to those of traditional genome-wide selection methods for all three different traits. In the prediction of the three traits, the traditional methods all exhibited low prediction performance and large fluctuations in results, while the method of this invention maintained a stable positive correlation prediction ability for all three traits.

[0071] The data in Table 1 further validates the adaptive adjustment of the regularization parameter in the method of this invention. This adjustment eliminates the constraint inhomogeneity caused by the Laplacian spectrum expansion of large blocks, suppresses its excessive smoothing effect, and allows blocks of different scales to obtain a relatively balanced penalty intensity in the overall optimization model, thereby improving the prediction accuracy of the model under different traits. Based on the results for the three traits, it can be seen that the method of this invention exhibits significant advantages in both prediction accuracy and stability. These experimental results demonstrate that the block-scale adaptive adjustment mechanism has good robustness under high-dimensional, small-sample conditions and can achieve stable performance improvement under different trait types and data scales of biological individuals, reflecting good generalization ability and practical application value.

[0072] In summary, this invention divides single nucleotide polymorphism sites into blocks based on the constructed linkage disequilibrium graph. However, during block modeling, instead of applying a uniform regularization strength to the linkage disequilibrium blocks, it adaptively adjusts the regularization parameters of the blocks using the negative correlation between the regularization constraint weights and the block size. Thus, in large-scale blocks, since the graph Laplacian regularization term naturally exhibits a stronger smoothing constraint effect as the block size increases, this invention compensates for this by reducing the regularization parameters in large-scale blocks with relatively small regularization constraint weights. This reduces the excessively strong smoothing constraint trend in large-scale blocks, preventing local genetic signals from being over-smoothed and weakening their resolving power. In small-scale blocks, the regularization parameters can be adjusted using relatively large regularization constraint weights to appropriately weaken the regularization constraint, overcoming the instability in model effect estimation caused by insufficient regularization. By adaptively adjusting the regularization parameters of linkage disequilibrium blocks, an adaptive balance of constraint strength is achieved at different linkage disequilibrium block scales, enabling subsequent genome-wide selection models to more fully estimate the effects of single nucleotide polymorphism sites and improve the accuracy of genome prediction.

[0073] The following section details the whole genome selection device with block adaptive graph regularization provided by this invention.

[0074] Figure 3 This is a schematic diagram of the structure of the whole genome selection device with block adaptive graph regularization provided by the present invention, as shown below. Figure 3 As shown, the whole genome selection device with block adaptive graph regularization specifically includes: a data acquisition module 301, a graph construction module 302, an adaptive regularization module 303, and a prediction output module 304.

[0075] Specifically, the data acquisition module 301 is used to acquire the genotype data and corresponding phenotypic data of an individual organism, and determine the corresponding genotype matrix based on the encoding of gene information in the genotype data; the graph construction module 302 is used to construct a linkage disequilibrium graph based on each single nucleotide polymorphism site in the genotype data, and divide the single nucleotide polymorphism sites in the linkage disequilibrium graph into blocks to obtain linkage disequilibrium blocks; the adaptive regularization module 303 is used to determine the regularization constraint weights based on the block size of the linkage disequilibrium blocks, and construct a whole-genome selection model according to the regularization constraint weights, the linkage disequilibrium blocks, and the genotype matrix, wherein the regularization constraint weights are adaptively determined based on the block size of the linkage disequilibrium blocks, and the regularization constraint weights are used to adjust the regularization parameters of the linkage disequilibrium blocks during whole-genome selection model prediction to balance the graph regularization contribution at different block sizes; the prediction output module 304 is used to call the whole-genome selection model to predict the phenotypic data to obtain the genome selection result of the individual organism.

[0076] In a specific genome selection scenario, by uploading the genotype data and corresponding phenotypic data of an individual organism, the data acquisition module 301 of the block-adaptive graph regularization whole-genome selection device acquires this data. Then, the linkage disequilibrium graph is constructed and divided into blocks sequentially through the graph construction module 302 and the adaptive regularization module 303. Regularization constraint weights are determined, and a whole-genome selection model is constructed by combining the genotype matrix. Finally, the prediction output module 304 calls the whole-genome selection model to predict the phenotypic data, obtaining the genome selection result of the individual organism. This selection result can be used for individual trait assessment, breeding decisions, or to execute decisions in other genome prediction application scenarios.

[0077] The block adaptive graph regularization whole genome selection device provided in the above embodiments can realize the technical solutions described in the above block adaptive graph regularization whole genome selection method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above block adaptive graph regularization whole genome selection method embodiments, and their technical effects can also be referred to each other, which will not be repeated here.

[0078] like Figure 4 As shown, the present invention also provides an electronic device 400. The electronic device 400 includes a processor 401, a memory 402, and a display 403. Figure 4 Only some components of the electronic device 400 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0079] In some embodiments, memory 402 may be an internal storage unit of electronic device 400, such as a hard disk or memory of electronic device 400. In other embodiments, memory 402 may also be an external storage device of electronic device 400, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 400.

[0080] Furthermore, the memory 402 may include both internal storage units of the electronic device 400 and external storage devices. The memory 402 is used to store application software installed on the electronic device 400 and various types of data, such as genotype data and corresponding phenotypic data of biological individuals.

[0081] In some embodiments, processor 401 may be a central processing unit (CPU), microprocessor, or other data processing chip for running program code stored in memory 402 or processing data, such as the block adaptive graph regularization whole genome selection method of the present invention.

[0082] In some embodiments, display 403 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 403 is used to display information from electronic device 400 and to display a visual user interface that can show the genomic selection results of an individual organism. Components 401-403 of electronic device 400 communicate with each other via a system bus.

[0083] In some embodiments of the present invention, when the processor 401 executes the computer program in the memory 402, the following steps can be implemented: acquiring the genotype data and corresponding phenotypic data of an individual organism, and determining the corresponding genotype matrix based on the encoding of gene information in the genotype data; constructing a linkage disequilibrium graph based on each single nucleotide polymorphism site in the genotype data, and dividing the single nucleotide polymorphism sites in the linkage disequilibrium graph into blocks to obtain linkage disequilibrium blocks; determining regularization constraint weights based on the block size of the linkage disequilibrium blocks, and constructing a whole-genome selection model according to the regularization constraint weights, the linkage disequilibrium blocks, and the genotype matrix, wherein the regularization constraint weights are negatively correlated with the block size, and the regularization constraint weights are used to adjust the regularization parameters of the linkage disequilibrium blocks during whole-genome selection model prediction; calling the whole-genome selection model to predict the phenotypic data to obtain the genome selection result of the individual organism.

[0084] It should be understood that when the processor 401 executes the computer program in the memory 402, in addition to the functions described above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.

[0085] Furthermore, the embodiments of the present invention do not specifically limit the type of electronic device 400 mentioned. Electronic device 400 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, electronic device 400 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0086] The present invention also provides a whole-genome selection system, including a gene sequencer, a phenotyping device, the electronic device 400 described in the above embodiments, and a display device. The gene sequencer and the phenotyping device are both communicatively connected to the electronic device, and the display device is housed within the electronic device.

[0087] A gene sequencer is used to collect genotype data of an individual organism and store it in the memory 402 of an electronic device 400. The gene sequencer is the core device for collecting genotype data of an individual organism; it can directly determine single nucleotide polymorphism sites on the chromosomes of an individual organism and obtain the coding of the site sequence as genotype data to provide high-precision genetic information.

[0088] Phenotyping devices are used to collect phenotypic data of biological individuals and store it in the memory 402 of electronic device 400. The type of phenotyping device can be selected according to the actual biological individual. For plant individuals, phenotyping devices can include plant precision identification platforms, crop phenotyping instruments, hyperspectral imaging systems, lidar systems, or drone-borne phenotyping systems. For animal individuals, phenotyping devices can include high-field magnetic resonance imaging scanners, three-dimensional body analyzers, wearable physiological monitoring systems, etc.

[0089] A display device is used to visualize the genomic selection results of a biological individual after the processor 401 of the electronic device 400 executes a program stored in the memory 402.

[0090] The display device can be an external device installed on the electronic device 400 or a display device that communicates remotely with the electronic device 400. It is similar to the display 403 in the electronic device 400, and in some embodiments can be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display device serves as the output of the whole-genome selection system, visualizing the genomic selection results of an individual organism and feeding them back to the user. This allows the user to perform trait assessments, breeding decisions, or other genomic prediction application decisions based on the visualized genomic selection results.

[0091] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a whole-genome selection method with block adaptive graph regularization provided by the methods described above. This method includes: acquiring genotype data and corresponding phenotypic data of an individual organism, and determining a corresponding genotype matrix based on the encoding of gene information in the genotype data; constructing a linkage disequilibrium graph based on each single nucleotide polymorphism (SNP) site in the genotype data, and dividing the SNP sites in the linkage disequilibrium graph into blocks to obtain linkage disequilibrium blocks; determining regularization constraint weights based on the block size of the linkage disequilibrium blocks, and constructing a whole-genome selection model based on the regularization constraint weights, the linkage disequilibrium blocks, and the genotype matrix, wherein the regularization constraint weights are negatively correlated with the block size, and the regularization constraint weights are used to adjust the regularization parameters of the linkage disequilibrium blocks during whole-genome selection model prediction; and calling the whole-genome selection model to predict the phenotypic data to obtain the genome selection result of the individual organism.

[0092] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0093] The above provides a detailed description of the block adaptive graph regularization whole-genome selection method and apparatus provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A block-adaptive graph regularized whole-genome selection method, characterized in that, include: Acquire genotype data and corresponding phenotypic data of biological individuals, and determine the corresponding genotype matrix based on the encoding of gene information in the genotype data; Based on each single nucleotide polymorphism site in the genotype data, a linkage disequilibrium graph is constructed, and the single nucleotide polymorphism sites in the linkage disequilibrium graph are divided into blocks to obtain linkage disequilibrium blocks. Regularization constraint weights are determined based on the block size of the linkage imbalance block, and a whole-genome selection model is constructed based on the regularization constraint weights, the linkage imbalance block, and the genotype matrix. The regularization constraint weights are negatively correlated with the block size, and the regularization constraint weights are used to adjust the regularization parameters of the linkage imbalance block during whole-genome selection model prediction. The genome-wide selection model is invoked to predict the phenotypic data, thereby obtaining the genome selection results of the biological individual.

2. The whole-genome selection method with block adaptive graph regularization according to claim 1, characterized in that, The chain imbalance diagram is constructed in the following way: The single nucleotide polymorphism sites in the genotype data are used as nodes in the linkage disequilibrium graph; Determine the imbalance coefficient between the nodes; When the imbalance coefficient is greater than a preset threshold, an edge is established between the nodes; Based on the nodes and edges, a chain imbalance graph is constructed.

3. The whole-genome selection method with block adaptive graph regularization according to claim 1, characterized in that, The process of dividing single nucleotide polymorphism sites in the linkage disequilibrium graph into blocks to obtain linkage disequilibrium blocks includes: Based on the physical location of the single nucleotide polymorphism sites on the chromosome in the linkage disequilibrium graph, the nodes in the linkage disequilibrium graph are divided to obtain linkage disequilibrium blocks; Alternatively, based on the linkage disequilibrium block information pre-labeled in the bioinformatics database of the biological individual, the single nucleotide polymorphism sites in the linkage disequilibrium graph are divided to obtain linkage disequilibrium blocks; Alternatively, a clustering model can be used to perform cluster analysis on the single nucleotide polymorphism sites in the linkage disequilibrium diagram to obtain linkage disequilibrium blocks.

4. The whole-genome selection method with block adaptive graph regularization according to claim 1, characterized in that, The adjustment process of the regularization constraint weights includes: The regularization parameters of the linkage imbalance block in the whole genome selection model prediction are weighted based on the regularization constraint weights to obtain the regularization parameters of the linkage imbalance block after adjustment. The weighting process of the regularization parameters is expressed by the following formula: in, This represents the regularization parameter used in genome-wide selection model prediction of the linkage disequilibrium blocks. This represents the regularization parameter after adjustment for the k-th chain imbalance block. This represents the weight of the regularization constraint. This represents the number of nodes in the k-th chain of unbalanced blocks. This is a preset positive real number, representing the scaling adjustment coefficient.

5. The whole-genome selection method with block adaptive graph regularization according to claim 1, characterized in that, The step of constructing a genome-wide selection model based on the regularization constraint weights, the linkage imbalance blocks, and the genotype matrix includes: Based on the regularization parameters of linkage disequilibrium blocks in the whole-genome selection model prediction, the regularization constraint weights, the effect parameters of single nucleotide polymorphism sites, the genotype matrix, and the phenotypic vectors corresponding to the phenotypic data, a target optimization function is constructed. A preset graph regularized regression model is obtained, and the objective optimization function is determined as the optimization function of the graph regularized regression model to obtain a genome-wide selection model.

6. The whole-genome selection method with block adaptive graph regularization according to claim 5, characterized in that, The objective optimization function is expressed as the following formula: in, This represents the genotype matrix. This represents the phenotypic vector corresponding to the phenotypic data. A vector representing the effector parameters of a single nucleotide polymorphism site. This represents the effect parameter vector of the single nucleotide polymorphism site in the k-th linkage disequilibrium block. This represents the regularization parameter after adjusting the k-th chain imbalance block. This represents the graph Laplacian matrix corresponding to the k-th chain imbalance block. This indicates the calculation of the L2 norm. The hyperparameter represents the regularization strength, and K represents the total number of chained unbalanced blocks.

7. The whole-genome selection method with block adaptive graph regularization according to claim 1, characterized in that, The step of calling the whole-genome selection model to predict the phenotypic data and obtaining the genomic selection result of the biological individual includes: The objective optimization function in the whole-genome selection model is solved iteratively to obtain the effect parameters of single nucleotide polymorphism sites; Based on the effect parameters, the phenotypic data are predicted to obtain the genomic selection results of the biological individual.

8. A block-adaptive graph regularization whole-genome selection device, characterized in that, include: The data acquisition module is used to acquire the genotype data and corresponding phenotypic data of biological individuals, and determine the corresponding genotype matrix based on the encoding of gene information in the genotype data; The graph construction module is used to construct a linkage disequilibrium graph based on each single nucleotide polymorphism site in the genotype data, and to divide the single nucleotide polymorphism sites in the linkage disequilibrium graph into blocks to obtain linkage disequilibrium blocks. An adaptive regularization module is used to determine regularization constraint weights based on the block size of the linkage imbalance block, and to construct a genome-wide selection model based on the regularization constraint weights, the linkage imbalance block, and the genotype matrix. The regularization constraint weights are negatively correlated with the block size, and the regularization constraint weights are used to adjust the regularization parameters of the linkage imbalance block during genome-wide selection model prediction. The prediction output module is used to call the whole genome selection model to predict the phenotypic data and obtain the genome selection result of the biological individual.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is configured to execute the program stored in the memory to implement the steps of the block adaptive graph regularization whole-genome selection method as described in any one of claims 1 to 7.

10. A genome-wide selection system, characterized in that, The device includes a gene sequencer, a phenotyping device, an electronic device as described in claim 9, and a display device, wherein the gene sequencer and the phenotyping device are both communicatively connected to the electronic device, and the display device is installed in the electronic device; The gene sequencer is used to collect genotype data of biological individuals and store it in the memory of the electronic device; The phenotyping device is used to collect phenotypic data of biological individuals and store it in the memory of the electronic device; The display device is used to visualize the genomic selection results of the obtained biological individual after the processor of the electronic device executes the program stored in the memory.

Images