A method of cross breeding by means of virtual traits
By using virtual traits for parent selection and hybridization, the uncertainties and blind spots in the hybridization breeding process are resolved, the breeding process is shortened, the number of hybridization combinations is reduced, and the breeding efficiency is improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG ACADEMY OF AGRICULTURE SCIENCES
- Filing Date
- 2023-04-07
- Publication Date
- 2026-07-10
AI Technical Summary
The existing technology for hybridization breeding is characterized by uncertainty and blindness. Direct selection and hybridization combinations based on parental performance lead to uncertainty in the breeding process and an excessive number of combinations.
Using virtual traits for parental selection and hybridization, a basic trait database is constructed, and virtual traits are established using machine learning methods. Virtual traits that are highly correlated with practical traits and have low heterosis rates are selected as target combinations for breeding and hybridization.
This method enables the screening of heterosis, avoids the uncertainty and blindness of parental performance, shortens the breeding process, and reduces the number of crossbreeding combinations.
Smart Images
Figure CN116597899B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hybridization breeding, and more particularly to a hybridization breeding method utilizing virtual traits. Background Technology
[0002] Hybrid breeding involves sexual crossbreeding between two individuals with different genetic structures. Due to the independent assortment of genes from both parents, new genotypes can emerge in the offspring. Superior combinations can then be selected to cultivate new varieties. First-generation hybrids (F1) are not simply a combination of the desirable traits of both parents, but can produce traits superior to those of both parents. Therefore, F1 hybrids are one of the most widely used and effective breeding methods both domestically and internationally, and most new varieties currently in production are developed using this method.
[0003] Hybridization, breeding, and selection are three closely linked steps in the hybridization breeding process. Selection is a fundamental breeding technique, and hybridization combines different genetic materials (genes) to produce segregation in the offspring. However, to select new varieties that meet breeding objectives from the breeding materials that have segregated traits after hybridization, it is also necessary to appropriately apply selection techniques based on breeding under certain environmental conditions, and to select superior individuals in a timely and continuous manner according to the breeding objectives, so as to turn the breeding goals and plans into reality.
[0004] In hybridization breeding, we often only obtain the parental performance results, while the hybrids used in production are the actual hybrids. Due to heterosis, the performance of the parents and hybrids often differs, leading to uncertainty and uncertainty in the breeding process. However, because the mechanism of heterosis is not fully understood, breeding experts currently often have to rely solely on parental performance to select parents and create hybrid combinations. For example, a "method for selecting parents for super hybrid rice breeding" disclosed in Chinese patent literature (CN100364385A) uses an intermediate sterile rice line as the female parent and an intermediate restorer line as the male parent to produce F1 hybrids. The male and female parents corresponding to the superior traits of the F1 hybrids are then selected for hybrid rice breeding. This method is used for parental selection in super hybrid rice breeding, but directly relying on parental performance for selection and hybrid combination presents uncertainties and uncertainties. Summary of the Invention
[0005] This invention primarily addresses the uncertainty and blind spots inherent in existing technologies that directly rely on parental performance for breeding and hybridization. It provides a hybridization breeding method utilizing virtual traits, enabling parental selection and hybridization to identify superior target combinations. This method avoids the influence of heterosis during parental selection and hybridization, eliminating the uncertainty and blind spots associated with current methods that rely solely on parental performance. It accelerates the parental selection process, reduces the number of hybridization combinations, and speeds up the breeding process.
[0006] The above-mentioned technical problems of the present invention are mainly solved by the following technical solutions:
[0007] A hybridization breeding method utilizing virtual traits includes the following steps:
[0008] S1: Establish a basic database containing the hybrid generation and corresponding parental performance; construct a set of basic traits and establish a set of computational methods;
[0009] S2: Randomly select basic traits and operation methods from the set of basic traits and the set of operation methods, and combine them to construct candidate virtual traits;
[0010] S3: Using data from the basic database, calculate the parameters and test results for the practical traits and alternative virtual traits;
[0011] S4: Based on the comparison of the calculated parameters, test results, and corresponding thresholds, select the target virtual trait for breeding or crossbreeding.
[0012] This scheme utilizes a large number of hybrid combinations and their parent data from the breeding process to construct a basic trait database. Then, it employs feature construction methods from machine learning to establish virtual traits, identifying those with very low heterosis rates but high correlation with practical traits. These virtual traits are then used for parent selection and hybrid combination formulation to screen for superior target combinations. This scheme avoids the influence of heterosis on parent selection and hybrid combination formulation, avoiding the uncertainty and blind spots of current methods that directly rely on parent performance for selection and hybrid combination formulation. It accelerates the parent selection process, reduces the number of hybrid combinations, and speeds up the breeding process.
[0013] As a preferred option, taking the silkworm as an example, the basic traits in the aforementioned basic trait set are practical traits, including cocooning rate, cage mortality rate, total cocoon quantity, cocoon layer quantity, cocoon production per 10,000 silkworms, pupa rate, cocoon layer rate, and cocoon layer quantity per 10,000 silkworms. These are used to establish virtual traits for analysis.
[0014] Preferably, the set of computational methods includes addition, subtraction, multiplication, division, reciprocal, square root, square, cube, logarithm, and exp. These are used to establish virtual traits for analysis.
[0015] As a preferred option, alternative virtual traits are constructed by randomly selecting basic traits and computational methods.
[0016] Preferably, the parameters and test results of the practical traits and candidate virtual traits include the heterosis rate of the candidate virtual traits, the correlation coefficient between the parental values of the candidate virtual traits and the F1 hybrid performance of the practical traits, and the normalized mean square error of the regression model between the parental values of the candidate virtual traits and the F1 hybrid performance of the practical traits. Virtual traits with very low heterosis rates and high correlation with practical traits are identified. Parental selection and hybridization are then conducted using virtual traits to screen for superior target combinations.
[0017] Preferably, the heterosis rate VR of the candidate virtual traits is calculated as follows:
[0018] VR = (F1 - MP) / MP;
[0019] Among them, F1 represents the results of the first generation of hybrids with virtual traits;
[0020] MP represents the midparent value of a virtual trait.
[0021] Preferably, the correlation coefficient R between the parent values of the candidate virtual traits and the F1 hybrid performance of the practical traits is calculated as follows:
[0022]
[0023] Where, x i Let be the virtual trait value of the i-th sample;
[0024] y i Let be the practical trait value of the i-th sample;
[0025] This represents the virtual trait mean of the sample population;
[0026] This represents the average practical traits of the sample population.
[0027] As a preferred approach, using data from the basic database, a univariate regression model is established with the middle parent of the virtual trait as the independent variable and the corresponding hybrid F1 performance as the dependent variable. The normalized mean square error (NMSE) of the model is estimated using the leave-one-out cross-test method of the regression model.
[0028]
[0029] Among them, y i Let be the measured value of the i-th sample;
[0030] y i ^Let be the predicted value for the i-th sample;
[0031] This represents the population mean of the tested sample.
[0032] Preferably, when the heterosis rate of the candidate virtual trait is less than the first threshold, and the correlation coefficient between the parental value of the candidate virtual trait and the F1 hybrid performance of the practical trait is greater than the second threshold, and the normalized mean square error of the regression model between the parental value of the candidate virtual trait and the F1 hybrid performance of the practical trait is less than the third threshold, the corresponding candidate virtual trait is selected as the target virtual trait for breeding or crossbreeding; otherwise, the process returns to step S2 for recalculation.
[0033] This approach establishes virtual traits to identify those with low heterosis rates and high correlation with practical traits. It avoids the influence of heterosis on parent selection and hybridization, mitigating the uncertainty and blind spots of current methods that rely solely on parental performance for selection and hybridization. This accelerates parent selection, reduces the number of hybridization combinations, and speeds up the breeding process.
[0034] The beneficial effects of this invention are:
[0035] By utilizing virtual traits for parent selection and hybridization, superior target combinations can be screened out. This avoids the influence of heterosis on parent selection and hybridization, and avoids the uncertainty and blindness of current methods that rely directly on parent performance for selection and hybridization. It can accelerate the parent selection process, reduce the number of hybridization combinations, and speed up the breeding process. Attached Figure Description
[0036] Figure 1 This is a flowchart of the hybridization breeding method of the present invention.
[0037] Figure 2 This invention relates to a binary tree representing the virtual trait of cocoon production per 10,000 silkworms.
[0038] Figure 3 This invention relates to a binary tree of virtual traits of the number of cocoon layers produced by 10,000 silkworms.
[0039] Figure 4 This is a comparison chart of the measured values of the first-generation hybrid silkworm cocoon production per 10,000 silkworms, the predicted values of the mid-parent values of the virtual trait, and the predicted values of the mid-parent values of the practical trait.
[0040] Figure 5 This is a comparison chart of the measured values of the cocoon production of F1 hybrid silkworms and the predicted values of the virtual trait midparent values and the practical trait midparent values. Detailed Implementation
[0041] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.
[0042] Example:
[0043] This embodiment describes a hybridization breeding method utilizing virtual traits, such as... Figure 1 As shown, it includes the following steps:
[0044] S1: Establish a basic database containing hybrid generation 1 and corresponding parental performance.
[0045] This embodiment uses silkworms as experimental material and employs the hybrids from 7 historically collected hybridization experiments, totaling 326 hybridization combinations, along with the corresponding parental results from simultaneous rearing, as the basic database.
[0046] Construct a set of basic traits and establish a set of computational methods.
[0047] A candidate trait set is generated, called the basic trait set, which usually uses commonly used practical traits.
[0048] This embodiment includes cocooning rate, cage death rate, total cocoon quantity, cocoon layer quantity, cocoon production per 10,000 silkworms, pupa rate, cocoon layer rate, and cocoon layer quantity per 10,000 silkworms.
[0049] The set of arithmetic operations includes addition, subtraction, multiplication, division, reciprocal, square root, square, cube, logarithm, and exp.
[0050] S2: Randomly select basic traits and operation methods from the set of basic traits and the set of operation methods, and combine them to construct candidate virtual traits.
[0051] By randomly selecting basic traits and computational methods, and employing feature construction methods from machine learning, candidate virtual traits are constructed. In this embodiment, two important yield traits of silkworms are constructed: the number of cocoons produced per 10,000 silkworms and the number of cocoon layers produced per 10,000 silkworms.
[0052] like Figure 2 and Figure 3 As shown, structural binary trees were constructed to represent the virtual traits of cocoon production per 10,000 silkworms and cocoon layer production per 10,000 silkworms. The symbols for the binary tree species (i.e., the basic traits are as follows): z[,1] = cocooning rate; z[,2] = cage death rate; z[,3] = total cocoon production; z[,4] = cocoon layer production; z[,5] = cocoon production per 10,000 silkworms; z[,6] = pupa rate; z[,7] = cocoon layer rate; z[,8] = cocoon layer production per 10,000 silkworms.
[0053] like Figure 2 The mathematical expression for the binary tree of cocoon production per 10,000 silkworms is as follows:
[0054]
[0055] like Figure 3 The mathematical expression for the binary tree of cocoon production per 10,000 silkworms is as follows:
[0056]
[0057] S3: Using data from the basic database, calculate the parameters and test results for the practical traits and alternative virtual traits.
[0058] The parameters and test results of the practical traits and alternative virtual traits include the heterosis rate (VR) of the alternative virtual traits, the correlation coefficient (R) between the parental value of the alternative virtual traits and the F1 hybrid performance of the practical traits, and the normalized mean square error (NMSE) of the regression model between the parental value of the alternative virtual traits and the F1 hybrid performance of the practical traits.
[0059] As is well known, the calculation process for the heterosis rate (VR) of a typical (utility trait) is as follows:
[0060] VR = (F1 - MP) / MP;
[0061] Among them, F1 represents the results of hybrid generations;
[0062] MP is the midparent value of the two parents.
[0063] After the formula transformation, F1 = MP(1 + VR);
[0064] When VR is very small, close to 0, F1≈MP, so that the performance of the parents and the hybrid can be basically consistent. Therefore, if a trait with a small heterosis rate and a high correlation with the practical trait can be constructed, its F1 hybrid performance is almost equal to the mid-parent value of the two parents. Thus, the performance of the virtual traits of the two parents can be used to select parents and create hybrid combinations.
[0065] Using data from the basic database, a univariate regression model was established with the middle parent of the virtual trait as the independent variable and the corresponding hybrid F1 performance as the dependent variable. The normalized mean square error (NMSE) of the model was estimated using the leave-one-out cross-test method of the regression model.
[0066]
[0067] Among them, y i Let be the measured value of the i-th sample;
[0068] y i ^ Let be the predicted value for the i-th sample;
[0069] This represents the population mean of the tested sample.
[0070] As shown in Table 1, the average heterosis of the two practical traits, the average heterosis rate of the virtual traits, and the correlation coefficients between the mid-parent values of the virtual traits and the F1 generation performance of the corresponding practical traits are listed in columns 1 to 3 of Table 1. It can be seen from the table that the heterosis rates of these virtual traits are all very small, while the mid-parent values of the virtual traits also show a high correlation with the F1 generation performance of the corresponding practical traits.
[0071] Table 1. Parameters and test results of practical traits and their virtual traits.
[0072]
[0073] In this embodiment, 100 samples were randomly selected from the 326 experimental data points used as validation samples for the method. A univariate regression model was established with the median parent value of the virtual trait as the independent variable and the F1 generation score of the corresponding practical trait as the dependent variable. The method was validated using leave-one-out cross-validation. The validation index was normalized mean squared error (NMSE), and the results are listed in column 4 of Table 1. As can be seen from Table 1, the NMSE values for both traits are less than 0.3, and significantly less than the NMSE predicted using the median parent value of the practical trait (listed in column 5 of Table 1) (the smaller the NMSE value, the better the model fit).
[0074] Figure 4 , Figure 5 The figure compares the predicted and measured values of two traits: the number of cocoons produced per 10,000 silkworms and the number of cocoons produced per 10,000 silkworms. As can be seen from the figure, the predicted and measured values of the two traits fit well, which is significantly better than the method commonly used by breeders based on parental performance (mid-parental value).
[0075] S4: Based on the comparison of the calculated parameters, test results, and corresponding thresholds, select the target virtual trait for breeding or crossbreeding.
[0076] When the heterosis rate of the candidate virtual trait is less than the first threshold, and the correlation coefficient between the parental value and the performance of the first generation of hybrids of the candidate virtual trait is greater than the second threshold, and the normalized mean square error of the regression model between the parental value and the performance of the first generation of hybrids of the candidate virtual trait is less than the third threshold, the corresponding candidate virtual trait is selected as the target virtual trait for breeding or crossbreeding; otherwise, return to step S2 to recalculate.
[0077] In this embodiment, the first threshold is 0.05 (or 5%), the second threshold is 0.8, and the third threshold is 0.3. When VR < 0.05, R > 0.8, and NMSE < 0.3, it is selected as the target trait (virtual trait); when one of the conditions "VR < 0.05, R > 0.8, and NMSE < 0.3" is not met, it returns to the basic database and is recalculated.
[0078] This example demonstrates a method that, after constructing a basic trait database using a large number of hybrid combinations and their parental data from the breeding process, employs feature construction methods from machine learning to establish virtual traits. This identifies virtual traits with very low heterosis rates and high correlation to practical traits. These virtual traits are then used for parental selection and hybrid combination formulation, thereby screening out advantageous target combinations. This invention avoids the influence of heterosis on parental selection and hybrid combination formulation, avoiding the uncertainty and blind spots of current methods that directly rely on parental performance for selection and hybrid combination formulation. It accelerates the parental selection process, reduces the number of hybrid combinations, and speeds up the breeding process.
[0079] It should be understood that the embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.
Claims
1. A hybridization breeding method utilizing virtual traits, characterized in that, Includes the following steps: S1: Establish a basic database containing the hybrid generation and corresponding parental performance; construct a set of basic traits and establish a set of computational methods; S2: Basic traits and operation methods are randomly selected from the basic trait set and the operation method set. The feature construction method in machine learning is used to combine and construct candidate virtual traits, and construct a structural binary tree of virtual traits of the number of cocoons produced per 10,000 silkworms and the number of cocoon layers produced per 10,000 silkworms. S3: Using data from the basic database, a univariate regression model is established with the middle parent of the virtual trait as the independent variable and the corresponding hybrid F1 performance of the practical trait as the dependent variable. The normalized mean square error (NMSE) of the model is estimated using the leave-one-out cross-test method of the regression model. The parameters and test results of the practical trait and the alternative virtual trait are calculated. S4: Based on the comparison of the calculated parameters, test results, and corresponding thresholds, select the target virtual trait for breeding or crossbreeding.
2. The hybridization breeding method using virtual traits according to claim 1, characterized in that, The basic traits in the basic trait set are practical traits, including cocooning rate, cage death rate, total cocoon quantity, cocoon layer quantity, cocoon production per 10,000 silkworms, pupa rate, cocoon layer rate, and cocoon layer quantity per 10,000 silkworms.
3. A hybridization breeding method using virtual traits according to claim 1 or 2, characterized in that, The arithmetic operations in the set of arithmetic operations include addition, subtraction, multiplication, division, reciprocal, square root, square, cube, logarithm, and exp.
4. A hybridization breeding method using virtual traits according to claim 1 or 2, characterized in that, The parameters and test results of the practical traits and alternative virtual traits include the heterosis rate of the alternative virtual traits, the correlation coefficient between the parental values of the alternative virtual traits and the F1 hybrid performance of the practical traits, and the normalized mean square error of the regression model between the parental values of the alternative virtual traits and the F1 hybrid performance of the practical traits.
5. A hybridization breeding method using virtual traits according to claim 4, characterized in that, The calculation process for the heterosis rate (VR) of the candidate virtual traits is as follows: VR = (F1 - MP) / MP; Among them, F1 represents the results of the first generation of hybrids with virtual traits; MP represents the midparent value of a virtual trait.
6. The hybridization breeding method using virtual traits according to claim 4, characterized in that, The calculation process for the correlation coefficient R between the parent values of the candidate virtual traits and the F1 hybrid performance of the practical traits is as follows: in, Let be the virtual trait value of the i-th sample; Let be the practical trait value of the i-th sample; This represents the virtual trait mean of the sample population; This represents the average practical traits of the sample population.
7. The hybridization breeding method using virtual traits according to claim 4, characterized in that, The NMSE equation of the method is as follows: in, Let be the measured value of the i-th sample; Let be the predicted value for the i-th sample; This represents the population mean of the tested sample.
8. A hybridization breeding method using virtual traits according to claim 1, 5, 6, or 7, characterized in that, When the heterosis rate of the candidate virtual trait is less than the first threshold, and the correlation coefficient between the parental value and the performance of the first generation of hybrids of the candidate virtual trait is greater than the second threshold, and the normalized mean square error of the regression model between the parental value and the performance of the first generation of hybrids of the candidate virtual trait is less than the third threshold, the corresponding candidate virtual trait is selected as the target virtual trait for breeding or crossbreeding; otherwise, return to step S2 to recalculate.