GS breeding method based on exponential decay normalization weighting, electronic equipment and medium
By using exponential attenuation normal weighting algorithm to fuse multiple machine learning regression models in the GS breeding method, the problem of inconsistent performance of a single algorithm model on different data sets is solved, and the accuracy and efficiency of breeding are improved.
Patent Information
- Application Number
- CN202411952548.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-27
- Publication Date
- 2025-04-25
AI Technical Summary
The use of a single algorithm model in the prior art reduces the accuracy of breeding and makes it difficult to perform consistently on different data sets.
A GS breeding method based on exponential decay normal weighting is adopted to build a fusion model by fusing multiple machine learning regression algorithm models to improve the accuracy of breeding.
It improves the generalization ability of the model, can predict the relationship between genotype and phenotype more accurately, and improves the accuracy and efficiency of breeding.
Smart Images
Figure CN119889455A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of biological breeding, and in particular relates to a GS breeding method, method and medium based on exponential decay normalized weighting. Background Art
[0002] With the development of sequencing technology, sequencing throughput has been greatly improved, costs have been significantly reduced, and the computing power of computers has been continuously enhanced, which has promoted the development of new breeding technologies, especially the rise of genomic selection (GS) breeding. Genomic selection breeding overcomes the problems of difficult-to-measure traits, high contingency, long breeding cycle and high technical threshold through high-density molecular markers covering the entire genome, effectively accelerating the pace of breeding.
[0003] In the related technology, GS software with a single algorithm model is used to construct a GS model. Since different types of single algorithm models have their own advantages and disadvantages when processing genomic data with different characteristics, and different single algorithm models have large differences in performance on different data sets, it is often difficult to select an optimal single algorithm model in practical applications, which also reduces the accuracy of breeding. Summary of the invention
[0004] In view of this, the present invention provides a GS breeding method based on exponential decay normalized weighting to solve the problem in the prior art that the use of a single algorithm model reduces the accuracy of breeding.
[0005] In the first aspect, the present invention provides a GS breeding method based on exponential decay normalized weighting, comprising: dividing a plurality of samples based on species into a population data set; cleaning the population data set to obtain preferred data, the preferred data comprising non-missing genotype data and phenotypic data; converting the preferred data into a format recognized by machine learning to obtain converted format data; constructing a plurality of machine learning regression algorithm models based on the converted format data; fusing the plurality of machine learning regression algorithm models using an exponential decay normalized weighted algorithm to construct a fusion model; and predicting the breeding value of the breeding population based on the fusion model.
[0006] As an optional implementation, an exponential decay normalized weighted algorithm is used to fuse the multiple machine learning regression algorithm models to construct a fusion model, which specifically includes: obtaining the prediction result of each of the machine learning regression algorithm models; calculating the exponential decay weight according to the correlation distance of each of the machine learning regression algorithm models; and performing weighted averaging on the prediction results based on the exponential decay weight to obtain the fusion model. The calculation formula is as follows: In the formula, w i is the weight, y iis the prediction result; the weight is calculated according to the exponential decay function as follows:
[0007] Where λ is the decay rate parameter, d i is the correlation distance of the i-th machine learning regression algorithm model.
[0008] As an optional implementation, the preferred data is converted into a format recognized by machine learning to obtain converted format data, specifically including: performing dimensionality reduction processing on the genotype data of the preferred data to obtain an intermediate processing result; converting the intermediate processing result into a format recognized by machine learning to obtain the converted format data.
[0009] As an optional implementation, based on the converted format data, multiple machine learning regression algorithm models are constructed, specifically including: using a 10-fold cross-validation method to train the converted format data to obtain training data; based on the training data, finding the optimal hyperparameters for each of the machine learning regression algorithm models; and constructing a machine learning regression algorithm model based on the optimal hyperparameters.
[0010] As an optional implementation, a 10-fold cross-validation method is used to train the converted format data to obtain training data, specifically including: dividing the converted format data into ten subsets of equal size; training and validating the subsets ten times, wherein a different subset is selected each time as a validation set and the remaining subsets are used as training sets; and averaging the results of the ten validations to obtain the training data.
[0011] As an optional embodiment, the population data set includes a training population, and the training population includes genotype data and phenotypic data; or the population data includes a training population and a breeding population, and the breeding population includes genotype data.
[0012] As an optional implementation, the multiple machine learning regression algorithm models include lightGBM algorithm, XGBoost algorithm, HGBoost algorithm and ET algorithm.
[0013] As an optional implementation, the GS breeding method based on exponential decay normalized weighting also includes: using the Pearson correlation coefficient to evaluate the accuracy of the fusion model.
[0014] In a second aspect, the present invention provides an electronic device comprising: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: divide a plurality of samples based on species into a population data set; clean the population data set to obtain preferred data, the preferred data comprising non-missing genotype data and phenotypic data; convert the preferred data into a format recognizable by machine learning to obtain converted format data; based on the converted format data, construct a plurality of machine learning regression algorithm models; use an exponential decay normalized weighted algorithm to fuse the plurality of machine learning regression algorithm models to construct a fused model; and predict the breeding value of a breeding population based on the fused model.
[0015] In a third aspect, the present invention provides a computer-readable storage medium, characterized in that a computer program is stored therein, and when the computer program is executed by a processor, the GS breeding method based on exponential decay normalized weighting as described in any one of claims 1 to 8 is implemented.
[0016] The beneficial effects of the present invention are:
[0017] The present invention predicts the breeding value of a breeding population by constructing a fusion model, thereby improving the generalization ability of the model, being able to more accurately predict the relationship between genotype and phenotype, and improving the accuracy and efficiency of breeding. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the specific implementation of the present invention or the technical solution in the prior art, the following briefly introduces the drawings required for use in the specific implementation or the prior art description. Obviously, the drawings described below are some implementations of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0019] Figure 1 This is one of the flow charts of the GS breeding method based on exponential decay normalized weighting according to one embodiment of the present invention;
[0020] Figure 2 This is a second flow chart of the GS breeding method based on exponential decay normalized weighting according to an embodiment of the present invention;
[0021] Figure 3 This is one of the training result diagrams of the GS breeding method based on exponential decay normalized weighting according to one embodiment of the present invention;
[0022] Figure 4 This is the second training result diagram of the GS breeding method based on exponential decay normalized weighting according to one embodiment of the present invention;
[0023] Figure 5 This is the third figure of the training results of the GS breeding method based on exponential decay normalized weighting according to one embodiment of the present invention;
[0024] Figure 6 FIG4 is a fourth training result diagram of the GS breeding method based on exponential decay normalized weighting according to an embodiment of the present invention;
[0025] Figure 7 FIG5 is a fifth training result diagram of the GS breeding method based on exponential decay normalized weighting according to an embodiment of the present invention;
[0026] Figure 8 FIG6 is a sixth training result diagram of the GS breeding method based on exponential decay normalized weighting according to an embodiment of the present invention;
[0027] Fig. 9 This is a diagram showing the prediction results of the GS breeding method based on exponential decay normalized weighting according to an embodiment of the present invention;
[0028] Fig.10 This is a prediction result diagram of Case 1 of the GS breeding method based on exponential decay normalized weighting according to an embodiment of the present invention;
[0029] Fig.11 This is a prediction result diagram of Case 2 of the GS breeding method based on exponential decay normalized weighting according to an embodiment of the present invention;
[0030] Fig.12 This is a prediction result diagram of Case 3 of the GS breeding method based on exponential decay normalized weighting according to an embodiment of the present invention;
[0031] Fig.13 The figure is a block diagram of an electronic device according to an embodiment of the present invention.
[0032] Description of reference numerals:
[0033] 110. Processor; 120 - Memory. DETAILED DESCRIPTION
[0034] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0035] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.
[0036] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0037] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0038] In related technologies, breeding history has undergone a gradual exploration from empirical breeding to systematic breeding theories and methods, covering classical methods such as selection theory, pure line theory, backcross breeding, cyclic breeding, mutagenesis breeding, single seed transmission, ideal plant type, etc., and then entered the molecular marker-assisted selection stage, using various molecular markers, such as AFLP, SSR and SNP, to accelerate the improvement process.
[0039] With the rise of GS breeding, GS breeding has overcome problems such as difficult-to-measure traits, high randomness, long breeding cycles and high technical barriers through high-density molecular markers covering the entire genome, effectively accelerating the pace of breeding.
[0040] On this basis, we move towards smart breeding. The "scientific" part of smart breeding is mainly reflected in the construction and upgrading of the platform, establishing a crop breeding database with the help of germplasm resource information, integrating multi-omics data for joint analysis, breaking through technical bottlenecks such as biological big data acquisition, analysis and mining, and forming an integrated smart breeding strategy of "data-technology-algorithm-decision-making". This system not only improves the accuracy of breeding and shortens the breeding cycle, but also lays the foundation for the establishment of an efficient crop smart breeding system.
[0041] Currently, there are many machine learning and deep learning algorithms used to build GS models. Single algorithm models include GS models built by software such as cropGBM, DNNGP and DeepCCR. Among them, cropGBM is a GS model built based on the lightGBM algorithm of machine learning. Its default hyperparameters are preset values. There is a lack of sufficient hyperparameter optimization, which may affect the performance of the model; DNNGP is based on the convolutional neural network (CNN) framework to build a GS model. Its accuracy and applicability are relatively low, the hardware configuration requirements are high, and the time efficiency is poor; DeepCCR is based on the CNN+long short-term memory network (LSTM) framework to build a GS model. It still has limitations in applicability, and has high requirements for hardware equipment, and the processing speed is relatively slow.
[0042] These softwares play a role in different GS applications, but there is still much room for improvement in terms of accuracy, applicability and hardware requirements.
[0043] Different types of single algorithm models have their own advantages and disadvantages when processing genomic data with different characteristics. For example, linear models such as GBLUP handle linear relationships better, but have limited handling of nonlinear relationships; nonlinear models such as random forests, support vector machines, XGBoost, and LGBM handle nonlinear relationships better, but may lack the ability to generalize high-dimensional data.
[0044] The single algorithm model uses the parameters set by the algorithm as the default parameters, and does not seek the optimal hyperparameters to reduce the overfitting of the model to achieve the optimal model. When the amount of data is very large, the algorithm cannot run, so it is necessary to reduce the data dimension through the PCA dimensionality reduction method.
[0045] To this end, the present invention proposes a GS breeding method based on exponential decay normalized weighting, in order to effectively reduce cluster resource consumption, reduce labor costs and time costs, improve the accuracy of breeding, and promote the progress of breeding.
[0046] Combine the following Figures 1 to 13 , describing an embodiment of the present invention.
[0047] like Figure 1 As shown, according to an embodiment of the invention, a GS breeding method based on exponential decay normalized weighting is provided, comprising the following steps:
[0048] Step S101: Divide a plurality of samples based on species into group data sets.
[0049] Step S102: Clean the population data set to obtain preferred data, which includes genotype data and phenotype data without missing data.
[0050] Step S103: Convert the preferred data into a format recognized by machine learning to obtain converted format data.
[0051] Step S104: construct multiple machine learning regression algorithm models based on the transformed format data.
[0052] Step S105: using an exponential decay normalized weighted algorithm to fuse multiple machine learning regression algorithm models to construct a fusion model.
[0053] Step S106: predicting the breeding value of the breeding population based on the fusion model.
[0054] The GS breeding method based on exponential decay normalized weighting aims to improve breeding efficiency and accuracy through advanced data analysis technology. First, multiple samples based on a specific species need to be divided to create a population data set, which contains genetic information from different individuals.
[0055] Thorough cleaning of the above population data sets can ensure the quality of the data. The purpose of the cleaning process is to remove any incomplete or erroneous data that may affect the accuracy of the analysis, and finally obtain the optimal data set. The optimal data set includes complete genotype data and phenotypic data, and there are no missing values, which provides a solid foundation for subsequent analysis.
[0056] The optimal data set needs to be converted into a format suitable for machine learning algorithm recognition and processing. This conversion process involves data standardization, encoding and other technologies, and finally obtains the converted format data to prepare for the construction of the machine learning model.
[0057] After obtaining the transformed format data, multiple machine learning regression algorithm models are constructed based on the transformed format data, and then the exponential decay normalized weighted algorithm is used to fuse the multiple machine learning regression algorithm models constructed above. Through this weighted method, different weights can be assigned to each machine learning regression algorithm model, where the weights will be adjusted exponentially according to the performance and prediction accuracy of the machine learning regression algorithm model, and finally a fusion model with better overall performance will be constructed.
[0058] Based on the fusion model, the breeding value of the breeding population can be predicted. The breeding value is an important indicator to measure the genetic potential of individuals. By predicting this value, breeders can more effectively select individuals with excellent genetic characteristics for breeding, thereby accelerating the transmission and fixation of excellent traits and improving breeding efficiency.
[0059] Among them, samples include most plants such as rice, corn, wheat, cotton, etc., and can also include animals such as pigs, cattle, sheep, chickens, ducks, geese, etc., and can also include most types of variations, such as SNP, InDel and SV of resequencing or chip data, CNV data, etc.
[0060] Further, based on step S105, multiple machine learning regression algorithm models are fused using an exponential decay normalized weighted algorithm to construct a fusion model, which specifically includes the following steps:
[0061] Step S1051: Obtain the prediction results of each machine learning regression algorithm model;
[0062] Step S1052: Calculate the exponential decay weight according to the correlation distance of each machine learning regression algorithm model;
[0063] Step S1053: Perform weighted averaging on the prediction results based on the exponential decay weight to obtain a fusion model. The calculation formula is as follows:
[0064]
[0065] In the formula, w i is the weight, y i is the predicted result;
[0066] The weight is calculated according to the exponential decay function as follows:
[0067]
[0068] Where λ is the decay rate parameter, d i is the correlation distance of the i-th machine learning regression algorithm model.
[0069] In this embodiment, an exponential decay normalized weighting method is used to fuse multiple machine learning regression algorithm models to obtain a fusion model. The detailed method is:
[0070] 1) Determine the model prediction results and obtain the prediction results y of n machine learning regression algorithm models i .
[0071] 2) Calculate the weights based on the correlation distance d of each machine learning regression algorithm model i , calculate the exponential decay weight w i .
[0072] 3) Weighted fusion: Use the calculated weights to perform weighted averaging on the model prediction results to obtain the final fusion result y.
[0073] Suppose there are n independent machine learning models, and their prediction results are y1, y2, ..., y i ,...,y n . Use weights w1,w2,...,w i ,...,w n These models are weighted and fused to obtain the final prediction result y.
[0074] λ controls the decay rate of the weight, and a larger value of λ will make the weight decay faster. i It reflects the relevance of the model's prediction results to the current task.
[0075] Furthermore, if Figure 2 As shown, the population data set includes a training population, and the training population includes genotype data and phenotypic data; or the population data includes a training population and a breeding population, and the breeding population includes genotype data.
[0076] At least a training population is required, and a breeding population can also be provided if conditions permit. The training population requires genotype data and phenotypic data; after data cleaning, a training population with high-quality phenotypic data and genotype data is obtained. The PCA method is then used for dimensionality reduction, which greatly reduces the dimension of the genotype matrix.
[0077] The phenotypic and genotypic data of the training population (TP) and the genotypic data of the breeding population (BP) were cleaned to obtain high-quality data without missing data. The genotypes of TP and BP were processed together using the PCA dimensionality reduction method, in which the principal components with a cumulative variance of 95% were selected to obtain low-dimensional data, which were then converted into a format recognized by machine learning, including the conversion of phenotypic data. The converted data were trained using the lightGBM algorithm, XGBoost algorithm, HGBoost algorithm, ET algorithm and other four machine learning regressors using a 10-fold cross-validation method. Each algorithm used the Bayesian optimization method to find the optimal hyperparameters, and constructed its own model. Finally, the four models were fused using the exponential decay normalized weighting algorithm to construct a fusion model; the breeding value of the breeding population was predicted by the constructed fusion model, and individuals with high or low breeding values were selected for seed preservation and breeding as needed.
[0078] Specifically, multiple machine learning regression algorithm models include lightGBM algorithm, XGBoost algorithm, HGBoost algorithm and ET algorithm. The lightGBM (Light Gradient Boosting Machine) algorithm, XGBoost (Extreme Gradient Boosting) algorithm, HGBoost (Histogram-based Gradient Boosting) algorithm and ET (Extra Trees) algorithm are all machine learning regressors.
[0079] Further, based on step S103, the preferred data is converted into a format recognized by machine learning to obtain converted format data, which specifically includes the following steps:
[0080] Step S1031: performing dimension reduction processing on the genotype data in the preferred data to obtain an intermediate processing result;
[0081] Step S1033: Convert the intermediate processing result into a format recognized by machine learning to obtain converted format data.
[0082] Perform dimensionality reduction on the genotype data of the training population and the genotype data of the breeding population. The purpose of dimensionality reduction is to reduce the complexity of the data while retaining the most valuable information for subsequent analysis. Through this process, an intermediate processing result can be obtained, which will lay the foundation for the next step of format conversion. The intermediate processing result obtained in the previous step is further converted into a format that can be recognized and processed by the machine learning algorithm. This conversion process ensures that the structure and content of the data can be understood by the machine learning model, thereby obtaining the final converted format data. This step is a crucial link in the entire data preprocessing process, which directly affects the performance of the machine learning model and the final analysis results.
[0083] Further, based on step S104, multiple machine learning regression algorithm models are constructed based on the transformed format data, specifically including the following steps:
[0084] Step S1041: Use a 10-fold cross validation method to train the transformed format data to obtain training data.
[0085] Step S1042: Based on the training data, find the optimal hyperparameters for each machine learning regression algorithm model.
[0086] Step S1043: Construct a machine learning regression algorithm model based on the optimal hyperparameters.
[0087] In this embodiment, a series of machine learning regression algorithm models will be constructed based on the data set in the transformed format. This process involves multiple steps, aiming to achieve the best prediction performance through precise model training and parameter optimization. First, a 10-fold cross-validation method is used to train the data set in the transformed format. Next, the training data obtained in the previous step is used to make detailed parameter adjustments for each machine learning regression algorithm model. This process includes finding the optimal hyperparameters, i.e., those parameters that can control the complexity and learning ability of the model, such as learning rate, regularization strength, tree depth, etc. By systematically adjusting these parameters, we can find the model configuration that best suits the current data set.
[0088] Based on the optimal hyperparameters determined in step S1042, a final machine learning regression algorithm model is constructed. This model will integrate all optimized parameter settings in order to achieve the best prediction effect. The constructed model will be able to perform accurate regression analysis on new data.
[0089] Further, based on step S1041, a 10-fold cross validation method is used to train the transformed format data to obtain training data, which specifically includes the following steps:
[0090] Step S10411: Divide the converted format data into ten subsets of equal size.
[0091] Step S10412: The subsets are trained and validated ten times, wherein a different subset is selected as the validation set each time, and the remaining subsets are used as the training set.
[0092] Step S10413: Average the results of ten verifications to obtain training data.
[0093] In this embodiment, the data set is divided into 10 parts, 9 of which are used as training data and the remaining 1 part is used as verification data in turn, so as to ensure that each data point has the opportunity to be used for verification, thereby evaluating the generalization ability of the model and obtaining a fully trained data set. The training results are shown in Figure 2. Figures 3 to 8 shown.
[0094] Furthermore, the GS breeding method based on exponential decay normalized weighting also includes: using the Pearson correlation coefficient to evaluate the accuracy of the fusion model.
[0095] By combining the prediction results of multiple single models, the present invention can integrate the advantages of each model and reduce the possible deviations of individual models. This method can significantly improve the accuracy and stability of prediction.
[0096] Among them, a single model sometimes overfits the training data, but the present invention can alleviate this problem by integrating the prediction results of multiple models, reducing the risk of overfitting and making the fusion model more robust on new data. The present invention can also better capture the complex patterns and diversity in the data, thereby improving the generalization ability of the model and making it perform well on different data sets.
[0097] In addition, since a single model may perform poorly due to some abnormal data or special circumstances, the present invention can effectively reduce the prediction error of a single model caused by special circumstances by integrating the prediction results of multiple models, thereby enhancing the fault tolerance.
[0098] Different models may focus on different features and patterns when processing data. The present invention utilizes the complementarity of the models, and the fusion model can make full use of this complementarity, making the final prediction result more comprehensive and reliable.
[0099] like Fig.13As shown, the present invention also provides an electronic device, comprising: at least one processor 110; a memory 120 connected to the at least one processor in communication; wherein the memory 120 stores instructions executable by the at least one processor 110, and the instructions are executed by the at least one processor 110, so that the at least one processor 110 can implement the following steps:
[0100] Step S101: Divide a plurality of samples based on species into group data sets.
[0101] Step S102: Clean the population data set to obtain preferred data, which includes genotype data and phenotype data without missing data.
[0102] Step S103: Convert the preferred data into a format recognized by machine learning to obtain converted format data.
[0103] Step S104: construct multiple machine learning regression algorithm models based on the transformed format data.
[0104] Step S105: using an exponential decay normalized weighted algorithm to fuse multiple machine learning regression algorithm models to construct a fusion model.
[0105] Step S106: predicting the breeding value of the breeding population based on the fusion model.
[0106] The present invention also provides a readable storage medium storing a computer program. When the computer program is executed by a processor, the steps of implementing any one of the GS breeding methods based on exponential decay normalized weighting are as follows:
[0107] Step S101: Acquire gas data, the gas data including greenhouse gas concentration, wind speed and wind direction;
[0108] Step S103: Calculating the greenhouse gas concentration distribution in the monitoring area according to the gas data;
[0109] Step S105: obtaining the real-time total amount of greenhouse gases according to the greenhouse gas concentration distribution;
[0110] Step S107: Obtain greenhouse gas emissions based on the real-time total amount of greenhouse gases and the greenhouse gas concentration distribution.
[0111] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.
[0112] The computer program of the present invention obtains a training group of high-quality phenotypic and genotypic data after data cleaning; then uses the PCA method for dimensionality reduction, greatly reduces the dimension of the genotype matrix, and significantly reduces the training time and prediction time. Then, an exponential decay normalized weighted algorithm is used to build a model, and after Bayesian algorithm optimization, the optimal model is built after finding the optimal hyperparameter; finally, the breeding value of the breeding group with unknown phenotype but known genotype is predicted by the constructed model, and the best samples are selected for breeding, or individuals with high breeding value or low breeding value are taken for seed preservation as needed, and generally the individuals with top 5% or top 1% breeding value are taken for breeding.
[0113] In the context of the present invention, a computer readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, device, or apparatus. A computer readable storage medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing.
[0114] Alternatively, the computer readable storage medium may be a machine readable signal medium. More specific examples of machine readable storage media may include electrical connections based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
[0115] The present invention is based on the exponential decay normalized weighted algorithm to fuse multiple machine learning algorithms to construct a GS model with better comprehensive ability, and uses the Bayes optimization algorithm to find the optimal hyperparameters and the optimal model, thereby improving the accuracy, applicability and generalization ability of constructing the GS model. The limitations of a single model, the difficulty in model selection, the optimal solution of the algorithm, and the time efficiency problems are solved, and the accuracy of the fusion model is also improved. On the one hand, the present invention greatly improves the adaptability of the GS model in different scenarios, and on the other hand, it improves the generalization ability of the GS model, thereby improving the accuracy of offspring selection, correctly selecting offspring for seed preservation, shortening the breeding cycle, and reducing the time cost and labor cost of breeding. On the other hand, it greatly reduces server resource consumption and computing time.
[0116] The present invention is not only applicable to high-density markers (SNPs, InDels, SVs) for high-throughput sequencing, but can also be used for genomic breeding needs of chips and simplified genome sequencing, providing a rapid selection platform for breeders to cultivate their own varieties.
[0117] The software program can be set up in four steps: the first step is to fill in the genotype files of the training population and the breeding population to obtain non-missing genotypes; the second step is to process the phenotype files and genotype files of the training population, remove the missing phenotype samples, perform or not perform filtering or PCA dimensionality reduction based on GWAS results on the genotype files, and then convert the filtered vcf into a machine learning recognition format; the third step is to train the model, using a 10-fold cross-validation method, using an exponentially decayed normalized weighted fusion machine algorithm for training, whether to use the Bayesian optimization algorithm hyperparameters, and using 4 indicators to evaluate the model, and statistically and graphically; the fourth step is to predict the breeding value of the breeding population. The training results are as follows: Figures 3 to 8 As shown, the prediction results are Fig. 9 shown.
[0118] The specific implementation cases of the breeding method of the present invention using 3k rice data, wheat data, cotton data, corn data and other public data are as follows.
[0119] Case 1, combined Fig.10 As shown, 13 yield-related phenotypes and SNP data of 2110 samples of 3k rice data were used to construct the GS model using software, and the Pearson correlation coefficient (PCCs) was used to evaluate the model accuracy. Each model was subjected to a 10-fold cross-validation test and the average was calculated. Figure 6 The horizontal axis is 13 traits, the vertical axis is PCCs, and the inner column is the PCC value of all models (0-1). Species: rice; Samples: 2,110 samples; Variation: 1,011,602 SNP loci; Traits: 13 traits; Indicator: PCC.
[0120] From left to right, the gray columns are traditional GBLUP models, the green and purple columns are deep learning DNNGP and DeepCCR models, the light orange, yellow, magenta, and dark orange are single machine learning models, and the blue columns are fusion models. From the figure, we can see that the PCCs value of the fusion model is the highest in 13 traits, which is greater than the traditional model, single machine learning model, and deep learning model.
[0121] like Fig.11 As shown in the figure, Case 2: Using 6 yield-related phenotypes and SNP data of 2,000 wheat samples, the software was used to build a GS model, and the Pearson correlation coefficient (PCCs) was used to evaluate the accuracy of the model. Each model was tested with 10-fold cross validation and the average was calculated. Species: wheat; Samples: 2,000 samples; Variation: 33,709 SNP loci; Traits: 6 traits; Indicator: PCC.
[0122] In the figure, the horizontal axis is the six traits, the vertical axis is the PCCs value, and the inner column is the PCC value of all models (0-0.8). From left to right, the gray column is the traditional GBLUP model, the green and purple columns are the deep learning DNNGP and DeepCCR models, the green and purple columns are the deep learning DNNGP and DeepCCR models, the light orange, yellow, magenta, and dark orange are single machine learning models, and the blue column is the fusion model. From the figure, we can see that the PCCs value of the fusion model is the highest in the six traits, which is greater than the traditional model, the single machine learning model, and the deep learning model.
[0123] like Fig.12 As shown in the figure, Case 3: Using 4 cotton fiber-related phenotypes and SNP data of 1,245 cotton data samples, the software was used to build a GS model, and the Pearson correlation coefficient (PCCs) was used to evaluate the accuracy of the model. Each model was tested with 10-fold cross validation and the average was calculated. Species: cotton; Samples: 1,245 samples; Variation: 1,112,352 SNP loci Traits: 4 traits; Indicator: PCC. The horizontal axis in the figure is the 4 traits, the vertical axis is the PCCs value, and the inner column is the PCC value of all models (0-1). From left to right, the gray columns are traditional GBLUP models, the green and purple columns are deep learning DNNGP and DeepCCR models, the light orange, yellow, magenta, and dark orange are single machine learning models, and the blue columns are fusion models. From the figure, it can be seen that the PCCs value of the fusion model is the highest in the four traits, which is greater than the traditional model, single machine learning model, and deep learning model.
[0124] Obviously, the above embodiments are merely examples for clear description and are not limitations of the implementation methods.
[0125] For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. However, the obvious changes or modifications derived from this are still within the protection scope of the invention.
Claims
1. A GS breeding method based on exponential decay normalized weighting, characterized in that: include: Divide the population datasets into multiple samples based on species; Cleaning the population data set to obtain preferred data, wherein the preferred data includes non-missing genotype data and phenotype data; Converting the preferred data into a format recognized by machine learning to obtain converted format data; Based on the converted format data, construct multiple machine learning regression algorithm models; The multiple machine learning regression algorithm models are fused using an exponential decay normalized weighted algorithm to construct a fusion model; The breeding values of the breeding population are predicted based on the fusion model.
2. The GS breeding method based on exponential decay normalized weighting according to claim 1, characterized in that: The multiple machine learning regression algorithm models are fused by using an exponential decay normalized weighted algorithm to construct a fusion model, which specifically includes: Obtaining prediction results of each of the machine learning regression algorithm models; Calculating an exponential decay weight according to the correlation distance of each of the machine learning regression algorithm models; The prediction results are weighted averaged based on the exponential decay weight to obtain the fusion model, and the calculation formula is as follows: In the formula, w i is the weight, y i is the predicted result; The weight is calculated according to the exponential decay function as follows: Where λ is the decay rate parameter, d i is the correlation distance of the i-th machine learning regression algorithm model.
3. The GS breeding method based on exponential decay normalized weighting according to claim 1, characterized in that: The preferred data is converted into a format recognized by machine learning to obtain converted format data, specifically including: Performing dimension reduction processing on the genotype data of the preferred data to obtain an intermediate processing result; The intermediate processing result is converted into a format recognized by machine learning to obtain the converted format data.
4. The GS breeding method based on exponential decay normalized weighting according to claim 1, characterized in that: Based on the converted format data, multiple machine learning regression algorithm models are constructed, including: The transformed format data is trained using a 10-fold cross validation method to obtain training data; Based on the training data, finding the optimal hyperparameters for each of the machine learning regression algorithm models; A machine learning regression algorithm model is constructed based on the optimal hyperparameters.
5. The GS breeding method based on exponential decay normalized weighting according to claim 4, characterized in that: The transformed format data is trained using a 10-fold cross validation method to obtain training data, specifically including: Dividing the converted format data into ten subsets of equal size; The subsets are trained and validated ten times, wherein a different subset is selected as a validation set each time, and the remaining subset is used as a training set; The results of ten verifications are averaged to obtain the training data.
6. The GS breeding method based on exponential decay normalized weighting according to any one of claims 1 to 5, characterized in that: The population data set includes a training population, the training population includes genotype data and phenotype data; or the population data includes a training population and a breeding population, the breeding population includes genotype data.
7. The GS breeding method based on exponential decay normalized weighting according to any one of claims 1 to 5, characterized in that: The multiple machine learning regression algorithm models include lightGBM algorithm, XGBoost algorithm, HGBoost algorithm and ET algorithm.
8. The GS breeding method based on exponential decay normalized weighting according to any one of claims 1 to 5, characterized in that: Also includes: The Pearson correlation coefficient was used to evaluate the accuracy of the fusion model.
9. An electronic device, characterized in that: include: at least one processor (110); a memory (120) communicatively coupled to the at least one processor (110); The memory (120) stores instructions executable by the at least one processor (110), and the instructions are executed by the at least one processor (110) to enable the at least one processor (110) to: Divide the population datasets into multiple samples based on species; Cleaning the population data set to obtain preferred data, wherein the preferred data includes non-missing genotype data and phenotype data; Converting the preferred data into a format recognized by machine learning to obtain converted format data; Based on the converted format data, construct multiple machine learning regression algorithm models; The multiple machine learning regression algorithm models are fused using an exponential decay normalized weighted algorithm to construct a fusion model; The breeding values of the breeding population are predicted based on the fusion model.
10. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the GS breeding method based on exponential decay normalized weighting as described in any one of claims 1 to 8 is implemented.