Breeding phenotype prediction method and device of whole genome data based on machine learning, electronic equipment, medium and product

Abstract

The invention discloses a breeding phenotype prediction method and device of whole genome data based on machine learning, electronic equipment, a medium and a product. The method comprises the following steps: acquiring SNP locus data of any organism; the organism comprises a plant or an animal; and according to the SNP locus data of the organism and the phenotype prediction model, realizing phenotype prediction of any organism. When the breeding phenotype prediction method is adopted for breeding, the Pearson correlation coefficient of the breeding genotype and the phenotypic character is high, precise breeding can be achieved, and phenotypic prediction of a specific genotype organism can be improved.

Description

Technical Field

[0001] The present application belongs to the field of biological breeding technology, and in particular relates to a breeding phenotype prediction method, device, electronic equipment, medium and product based on whole genome data of machine learning. Background Art

[0002] With the further development of high-throughput sequencing technology and the continuous reduction of sequencing costs, a large amount of key breeding data has been accumulated. Biological breeding has gradually developed from traditional breeding based on breeders' experience to whole genome selection assisted breeding with the help of high-density molecular markers, realizing digital and intelligent breeding. Genome selection technology is roughly divided into two parts at the platform level: one is the acquisition of genotypes, and the other is the prediction from genotype to phenotype. At present, in the field of animal and plant breeding, gene chip technology is widely used for genotype acquisition; for phenotype prediction, traditional statistical algorithm models are generally used. This strategy has accumulated a lot of fruitful experience for the practical work of biological breeding, laid a solid theoretical foundation for precision molecular design breeding, and promoted the establishment and improvement of a modern intelligent breeding system. However, with the continuous deepening of biological breeding work, the shortcomings of this system have gradually emerged, mainly reflected in the following aspects:

[0003] First, the number of molecular markers obtained based on gene chip technology is limited. Many of them are group functional sites selected based on the main effect sites. When faced with whole genome molecular markers, there will be a lack of functional sites. Secondly, models built based on traditional statistical algorithms, such as rrBLUP, are built based on statistical regression algorithms and belong to the category of linear models, which are mainly aimed at capturing the linear association between genotype and phenotype.

[0004] Therefore, how to speed up the acquisition of organisms with expected characteristics, improve breeding efficiency and increase breeding accuracy has become an urgent problem to be solved in the breeding field. Summary of the invention

[0005] The breeding phenotype prediction method, device, electronic device, medium and product based on machine learning whole genome data provided in the embodiments of the present application continuously learn and optimize the capture of complex associations between genotypes and phenotypes, filter redundant information, improve the utilization rate of effective information, and comprehensively improve the accuracy and operation speed of genome selection. It can accurately predict the phenotype of an organism, thereby improving breeding efficiency and breeding accuracy.

[0006] In a first aspect, the present application provides a breeding phenotype prediction method based on whole genome data of machine learning, the method comprising:

[0007] Obtaining SNP site data of any organism; the organism includes plants or animals;

[0008] Based on the SNP site data of the organism and the phenotype prediction model, the phenotype prediction of any organism can be achieved.

[0009] In some optional embodiments, the method for constructing the phenotype prediction model includes:

[0010] Construct a source model, wherein the source model includes at least seven data processing levels; wherein each data processing level includes calculating using formula (1),

[0011]

[0012] in, represents the weight vector, Represents the input data vector represents the bias vector, n represents the number of input data, x i represents the digitally encoded input data of the i-th gene, W i Corresponding to x i The weight of B i Corresponding to x i The bias of , · represents the inner product operation, × represents the multiplication operation, ∑ represents the summation operation, and f(x) represents the output; the output data of the previous level corresponds to the input data of the next level; and The value of is updated according to the following formula:

[0013]

[0014] Among them, loss represents the overall loss value, y real represents the observed true phenotypic value, y predict represents the predicted phenotypic value, ||W|| 2 represents the bi-norm of all weights, η represents the learning rate, represents the partial derivative operation, W i0 Indicates the initialized W i The value of B i0 B represents the initialization i The last level of data includes the phenotypic trait values ​​of n input data;

[0015] The source model is trained using the SNP site data of a specific population and its phenotype data as training data to obtain a phenotype prediction model.

[0016] In some optional embodiments, obtaining SNP site data of any organism includes:

[0017] Obtain whole genome sequencing data for any organism;

[0018] Based on the whole genome sequencing data, the SNP site data of the organism is obtained.

[0019] In some optional embodiments, the SNP site data includes first SNP site data; the first SNP site data is obtained by functionally scoring single nucleotide polymorphisms in whole genome sequencing data.

[0020] In some optional embodiments, the first SNP site data is obtained by functionally scoring single nucleotide polymorphisms in whole genome sequencing data, including:

[0021] Performing genome-wide association analysis on the genome sequencing data to calculate the significance level of SNP;

[0022] According to the SNP significance level p value and the SNP preset significance level interval, the first SNP site data with potential functionality is obtained.

[0023] In some optional embodiments, the SNP site data includes second SNP site data; the second SNP site data is obtained by sorting the SNPs by the difference level of the epigenetic group data to obtain the second SNP site data matching the epigenetic group data.

[0024] In some optional embodiments, the second SNP site data is obtained by sorting the SNPs by the difference level of the epigenomic data to obtain the second SNP site data matching the epigenomic data, including:

[0025] The epigenomic data were processed using epigenomic functional annotation software to obtain the epigenomic SNP sites corresponding to the epigenomic data;

[0026] Sorting the apparent SNP sites according to the openness of the quantitative SNP to obtain an apparent SNP sorting result;

[0027] According to the set threshold of the epigenomic functional annotation software and the epigenomic SNP ranking result, second SNP site data matching the epigenomic data are obtained.

[0028] In some optional embodiments, before the source model is trained using the SNP site data of a specific population and its phenotype data to obtain a phenotype prediction model, the method further includes:

[0029] Based on the SNP site data, the SNP site data is converted into a processable information file using conversion software.

[0030] In some optional embodiments, the source model includes nine data processing levels, wherein the input values ​​of the first processing level are behavioral sample names, the columns of the input values ​​of the first processing level are SNP site data, the output values ​​of the last processing level are behavioral sample names, and the columns of the output values ​​of the last processing level are phenotypic trait values.

[0031] In some optional embodiments, between the last processing level and the first data processing level, the construction method further includes:

[0032] The training data is subjected to penalty correction of feature weights to remove redundant information and achieve association between input data and phenotypic trait values; the feature weight penalty correction is performed using the following formula:

[0033]

[0034] in Represents the bi-norm operation, and n represents the number of input data.

[0035] In some optional embodiments, after performing penalty correction on the feature weights of the training data (whole genome sequencing data), the construction method further comprises:

[0036] The training data is batch normalized, and the output data is corrected by a ReLU activation function.

[0037] Among them, batch standardization is performed by the following formula:

[0038]

[0039] Among them, mean(x) represents the mean of the input data values, and var(x) represents the variance of the input data values.

[0040] Among them, the ReLU activation function is performed by the following formula:

[0041]

[0042] In some optional embodiments, the feature numbers of the output values ​​of the nine-level data processing hierarchy are 90, 180, 360, 450, 540, 630, 720, 810, and 1 in the first to ninth levels respectively.

[0043] In a second aspect, the present application provides a breeding phenotype prediction device based on whole genome data of machine learning, the device comprising:

[0044] An acquisition module, used to acquire SNP site data of any organism; the organism includes plants or animals;

[0045] The prediction module is used to predict the phenotype of any organism based on the SNP site data of the organism and the phenotype prediction model.

[0046] In a third aspect, an embodiment of the present application provides an electronic device, the device comprising: a processor and a memory storing computer program instructions;

[0047] When the processor executes the computer program instructions, the method of the first aspect is implemented.

[0048] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the method of the first aspect is implemented.

[0049] In a fifth aspect, an embodiment of the present application provides a computer program product, which is stored in a storage medium. When the computer program product is executed by at least one processor, it performs the method of the first aspect.

[0050] The prediction method of the embodiment of the present application can quickly predict the phenotype of the organism by inputting the SNP site data of the organism. The Pearson correlation coefficient between the breeding genotype and the phenotypic trait is high, and the phenotype of the organism can be accurately predicted. Applying this method to breeding can improve breeding efficiency and breeding accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0052] Figure 1 A schematic diagram of the breeding method flow chart of some embodiments of the present application;

[0053] Figure 2 The hyperparameter combination of the breeding phenotype prediction method of the present application is shown;

[0054] Figure 3 The evaluation result of the prediction performance of soybean 100-grain weight trait according to one embodiment is shown;

[0055] Figure 4 The evaluation results of the prediction performance of Duroc pig eye loin depth trait in one embodiment are shown;

[0056] Figure 5 A schematic diagram of the structure of a whole genome selection breeding device based on machine learning provided in an embodiment of the present application is shown;

[0057] Figure 6A schematic structural diagram of an embodiment of an electronic device provided by the present application is shown. DETAILED DESCRIPTION

[0058] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0059] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0060] As the cost of sequencing continues to decrease, the application of obtaining genetic information at the whole genome scale at the individual level has become popular, which includes all the genotype information of the individual. For important economic traits of plants and animals, not all SNP sites have a large genetic effect, which means that the whole genome genetic information is redundant and it is necessary to screen functional SNPs at the whole genome scale. In fact, the sites contained in the gene chip are also a collection of functional SNP sites in a specific scenario. Generally speaking, the potential causal molecular marker sites related to phenotypes can be screened through whole genome association analysis. Introducing such functional SNP sites in phenotypic prediction work can improve the accuracy of genome selection. In addition, more and more studies have shown that in addition to being distributed in the coding region, most of the causal sites related to economic traits are located in the non-coding region. Introducing such functional SNP sites with regulatory properties can also improve the accuracy of genome selection.

[0061] Secondly, models built based on traditional statistical algorithms, such as rrBLUP, are built based on statistical regression algorithms and belong to the category of linear models, which are mainly used to capture the linear association between genotype and phenotype. From the perspective of SNP interaction, the additive effect is the superposition of SNP effects. When faced with traits dominated by main effects (such as quality traits), the prediction performance is relatively good. However, due to the existence of dominant effects, epistatic effects, etc., there are still complex nonlinear associations between genotypes and phenotypes. When faced with traits dominated by micro-effect polygenic effects (such as quantitative traits), the prediction performance is relatively poor.

[0062] The genomic selection system based on functional variant sites and deep learning technology is still in its infancy. Although existing studies have shown that the use of functional variant sites screened from the whole genome scale and the use of deep learning technology can improve the accuracy of genomic selection, the exploration of the combination of the two is still not in-depth.

[0063] In order to solve the technical problems involved in the background technology, the present application embodiment provides a breeding phenotype prediction method based on whole genome data of machine learning, such as Figure 1 , the method comprising:

[0064] S100, obtaining SNP site data of any organism; the organism includes a plant or an animal;

[0065] S200, realizing phenotype prediction of any organism based on the SNP site data of the organism and the phenotype prediction model.

[0066] In breeding, in view of the complex nonlinear relationship between genotype and phenotype, traditional statistical models, which model linear relationships, limit the accuracy and speed of phenotypic prediction results to a certain extent.

[0067] The phenotype prediction method of the present application can utilize the mapping relationship between SNP site data and phenotype to predict the individual phenotype of a specific organism by understanding its functional variation sites. Due to the existence of the activation function, nonlinear associations can be captured on a linear basis. Since the above-mentioned machine learning realizes the capture of complex associations between genotype and phenotype, the Pearson correlation coefficient of the prediction results for the candidate individual phenotype is relatively high, thereby improving the accuracy and speed of the phenotype prediction results.

[0068] In addition, the breeding phenotype prediction method of the present application solves the problem in the prior art that the number of functional variation sites is too small and cannot fully capture the complex relationship between genotype and phenotype, thereby improving the accuracy of phenotype prediction.

[0069] In some optional embodiments, the method for constructing the phenotype prediction model includes:

[0070] Construct a source model, wherein the source model includes at least seven data processing levels; wherein each data processing level includes calculating using formula (1),

[0071]

[0072] in, represents the weight vector, represents the input data vector, represents the bias vector, n represents the number of input data, x i represents the digitally encoded input data of the i-th gene, W i Corresponding to x i The weight of B i Corresponding to x i The bias of , · represents the inner product operation, × represents the multiplication operation, ∑ represents the summation operation, and f(x) represents the output; the output data of the previous level corresponds to the input data of the next level; and The value of is updated according to the following formula:

[0073]

[0074] Among them, loss represents the overall loss value, y real represents the observed true phenotypic value, y predict represents the predicted phenotypic value, ||W|| 2 represents the bi-norm of all weights, η represents the learning rate, represents the partial derivative operation, W i0 Indicates the initialized W i The value of B i0 B represents the initialization i The last level of data includes the phenotypic trait values ​​of n input data;

[0075] The source model is trained using the SNP site data of a specific population and its phenotype data as training data to obtain a phenotype prediction model.

[0076] For example, the input data can have SNP site data of n samples, the columns of the input data are the names of n samples, and the rows of the input data are the digitally encoded genotypes of the SNPs. After the first layer inputs this rectangular data, according to the updated parameters, the output results of the first layer are used as the second input data, and the cycle is repeated. The input data of the last level is the output result of the previous level, and the output data of the last level corresponds to the actual phenotype of the actual population or individual.

[0077] The bi-norm is a penalty method or a calculation rule. η represents the learning rate, and its value can be 0.01.

[0078] For example, n samples, that is, x i The genotype matrix of this sample can be a vector with a length equal to the number of SNPs.

[0079] Loss can be understood as a correction of loss, a correction of weights, and a reduction in the risk of overfitting the model.

[0080] For the basic output f(x), batch normalization BN is performed to standardize the mean and variance of the basic output so that it obeys a normal distribution with a variance of 0 and a mean of 1, reducing overfitting and improving the generalization ability of the model.

[0081] For the batch normalized output f(x), ReLU activation is performed to keep the gradient from decaying, thereby alleviating the gradient vanishing problem, and also allowing for faster convergence and improved model computing efficiency, that is, according to the following rules:

[0082]

[0083] For the output after ReLU activation, a certain proportion can be randomly discarded, such as 0.1, to prevent overfitting and improve the generalization ability of the model.

[0084] The training data may include whole genome sequencing data and SNP site data and phenotypic data therein. Therefore, the whole genome sequencing data and phenotypic data in the training data may be used to train and adjust the algorithm model, so as to capture the complex association between genotype and phenotype and improve the efficiency and accuracy of the breeding method.

[0085] In this embodiment, by adopting this digitally encoded genotype information or phenotypic information input method and through multi-level construction, the complex association between genotype and phenotype can be effectively realized, thereby making the breeding method and phenotypic prediction results of the algorithm model more efficient and accurate.

[0086] In some optional embodiments, S100, obtaining SNP site data of any organism includes:

[0087] S110, obtaining whole genome sequencing data of any organism;

[0088] S120, obtaining SNP site data of the organism according to the whole genome sequencing data.

[0089] In some optional embodiments, the SNP site data includes first SNP site data; the first SNP site data is obtained by functionally scoring single nucleotide polymorphisms in whole genome sequencing data.

[0090] In some optional embodiments, the first SNP site data is obtained by functionally scoring single nucleotide polymorphisms in whole genome sequencing data, including:

[0091] Performing genome-wide association analysis on the genome sequencing data to calculate the significance level of SNP;

[0092] According to the SNP significance level p value and the SNP preset significance level interval, the first SNP site data with potential functionality is obtained.

[0093] It can be understood that: starting from the original vcf file, plink can be used to perform GWAS analysis, the significance level p can be set, potential functional sites, that is, the first SNP site data can be obtained, the first SNP site data can be extracted from the original vcf file using plink software, and it can be input into the algorithm model to achieve modeling.

[0094] In this embodiment, GWAS software can be used to calculate the significance level of SNP, for example, the significance level, that is, the p value can be <1.0e-24, etc.

[0095] In this embodiment, by functionally scoring the genetic molecular markers of the whole genome, a method for screening functional variation sites is proposed, which solves the limitation of too few potential causal markers in gene chips. Through algorithm models, training data and deep learning technology, the limitation of capturing complex associations between genotypes and phenotypes can be solved.

[0096] In some optional embodiments, the SNP site data includes second SNP site data; the second SNP site data is obtained by sorting the SNPs by the difference level of the epigenetic group data to obtain the second SNP site data matching the epigenetic group data.

[0097] In some optional embodiments, the second SNP site data is obtained by sorting the SNPs by the difference level of the epigenomic data to obtain the second SNP site data matching the epigenomic data, including:

[0098] The epigenomic data were processed using epigenomic functional annotation software to obtain the epigenomic SNP sites corresponding to the epigenomic data;

[0099] Sorting the apparent SNP sites according to the openness of the quantitative SNP to obtain an apparent SNP sorting result;

[0100] According to the set threshold of the epigenomic functional annotation software and the epigenomic SNP ranking result, second SNP site data matching the epigenomic data are obtained.

[0101] It can be understood that: starting from the peak file of the phenotype data, DeepSATA can be used to annotate the regulatory function, set the number threshold n, obtain the potential functional sites, that is, the second SNP site data, use plink software to extract the second SNP site data from the original vcf file, and input it into the algorithm model to achieve modeling.

[0102] In the embodiments of the present application, the SNP openness is a general term. According to the epigenomic data, the number of reads at the SNP site can be counted. The more reads, the more open it is. A standardized process will be performed based on the epigenomic data, and then the epigenomic data will be modeled using a model to calculate the epigenomic signal value at the SNP site. The higher the signal value, the more open it is, and it is considered to have a regulatory function.

[0103] In this embodiment, the epigenetic group function annotation software may be DeepSATA software, etc. Exemplarily, the difference level of the functional annotation of the reference allele and the alternative allele in the DeepSATA software may be used to sort the SNPs. For example, the threshold value may be set to 32451, etc.

[0104] In some optional embodiments, before the source model is trained using the SNP site data of a specific population and its phenotype data to obtain a phenotype prediction model, the method further includes:

[0105] Based on the SNP site data, the SNP site data is converted into a processable information file using conversion software.

[0106] In this embodiment, this step can realize data collection and conversion, and convert the first SNP site data and the second SNP site data into information files that can be independently operated and processed. For example, plink software can be used to extract genotype data containing a list of functional SNP sites from the original vcf file, so that these genotype information can be run in subsequent methods or computer programs.

[0107] In some optional embodiments, the source model includes nine data processing levels, wherein the input values ​​of the first processing level are behavioral sample names, the columns of the input values ​​of the first processing level are SNP site data, the output values ​​of the last processing level are behavioral sample names, and the columns of the output values ​​of the last processing level are phenotypic trait values.

[0108] As an example, the rows of the first processing level in the nine-level data processing level can be samples, and its columns can be SNP site data; the rows of the output results of the ninth processing level can be samples, and the columns can be traits. For example, the number of features of the output values ​​of the nine-level data processing level is 90, 180, 360, 450, 540, 630, 720, 810, and 1 in the first to ninth levels, respectively.

[0109] As another example, the number of features extracted in the first eight layers is set to [90, 180, 360, 450, 540, 630, 720, 810], respectively, to ensure the continuous aggregation of effective information of functional SNP genotypes.

[0110] In some optional embodiments, between the last processing level and the first data processing level, the construction method further includes:

[0111] The training data is subjected to penalty correction of feature weights to remove redundant information and achieve association between input data and phenotypic trait values; the feature weight penalty correction is performed using the following formula:

[0112]

[0113] in Represents the bi-norm operation, and n represents the number of input data.

[0114] In some optional embodiments, after performing penalty correction on the feature weights of the training data (whole genome sequencing data), the construction method further comprises:

[0115] The training data is batch normalized, and the output data is corrected by a ReLU activation function.

[0116] Among them, batch standardization is performed by the following formula:

[0117]

[0118] Among them, mean(x) represents the mean of the input data values, and var(x) represents the variance of the input data values.

[0119] Among them, the ReLU activation function is performed by the following formula:

[0120]

[0121] In some optional embodiments, the feature numbers of the output values ​​of the nine-level data processing hierarchy are 90, 180, 360, 450, 540, 630, 720, 810, and 1 in the first to ninth levels respectively.

[0122] Figure 2 The hyperparameter combination of the breeding phenotype prediction method of the present application is shown. In the process of training the source model using the SNP site data of a specific population and its phenotype group data as training data to obtain a phenotype prediction model, different hyperparameter combinations of the training data have a great influence on the performance of the phenotype prediction model. In order to apply the potential functional first SNP site data and second SNP site data proposed in the present application, the present application conducts an optimal hyperparameter optimization to determine the hyperparameter combination finally used, so as to improve the accuracy of phenotypic prediction and help improve breeding efficiency.

[0123] First, different hyperparameter combinations are made for the learning rate, number of reference features, norm, dropout rate, and ratio of the first 8 levels to the number of reference features of the source model. Finally, 500 hyperparameter combinations are randomly selected for prediction performance evaluation, of which hyperparameter combinations 1 to 4 include 125 different random combinations. Secondly, from the results of the prediction method, two statistical indicators, such as median and average, are used to combine the optimal parameters. Finally, the number of votes for each hyperparameter to obtain the best performance is counted. The higher the number of votes, the better the effect, and the final optimal hyperparameter can be determined. Each box plot in the figure represents the prediction accuracy of 125 random combinations under a specific hyperparameter value. The yellow one represents the highest average value under the 125 random combinations, the purple one represents the highest median value under the 125 random combinations, and the pink one represents the highest average and median value under the 125 random combinations. Among them, the parameter value with the highest number of votes is determined as the optimal parameter value. For the hyperparameter of the drop rate, the drop rates of 0.2 and 0.1 both received 4 votes, but the overall prediction performance of the drop rate of 0.1 was higher than that of 0.2, so 0.1 was finally selected as the optimal drop rate. In this way, the optimal hyperparameter combination finally determined, that is, the final optimal hyperparameter combination used in this application, is: a learning rate of 0.01, a reference feature number of 90, a second norm of L2, a drop rate of 0.1, and the ratio of the first 8 levels to the reference feature number of 1:2:4:5:6:7:8:9. Considering that the number of features of the 9th level is 1 corresponding to the trait value, the number of features of the output value of the nine-level data processing level is 90, 180, 360, 450, 540, 630, 720, 810, 1 in the first to ninth levels, respectively.

[0124] In some optional embodiments, at least seven data processing levels are nine data processing levels. Functional variant sites are used as bottom-level labels, i.e., the first-level data processing level, and SNPs are encoded using digital coding, such as [0, 1, 2] corresponding to [AA, Aa, aa], where A is the reference allele and a is the alternative allele, and the SNP genotype matrix is ​​used as the input data of the first level.

[0125] There may be multiple layers of data processing methods between the first data processing level and the last data processing level, such as the second to eighth layers of data processing methods, which may be hidden layers for filtering redundant information and aggregating feature information.

[0126] For example, the breeding phenotype prediction method of the present application can be used for the soybean ZH13 dataset, which collects 100-grain weight trait values ​​of 1254 samples and whole genome sequencing data including about 30 million SNP sites. GWAS analysis is used to evaluate the significance level of all SNP sites. By setting the threshold P < 1.0e -24 , 47534 potential functional SNP sites were extracted using the above method of the embodiment of the present application. 1000 samples were randomly selected as the training population to train the deep learning model, and the phenotypic values ​​of the remaining 254 test samples were predicted. The prediction performance was evaluated by calculating the Pearson correlation coefficient between the predicted phenotypic values ​​and the true phenotypic values ​​of the test samples. In the comparison between the traditional classic statistical regression model rrBLUP and the breeding phenotype prediction method of the present application, the breeding method of the present application achieved an accuracy improvement of 17.9% and a running speed improvement of 42.6%. The results are as follows Figure 3 .

[0127] Figure 3 It represents a performance evaluation diagram for predicting the soybean 100-grain weight trait. It uses GWAS to screen functional SNP sites. The horizontal axis of the scatter plot in Figure A represents the observed true phenotypic values, and the vertical axis represents the predicted phenotypic values; the horizontal axis of the bar graph in Figure B represents the deep learning model constructed in this application, and the statistical model constructed using a statistical algorithm, the improvement of the deep learning model constructed in this application relative to the statistical model, and the vertical axis represents the Pearson correlation coefficient between the observed and predicted phenotypic values, the improvement range, and the running time. Among them, the deep learning model of this application can be understood as the breeding phenotype prediction method of this application.

[0128] Exemplarily, the breeding phenotype prediction method of the present application can be used for the Duroc pig data set, and the eye muscle depth trait values ​​of 1940 samples and the whole genome sequencing data of about 11 million SNP sites are collected. The ATAC data of Duroc tissues including fat, liver, muscle, heart, spleen, etc. are collected, and the openness of the quantitative SNPs of all SNP sites is sorted from large to small according to the difference between reference and alternative alleles using the DeepSATA model. In order to compare with the gene chip, the data of GeneSeekporcine 50K SNP was collected. Due to the difference in the reference genome version, the version conversion was performed through liftover, and finally 32451 gene chip SNP sites and 32451 functional SNP sites with regulatory properties were extracted. By randomly selecting 1700 samples as the training population, the deep learning model was trained, and the phenotypic value prediction was performed on the remaining 240 test samples, and the prediction performance was evaluated by calculating the Pearson correlation coefficient between the predicted phenotypic value and the true phenotypic value of the test sample. In the comparison between the traditional classic statistical regression model rrBLUP and the breeding phenotype prediction method of the present application, the breeding method of the present application achieved an accuracy improvement of 16.9% and an operating speed improvement of 36.5%. Figure 4 Among them, the deep learning model of the present application can be understood as the breeding phenotype prediction method of the present application.

[0129] Figure 4 It represents the performance evaluation diagram of Duroc pig eye muscle depth trait prediction. It uses DeepSATA to screen functional SNP sites. The horizontal axis of the scatter plot in Figure A represents the observed true phenotypic value, and the vertical axis represents the predicted phenotypic value; the horizontal axis of the bar graph in Figure B represents the deep learning model constructed by this application, and the statistical model constructed using the statistical algorithm, the improvement of the deep learning model constructed by this application relative to the statistical model, and the vertical axis represents the Pearson correlation coefficient between the observed and predicted phenotypic values, the improvement range and the running time. Among them, the deep learning model of this application can be understood as the breeding phenotype prediction method of this application.

[0130] The breeding phenotype prediction model based on whole genome data of machine learning provided in this application is illustrated as follows.

[0131] Please refer to, Figure 5 A schematic diagram of a breeding phenotype prediction device based on whole genome data of machine learning provided in an embodiment of the present application is shown. Figure 5 The breeding phenotype prediction device 400 of the whole genome data shown may include the following units:

[0132] The present application embodiment provides a breeding phenotype prediction device based on whole genome data of machine learning, the device comprising:

[0133] An acquisition module 410 is used to acquire SNP site data of any organism; the organism includes plants or animals;

[0134] The prediction module 420 is used to predict the phenotype of any organism based on the SNP site data of the organism and the phenotype prediction model.

[0135] In some optional embodiments, the prediction module 420 includes a first phenotype prediction module, which is used to construct the first phenotype prediction module, and includes:

[0136] The source model unit 430 is used to construct a source model, wherein the source model includes at least seven data processing levels; wherein each data processing level includes using formula (1) to calculate,

[0137]

[0138] in, represents the weight vector, Represents the input data vector represents the bias vector, n represents the number of input data, x i represents the digitally encoded input data of the i-th gene, W i Corresponding to x i The weight of B i Corresponding to x i The bias of , · represents the inner product operation, × represents the multiplication operation, ∑ represents the summation operation, and f(x) represents the output; the output data of the previous level corresponds to the input data of the next level; and The value of is updated according to the following formula:

[0139]

[0140] Among them, loss represents the overall loss value, y real represents the observed true phenotypic value, y predict represents the predicted phenotypic value, ||W|| 2 represents the bi-norm of all weights, η represents the learning rate, represents the partial derivative operation, W i0 Indicates the initialized W i The value of B i0 B represents the initialization i The last level of data includes the phenotypic trait values ​​of n input data;

[0141] The training unit 440 is used to train the source model using the SNP site data of a specific population and its phenotype data as training data to obtain a phenotype prediction model.

[0142] In some optional embodiments, the acquisition module 410 includes:

[0143] A first acquisition submodule 411 is used to acquire whole genome sequencing data of any organism;

[0144] The SNP site data generation submodule 412 is used to obtain the SNP site data of the organism according to the whole genome sequencing data.

[0145] In some optional embodiments, the SNP site data generation submodule 412 includes a first SNP site data unit 4121; the first SNP site data unit 4121 is used to obtain functional scores for single nucleotide polymorphisms in whole genome sequencing data.

[0146] In some optional embodiments, the first SNP site data unit 4121 includes:

[0147] An association module 41211 is used to perform genome-wide association analysis on the genome sequencing data and calculate the significance level of SNPs;

[0148] The first screening module 41212 is used to obtain first SNP site data with potential functionality according to the SNP significance level p value and the SNP preset significance level interval.

[0149] In some optional embodiments, the SNP site data generation submodule 412 includes a second SNP site data unit 4122; the second SNP site data unit 4122 sorts the SNPs by the difference level of the apparent group data to obtain second SNP site data matching the apparent group data.

[0150] In some optional embodiments, the second SNP site data unit 4122 includes:

[0151] The epigenomic data processing unit 41221 is used to process the epigenomic data using the epigenomic function annotation software to obtain the epigenomic SNP sites corresponding to the epigenomic data;

[0152] The SNP site sorting unit 41222 is used to sort the apparent SNP sites according to the openness of the quantitative SNP to obtain the apparent SNP sorting result;

[0153] The second screening module 41223 is used to obtain second SNP site data matching the epigenomic data according to the set threshold of the epigenomic functional annotation software and the epigenomic SNP sorting result.

[0154] In some optional embodiments, the prediction module 420 further includes:

[0155] The file conversion unit 450 converts the SNP site data into a processable information file based on the SNP site data using conversion software.

[0156] Figure 6 A schematic structural diagram of an embodiment of an electronic device provided by the present application is shown.

[0157] The electronic device may include a processor 801 and a memory 802 storing computer program instructions.

[0158] Specifically, the processor 801 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0159] The memory 802 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 802 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 802 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 802 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 802 is a non-volatile solid-state memory.

[0160] The memory may include a read-only memory (ROM), a random access memory (RAM), a disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform a breeding phenotype prediction method based on whole genome data based on machine learning according to one aspect of the present disclosure.

[0161] The processor 801 reads and executes computer program instructions stored in the memory 802 to implement any one of the breeding phenotype prediction methods based on whole genome data using machine learning in the above embodiments.

[0162] In one example, the electronic device may further include a communication interface 803 and a bus 810. Figure 6 As shown, the processor 801, the memory 802, and the communication interface 803 are connected via a bus 810 and communicate with each other.

[0163] The communication interface 803 is mainly used to implement communication between various units, devices, units and / or equipment in the embodiments of the present application.

[0164] Bus 810 includes hardware, software or both, and the parts of the detection equipment of abnormal behavior are coupled to each other.For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 810 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the application considers any suitable bus or interconnection.

[0165] In addition, the embodiments of the present application may be implemented by providing a computer storage medium having computer program instructions stored thereon; when the computer program instructions are executed by a processor, any breeding phenotype prediction method based on whole genome data of machine learning in the above embodiments is implemented.

[0166] An embodiment of the present application provides a computer program product, which is stored in a storage medium. The program product is executed by at least one processor to implement the various processes of any breeding phenotype prediction method based on whole genome data based on machine learning in the above embodiments, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0167] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0168] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0169] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0170] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0171] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the above-described systems, units and specific working processes of the units can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A breeding phenotype prediction method based on whole genome data of machine learning, characterized in that: The method comprises: Obtaining SNP site data of any organism, including plants or animals; Based on the SNP site data of the organism and the phenotype prediction model, the phenotype prediction of any organism can be achieved.

2. The method according to claim 1, characterized in that: The method for constructing the phenotype prediction model comprises: Construct a source model, wherein the source model includes at least seven data processing levels; wherein each data processing level includes calculating using formula (1), in, represents the weight vector, represents the input data vector, represents the bias vector, n represents the number of input data, x i represents the digitally encoded input data of the i-th gene, W i Corresponding to x i The weight of B i Indicates that it corresponds to x i The bias of , · represents the inner product operation, × represents the multiplication operation, ∑ represents the summation operation, and f(x) represents the output; the output data of the previous level corresponds to the input data of the next level; and The value of is updated according to the following formula: Among them, loss represents the overall loss value, y real represents the observed true phenotypic value, y predict represents the predicted phenotypic value, ||W|| 2 represents the bi-norm of all weights, η represents the learning rate, represents the partial derivative operation, W i0 Indicates the initialized W i The value of B i0 B represents the initialization i The last level of data includes the phenotypic trait values ​​of n input data; The source model is trained using the SNP site data of a specific population and its phenotype data as training data to obtain a phenotype prediction model.

3. The method according to claim 1, characterized in that The step of obtaining SNP site data of any organism comprises: Obtain whole genome sequencing data for any organism; Based on the whole genome sequencing data, the SNP site data of the organism is obtained.

4. The method according to claim 1, characterized in that: The SNP site data includes first SNP site data; the first SNP site data is obtained by functionally scoring single nucleotide polymorphisms in whole genome sequencing data.

5. The method according to claim 4, characterized in that The first SNP site data is obtained by functionally scoring single nucleotide polymorphisms in whole genome sequencing data, including: Performing genome-wide association analysis on the genome sequencing data to calculate the significance level of SNP; According to the significance level p value of the SNP and the preset significance level interval of the SNP, the first SNP site data with potential functionality is obtained.

6. The method according to claim 1, characterized in that The SNP site data includes second SNP site data; the second SNP site data is obtained by sorting the SNPs according to the difference level of the epigenetic group data to obtain the second SNP site data matching the epigenetic group data.

7. The method according to claim 6, characterized in that The second SNP site data is obtained by sorting the SNPs according to the difference level of the epigenetic data to obtain the second SNP site data matching the epigenetic data, including: The epigenomic data were processed using epigenomic functional annotation software to obtain the epigenomic SNP sites corresponding to the epigenomic data; Sorting the apparent SNP sites according to the openness of the quantitative SNP to obtain an apparent SNP sorting result; According to the set threshold of the epigenomic functional annotation software and the epigenomic SNP ranking result, second SNP site data matching the epigenomic data are obtained.

8. The method according to claim 2, characterized in that: Before the source model is trained using the SNP site data of a specific population and its phenotype data to obtain a phenotype prediction model, the method further comprises: Based on the SNP site data, the SNP site data is converted into a processable information file using conversion software.

9. The method according to claim 2, characterized in that: The source model includes nine levels of data processing layers, wherein the behavioral sample names of the input values ​​of the first level of processing layers, the columns of the input values ​​of the first level of processing layers are SNP site data, the behavioral sample names of the output values ​​of the last level of processing layers, and the columns of the output values ​​of the last level of processing layers are phenotypic trait values.

10. The method according to claim 9, characterized in that Between the last processing level and the first data processing level, the construction method further comprises: The training data is subjected to penalty correction of feature weights to remove redundant information and achieve association between input data and phenotypic trait values; the feature weight penalty correction is performed using the following formula: in, Represents the bi-norm operation, and n represents the number of input data.

11. The method according to claim 10, characterized in that After the feature weights of the training data are corrected by penalty, the construction method further includes: The training data is batch normalized, and the output data is corrected by a ReLU activation function. Among them, batch standardization is performed by the following formula: Among them, mean(x) represents the mean of the input data values, and var(x) represents the variance of the input data values. Among them, the ReLU activation function is performed by the following formula:

12. The method according to claim 9, characterized in that The feature numbers of the output values ​​of the nine-level data processing hierarchy are 90, 180, 360, 450, 540, 630, 720, 810, and 1 in the first to ninth levels respectively.

13. A breeding phenotype prediction device based on whole genome data of machine learning, characterized in that: The device comprises: An acquisition module, used to acquire SNP site data of any organism; the organism includes plants or animals; The prediction module is used to predict the phenotype of any organism based on the SNP site data of the organism and the phenotype prediction model.

14. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method according to any one of claims 1 to 12 is implemented.

15. A readable storage medium, characterized in that: The readable storage medium stores a program or an instruction, and when the program or the instruction is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

16. A computer program product, characterized in that The computer program product is stored in a storage medium, and when the computer program product is executed by at least one processor, the method according to any one of claims 1 to 12 is performed.

Images