Method, device, equipment and medium for predicting hundred-grain weight of corn
By constructing a prediction model on a deep space-time network, combining deep residual network and two-way long and short-term memory recurrent neural network, the problem of low efficiency and accuracy in the prediction of corn 100 grain weight is solved, and more efficient corn breeding is achieved choose.
Patent Information
- Application Number
- CN202411107059.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-13
- Publication Date
- 2025-01-03
AI Technical Summary
The genome-wide selection method based on traditional statistics is low efficiency and accuracy in corn 100 grain weight prediction, making it difficult to effectively deal with complex nonlinear relationships between molecular markers and genotype and environmental interactions.
Deep spatiotemporal networks are adopted, combined with deep residual networks and bidirectional long and short-term memory recurrent neural networks, and iterative training is carried out through multi-dimensional genotype data and phenotype data to construct a prediction model to improve the prediction accuracy of 100-grain weight corn.
It significantly improves the efficiency and accuracy of corn 100-grain weight prediction, optimizes the breeding selection process, saves manpower and material costs, and improves the efficiency and accuracy of corn breeding.
Smart Images
Figure CN119252337A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of plant breeding, and in particular to a method, a device, equipment and a medium for predicting 100-grain weight of corn. Background Art
[0002] Hundred-grain weight is an extremely important corn breeding trait, which reflects the size and quality of seeds and has a direct impact on yield.
[0003] In recent years, GS (genomic selection) as a new breeding method has gradually attracted widespread attention. GS uses molecular markers or genomic data to predict individual performance, thereby screening individuals with excellent potential, playing an important role in increasing yield, improving disease resistance and quality improvement, and providing an efficient, accurate and sustainable solution for crop breeding. However, the GS method based on traditional statistics has limitations in dealing with complex nonlinear relationships between molecular markers and interactions between genotypes and environments, resulting in low efficiency and accuracy of corn breeding selection. Summary of the invention
[0004] The present invention provides a method, device, equipment and medium for predicting corn 100-kernel weight, aiming to solve the shortcomings of the whole genome selection method based on traditional statistics in the prior art in predicting corn 100-kernel weight, thereby improving the efficiency and accuracy of corn breeding selection. Through the technical means of the present invention, the efficiency and accuracy of corn 100-kernel weight prediction can be significantly improved, thereby optimizing the selection process of corn breeding.
[0005] The present invention provides a method for predicting 100-kernel weight of corn, comprising: Acquire multidimensional genotype data of corn materials to be predicted; According to the genotype frequency corresponding to the genotype data in each dimension, the genotype data in each dimension is converted into a genotype value; Inputting the multi-dimensional genotype values into a prediction model to obtain a predicted value of the 100-grain weight of the corn material to be predicted; Among them, the prediction model is obtained by iterative training based on the deep spatiotemporal network through multidimensional sample genotype data of sample corn materials of multiple different varieties and the 100-grain weight phenotypic data of the sample corn materials; the deep spatiotemporal network is constructed by fusing a deep residual network and a bidirectional long short-term memory recurrent neural network.
[0006] According to a method for predicting 100-kernel weight of corn provided by the present invention, the training process of the prediction model comprises the following steps: Based on genotyping sequencing, genotyping and typing are performed on each of the sample corn materials to obtain multi-dimensional sample genotyping data of each of the sample corn materials; Convert the sample genotype data of each dimension into a corresponding numerical form to obtain the sample genotype value of each dimension; Marking the hundred-grain weight of each corn material sample according to the multi-dimensional sample genotype data to obtain the hundred-grain weight phenotypic data of each corn material sample; Preprocessing the sample genotype values of each dimension of each sample corn material and the 100-grain weight phenotypic data; the preprocessing includes outlier removal, missing value filling and data standardization; The sample genotype values of each dimension after preprocessing are used as samples, and the 100-grain weight phenotype data after preprocessing are used as label values to construct a sample data set; According to the sample data set, the deep spatiotemporal network is iteratively trained to obtain a final prediction model.
[0007] According to a method for predicting 100-kernel weight of corn provided by the present invention, the step of preprocessing the sample genotype values of each dimension of each sample corn material comprises the following steps: In the multi-dimensional sample genotype values of each sample corn material, abnormal sample genotype values are obtained, and the abnormal sample genotype values are eliminated; the abnormal sample genotype values are sample genotype values whose minimum allele frequency is lower than the first frequency and whose missing rate is higher than the second frequency; In the multi-dimensional sample genotype values of each sample corn material, missing sample genotype values are obtained, and the missing sample genotype values are filled; the missing sample genotype values are sample genotype values with missing values.
[0008] According to a method for predicting 100-kernel weight of corn provided by the present invention, the step of preprocessing the 100-kernel weight phenotypic data of each sample corn material comprises the following steps: Performing data standardization on the 100-grain weight phenotype data of each sample corn material; Acquire abnormal phenotypic data from the hundred-grain weight phenotypic data of the plurality of sample corn materials, and remove the abnormal phenotypic data; the abnormal phenotypic data is the hundred-grain weight phenotypic data outside the range of plus or minus three times the standard deviation of the mean value; In the hundred-grain weight phenotypic data of the plurality of sample corn materials, missing phenotypic data are obtained, and the missing phenotypic data are filled; the missing phenotypic data are the hundred-grain weight phenotypic data with missing values.
[0009] According to a method for predicting 100-grain weight of corn provided by the present invention, the deep spatiotemporal network is iteratively trained according to the sample data set to obtain a final prediction model, including: Based on a preset split ratio, the sample data set is randomly split into a training data set and a validation data set; Iteratively training the deep spatiotemporal network according to the training data set to obtain a candidate model; According to the validation data set, the candidate model is validated to obtain the predicted value of 100-grain weight of each sample corn material in the validation data set; Determining the prediction performance of the candidate model according to the similarity between the predicted value of the 100-grain weight of each sample corn material in the validation data set and the 100-grain weight label; If the prediction performance does not meet the preset performance condition, return to the step of randomly dividing the sample data set into a training data set and a validation data set based on the preset split ratio until a candidate model whose prediction performance meets the preset performance condition is obtained; The prediction model is constructed according to the candidate models whose prediction performance meets the preset performance conditions.
[0010] According to a method for predicting 100-grain weight of corn provided by the present invention, the deep spatiotemporal network is iteratively trained according to the training data set to obtain a candidate model, including: For the current round of training, the training data set is divided into a plurality of batches of sub-training data sets; Based on multiple batches of sub-training data sets, the deep spatiotemporal network after the previous round of training is iteratively trained in multiple batches to obtain the deep spatiotemporal network after the current training; If the training round of the current training round has not reached the maximum round, and the deep spatiotemporal network after the current training has not reached convergence, return to the step of dividing the training data set into a plurality of batches of sub-training data sets, and iterate the next round of training until the training round reaches the maximum round and / or the deep spatiotemporal network after training converges; The candidate model is constructed according to the deep spatiotemporal network after each round of training.
[0011] According to a method for predicting 100-grain weight of corn provided by the present invention, the deep spatiotemporal network includes an input layer, a residual network layer, a forward long short-term memory network layer, a reverse long short-term memory network layer, at least one fully connected layer and an output layer; The residual network layer includes multiple convolutional layers, batch normalization layers, pooling layers and multiple residual blocks.
[0012] The present invention also provides a device for predicting the 100-kernel weight of corn, comprising: An acquisition unit, used for acquiring multi-dimensional genotype data of the corn material to be predicted; A conversion unit, used for converting the genotype data of each dimension into a genotype value according to the genotype frequency corresponding to the genotype data of each dimension; A prediction unit, used for inputting the multi-dimensional genotype values into a prediction model to obtain a predicted value of 100-grain weight of the corn material to be predicted; Among them, the prediction model is obtained by iterative training based on the deep spatiotemporal network through multidimensional sample genotype data of sample corn materials of multiple different varieties and the 100-grain weight phenotypic data of the sample corn materials; the deep spatiotemporal network is constructed by fusing a deep residual network and a bidirectional long short-term memory recurrent neural network.
[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for predicting 100-grain weight of corn as described above is implemented.
[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for predicting the 100-grain weight of corn.
[0015] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described methods for predicting 100-kernel weight of corn.
[0016] The method, device, equipment and medium for predicting the 100-grain weight of corn provided by the present invention obtain a prediction model by training on the basis of a deep spatiotemporal network, and can fully learn and capture the nonlinear relationship between the multidimensional genotype data of corn materials of different strains, as well as the nonlinear relationship between the multidimensional genotype data and the 100-grain weight data. It can be applied to the prediction of the 100-grain weight of a wider range of traits and species, so as to effectively solve the limitations in the nonlinear relationship between complex molecular markers and the interaction between genotype and environment, optimize the prediction performance of the 100-grain weight of corn, thereby significantly improving the efficiency and accuracy of the prediction of the 100-grain weight of corn, and further improving the efficiency and accuracy of the corn breeding selection. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0018] Figure 1 The present invention is a schematic flow chart of a method for predicting 100-grain weight of corn.
[0019] Figure 2 It is a schematic diagram of the structure of the deep spatiotemporal network provided by the present invention.
[0020] Figure 3 It is a simulation schematic diagram of the method for predicting 100-grain weight of corn provided by the present invention.
[0021] Figure 4 It is a schematic diagram of the structure of the device for predicting the 100-grain weight of corn provided by the present invention.
[0022] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0023] 100-grain weight is an extremely important corn breeding trait, which reflects the size and quality of seeds and has a direct impact on yield. Although conventional breeding methods have achieved results to a certain extent, these methods have certain limitations due to low selection efficiency and long cycles. In recent years, whole genome selection (GS) has gradually attracted widespread attention as a new breeding method. GS uses molecular markers or genomic data to predict individual performance and thus screen out individuals with excellent potential. Compared with traditional methods, GS has obvious advantages in improving breeding efficiency and accelerating breeding speed, so it is widely used in modern breeding practices.
[0024] Although GS plays an important role in increasing yield, improving disease resistance and improving quality, and provides an efficient, accurate and sustainable method for crop breeding, GS methods based on traditional statistics, such as rrBLUP (Ridge Regression Best Linear Unbiased Prediction) and BayesA (Bayesian Method A), have certain limitations in dealing with complex nonlinear relationships between molecular markers and interactions between genotypes and environments, resulting in low efficiency and accuracy in the selection of maize 100-grain weight traits.
[0025] In recent years, with the rapid development of deep learning, especially ResNET (Residual Neural Network), CNN (Convolutional Neural Network), and LSTM (Long Short-Term Memory networks), they have shown great potential in the field of crop breeding and have been applied to weather forecasting, crop growth models, and pest and disease prediction.
[0026] Therefore, in view of the low accuracy of the whole genome selection of corn 100-grain weight traits by the current traditional breeding methods, the method provided in this embodiment collects genotype data and phenotypic data for the 100-grain weight of crops, and constructs and trains a prediction model for predicting corn 100-grain weight based on a deep spatiotemporal network that integrates a deep residual network and a bidirectional long short-term memory recurrent neural network. The trained prediction model can be used to predict the 100-grain weight of corn materials with different genotypes, so as to effectively improve the prediction accuracy and efficiency, thereby saving manpower and material costs, improving the efficiency and accuracy of corn breeding selection, promoting the improvement of crop yield and quality, and promoting the sustainable development of agriculture.
[0027] It should be noted that the execution subject of the method may be an electronic device, a component in an electronic device, an integrated circuit, or a chip. The electronic device may be a mobile electronic device or a non-mobile electronic device. For example, the mobile electronic device may be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, an ultra-mobile personal computer, etc., and the non-mobile electronic device may be a server, a network attached storage, a personal computer, etc., which is not specifically limited in the present invention.
[0028] Figure 1 A schematic flow chart of a method for predicting 100-grain weight of corn provided by the present invention, the method comprising step 110, step 120, and step 130.
[0029] Step 110, obtaining multi-dimensional genotype data of the corn material to be predicted.
[0030] The so-called corn material to be predicted is the corn material required for prediction of 100-grain weight in breeding selection, which may be a sample material containing one or more corn varieties for seed selection.
[0031] Optionally, in the breeding selection process, the sample materials of all strains in the predicted corn material can be genotyped and genotyped based on genotyping-by-Sequencing (GBS) to obtain multidimensional genotype data of the predicted corn material. The so-called multidimensional genotype data is formed by multiple different point SNP (Single Nucleotide Polymorphism) molecular markers, such as 5 point SNPs, namely SNP1, SNP2, SNP3, SNP4 and SNP5, and the marker types of the two alleles of the SNP molecule at each point can include XX, XY, YY, where XX corresponds to pure and type 1, the two alleles are the same; XY corresponds to heterozygous type, the two alleles are different; YY corresponds to pure and type 2, the two alleles are the same.
[0032] Step 120, converting the genotype data in each dimension into a genotype value according to the genotype frequency corresponding to the genotype data in each dimension.
[0033] Optionally, after obtaining the multi-dimensional genotype data of the corn material to be predicted, the genotype frequency corresponding to each dimensional genotype data can be counted, and each dimensional genotype data can be converted into a numerical type recognizable by the model according to the frequency to obtain the corresponding genotype numerical vector. For example, for the SNP genotype data {XX, XY, YY} of a certain point, the genotype frequency corresponding to XX is 0.29, the genotype frequency corresponding to XY is 0.61, and the genotype frequency corresponding to YY is 0.1, then the genotype data {XX, XY, YY} can be converted to a corresponding genotype value of {0.29, 0.61, 0.1}, that is, the genotype value corresponding to XX is 0.29, the genotype value corresponding to XY is 0.61, and the genotype value corresponding to YY is 0.1.
[0034] Step 130, input the multidimensional genotype value into the prediction model to obtain the predicted value of the 100-grain weight of the corn material to be predicted; wherein, the prediction model is obtained by iteratively training the multidimensional sample genotype data of sample corn materials of multiple different strains and the 100-grain weight phenotypic data of the sample corn materials on the basis of a deep spatiotemporal network; the deep spatiotemporal network is constructed by fusing a deep residual network and a bidirectional long short-term memory recurrent neural network.
[0035] Optionally, after obtaining the multidimensional genotype values of the corn material to be predicted, the multidimensional genotype values can be directly input into a prediction model pre-trained based on a deep spatiotemporal network, or the multidimensional genotype values can be pre-processed, such as at least one of normalization, outlier removal, and missing value filling, and then input into the prediction model as input information to improve the data quality input into the prediction model, thereby improving the prediction accuracy of the 100-grain weight prediction of the corn material.
[0036] Subsequently, nonlinear mapping learning is performed on the multidimensional genotype values based on the prediction model to predict the output value of the 100-grain weight of the corn material to be predicted.
[0037] Furthermore, breeding selection can be performed according to the predicted value of 100-grain weight of the corn material to be predicted, such as sorting or classifying the corn materials according to the predicted 100-grain weight value, so as to select corn materials with high predicted value of 100-grain weight, strong growth ability and good quality for breeding, without the need for corn cultivation, saving manpower and material costs, shortening the breeding cycle and improving breeding efficiency.
[0038] The prediction model here can be pre-trained on the basis of a deep spatiotemporal network, and the specific training steps include: collecting multi-dimensional sample genotype data of sample corn materials of multiple different strains, and 100-grain weight phenotypic data of the sample corn materials; then, constructing samples with the multi-dimensional sample genotype data of the sample corn materials, and constructing labels with the 100-grain weight phenotypic data of the sample corn materials, to form a sample data set; Next, the deep spatiotemporal network was optimized and trained based on the sample data set, so as to construct a prediction model based on the trained deep spatiotemporal network, which can accurately output the predicted value of the 100-grain weight of the corn material based on the multi-dimensional genotype data of the corn material.
[0039] Here, the optimization training can be based on the sample data set, and all layers of the network in the deep spatiotemporal network are trained as a whole; or the layers of the network in the deep spatiotemporal network are controlled to participate in the model training in a staged frozen fine-tuning manner, such as first fixing the parameters of a part of the network layer, optimizing the parameters of the other part of the network layer, and then fixing the other part of the network layer after training, and optimizing the parameters of the part of the network layer; in addition, during the training process, joint training or double-layer loop training can be performed. Among them, joint training can be combined with all sample data sets in each iteration to jointly train the bidirectional long short-term memory network; the double-layer loop training includes an outer loop of the overall training data set and an inner loop of the batch sub-training data set, that is, the outer loop controls the number of training rounds, and the inner loop iterates each batch of sub-training data sets in the training data set.
[0040] The method provided by the embodiment of the present invention obtains a prediction model through training based on a deep spatiotemporal network, and can fully learn and capture the nonlinear relationship between the multidimensional genotype data of corn materials of different strains, as well as the nonlinear relationship between the multidimensional genotype data and the 100-grain weight data. It can be applied to the prediction of 100-grain weight of a wider range of traits and species, so as to effectively solve the limitations in the nonlinear relationship between complex molecular markers and the interaction between genotype and environment, optimize the prediction performance of corn 100-grain weight, thereby significantly improving the efficiency and accuracy of corn 100-grain weight prediction, and further improving the efficiency and accuracy of corn breeding selection.
[0041] In some embodiments, the training process of the prediction model includes the following steps: based on genotyping sequencing, genotyping and typing of each sample corn material are performed to obtain multidimensional sample genotyping data of each sample corn material; each dimension of the sample genotyping data is converted into a corresponding numerical form to obtain each dimension of the sample genotyping value; 100-grain weight is marked according to the multidimensional sample genotyping data of each sample corn material to obtain the 100-grain weight phenotypic data of each sample corn material; each dimension of the sample genotyping value and the 100-grain weight phenotypic data of each sample corn material are preprocessed; the preprocessing includes outlier removal, missing value filling and data standardization; the preprocessed sample genotyping values of each dimension are used as samples, and the preprocessed 100-grain weight phenotypic data are used as label values to construct a sample data set; according to the sample data set, the deep spatiotemporal network is iteratively trained to obtain a final prediction model.
[0042] The following describes the training steps of the prediction model in specific steps.
[0043] Step 210 , obtaining a sample data set, specifically includes steps 211 , 212 , 213 , 214 , 215 and 216 .
[0044] Step 211, using GBS to perform genotype identification and genotyping on sample corn materials of all varieties to obtain multi-dimensional sample genotype data of each sample corn material; the multi-dimensional sample genotype data of all sample corn materials can form a matrix, including m corn materials and n molecular markers (that is, n-dimensional sample genotype data), forming an m×n matrix data; illustratively, the matrix formed by the genotype data is shown in Table 1.
[0045] Among them, the ID in Table 1 is the identification of the sample corn material, and SNP1, SNP2, SNP3, SNP4 and SNP5 are the multidimensional genotype data of the sample corn material.
[0046]
[0047] Step 212, the data digitization stage converts the sample genotype data of each dimension into a corresponding frequency value according to each SNP molecular marker, that is, the genotype frequency corresponding to the sample genotype data of each dimension, to obtain the corresponding sample genotype value.
[0048]
[0049] Step 213, preprocessing the sample genotype values of each dimension, including outlier removal, missing value filling and data standardization, to obtain high-quality sample data.
[0050] Step 214, marking the 100-grain weight according to the multidimensional genotype data of each sample corn material to obtain the 100-grain weight phenotypic data of each sample corn material. The 100-grain weight phenotypic data of all sample corn materials can form a matrix, including m corn materials and 1 phenotypic data, forming an m×1 data table; illustratively, the 100-grain weight phenotypic data table is shown in Table 2.
[0051] Among them, ID in Table 2 is the identification of the sample corn material, and GW100 is the phenotypic data of the hundred-grain weight of the sample corn material.
[0052] Step 215 , preprocessing each hundred-grain weight phenotypic data, including outlier removal, missing value filling and data standardization, to obtain high-quality sample labels.
[0053] Step 216, taking the sample genotype values of each dimension after preprocessing as samples, and taking the preprocessed 100-grain weight phenotype data as label values, thereby constructing a sample data set.
[0054] It should be noted that the methods of filling missing values include but are not limited to filling the average value of the same characteristic data in this column, or using a deep learning-based method to fill in the data. The so-called data standardization methods include but are not limited to range standardization, Z-score standardization, linear ratio standardization, log function standardization, inverse tangent function standardization, etc.
[0055] Step 220, establishing a deep learning network framework with parameter initialization using a deep spatiotemporal network.
[0056] In some embodiments, the deep spatiotemporal network includes an input layer, a residual network layer, a forward long short-term memory network layer, a reverse long short-term memory network layer, at least one fully connected layer and an output layer; the residual network layer includes multiple convolutional layers, batch normalization layers, pooling layers and multiple residual blocks.
[0057] Figure 2 Schematic diagram of the structure of the deep spatiotemporal network provided by the present invention. Figure 2As shown in the figure, the deep spatiotemporal network includes an input layer, a deep residual network and a bidirectional long short-term memory recurrent neural network, as well as at least one fully connected layer and an output layer; wherein the deep residual network is constructed by at least one residual network layer; the bidirectional long short-term memory recurrent neural network includes a forward long short-term memory network layer and a reverse long short-term memory network layer. Among them, the input layer (Input Layer) is the first layer of the neural network, which is used to receive and transmit input data to subsequent layers in the network. After being processed by the residual network layer (ResidualNetwork, ResNet layer), the forward LSTM layer and the reverse LSTM layer in sequence, and then passing through the fully connected layer, the prediction result is finally obtained in the output layer, which effectively enhances the model's ability to process time series data.
[0058] Optionally, a ResNet layer is added in front of the deep spatiotemporal network (also known as a deep structure model of spatiotemporal relationship), and the ResNet layer includes multiple convolutional layers, batch normalization layers, pooling layers, and multiple residual blocks. Exemplarily, the ResNet layer includes two convolutional layers {Conv1, Conv2}, a batch normalization (Batch Normalization, BN) layer, a rectified linear unit activation function (Rectified Linear Unit, ReLU) activation function layer, a maximum pooling layer {Maxpool1}, and n residual blocks {Resblock1, Resblock2, , Resblockn}. Among them, the role of the convolution layer in the ResNet layer is mainly to extract features, reduce and increase dimensions, perform nonlinear transformations, reduce parameters and computational complexity, and maintain spatial structure. These roles together ensure the high efficiency and powerful feature learning ability of the ResNet layer. The role of the BN layer in deep learning includes accelerating training, stabilizing gradient flow, reducing dependence on initialization, providing regularization effects, improving model generalization ability, and eliminating covariate shift. In the ResNet layer, the main functions of the ReLU activation function layer include introducing nonlinearity, solving the gradient vanishing problem, improving computational efficiency, achieving sparse activation, and easy optimization. The role of the pooling layer is to reduce the spatial dimension of data features by downsampling, reduce computational complexity, and enhance the invariance and robustness of features, reduce noise and overfitting risks, thereby enhancing the generalization ability and stability of the model.
[0059] Among them, the calculation formula of the convolution layer in the ResNet network structure is: Y=f(W X+b); Among them, X is the input feature, Y is the output feature, W is the convolution kernel, and b is the bias term. Optionally, the deep spatiotemporal network also adds memory units, input gates, forget gates, and output gates on the basis of traditional neural networks, so that the model can selectively save information, thereby realizing the simultaneous memory of long-term and short-term information. Its structure is as follows Figure 2 As shown, it includes 1 forward LSTM layer and 1 reverse LSTM layer.
[0060] A fully connected layer is a type of layer in a neural network. Its characteristic is that each neuron in this layer is connected to all neurons in the previous layer. In a fully connected layer, each neuron has a weight and a bias, which are learned through training data. These weights and biases are constantly adjusted during the training process to minimize the loss function. In order to increase the expressive power of the network and prevent overfitting, activation functions (such as ReLU) and regularization techniques (such as Dropout) are often added after the fully connected layer.
[0061] The output layer is the last layer in the neural network. Its main function is to convert the output of the fully connected layer into the final prediction result of the model.
[0062] Among them, the calculation formula of the bidirectional LSTM in the deep spatiotemporal network is: ; ; in, is the hidden state of the bidirectional LSTM (Bidirectional Long Short-Term Memory, BiLSTM) network, is a maize molecular marker feature vector of length n, that is, a feature vector formed by extracting the n-dimensional genotype value through the ResNet layer. is the output value at the previous moment, is the predicted value of 100-grain weight of corn, is the weight matrix connecting the deep spatiotemporal network and the output layer, is the bias vector of the output layer.
[0063] Among them, the choice of long short-term memory by the forward LSTM unit is mainly determined by the forget gate, input gate and output gate, which can effectively capture the long-term dependencies in the time series.
[0064] The calculation formula of the forget gate is as follows: ; In the formula, is the output of the forget gate at the current moment, with a value range of 0 to 1, 0 means completely relevant information, and 1 means completely retaining relevant information; σ is the sigmoid activation function; is the input value at the current moment; is the output value at the previous moment; is the weight value of the forget gate; is the bias coefficient of the forget gate.
[0065] The calculation formula of the input gate is as follows: ); In the formula, is the output of the input gate at the current moment, with a value range of 0 to 1, where 0 means completely forgetting the relevant information and 1 means completely retaining the relevant information; The sigmoid function is used to control the importance of information; is the weight coefficient of the input gate; is the bias term of the input gate.
[0066] The calculation formula for candidate memory unit information is as follows: ); is the weight value of the candidate information; is the bias term of candidate information; The hyperbolic tangent activation function is used to create a new candidate value.
[0067] The calculation formula for the new memory cell information is as follows: ; is the new memory unit information at the current moment; is the memory unit information of the previous moment; The output of the forward LSTM unit is calculated as follows: ; ; In the formula, is the sigmoid function, is the initial output at the current moment, is the weight value of the initial output; is the bias term of the initial output; is the output at the current moment.
[0068] Similarly, the calculation process of the reverse LSTM can refer to the calculation process of the forward LSTM, but the calculation order is from time t to time t-1.
[0069] For the fully connected layer, the calculation formula is: ; ; In the formula, is the output at the current moment, is the weight value; is the bias term, is the predicted value of 100-grain weight predicted by the model.
[0070] In step 230, the multidimensional genotype values in the sample data set obtained in step 210 are used as model inputs, and the 100-grain weight phenotype data are used as sample labels, and the bidirectional long short-term memory network constructed in step 220 is iteratively trained to obtain a prediction model.
[0071] Optionally, during the training process, stochastic gradient descent (SGD) can be used to optimize the model, aiming to obtain a training model with the minimum prediction loss.
[0072] The method provided in this embodiment is based on the detection data of a certain time period, and preprocesses to improve the data quality. A prediction model is constructed using a deep spatiotemporal network, and the model is trained using a gradient descent algorithm and a forward and backward propagation algorithm. Through repeated iterations, the model parameters are continuously updated until the loss model is minimized, and an optimized prediction model for the 100-grain weight trait of corn is obtained, so that the prediction model can accurately and efficiently predict the 100-grain weight trait of corn of different strains in a specific time period, thereby saving manpower and material costs.
[0073] In some embodiments, the step of preprocessing the sample genotype values of each dimension of each of the sample corn materials includes the following steps: obtaining abnormal sample genotype values in the multi-dimensional sample genotype values of each of the sample corn materials, and eliminating the abnormal sample genotype values; the abnormal sample genotype values are sample genotype values whose minimum allele frequency is lower than the first frequency and whose missing rate is higher than the second frequency; obtaining missing sample genotype values in the multi-dimensional sample genotype values of each of the sample corn materials, and filling the missing sample genotype values; the missing sample genotype values are sample genotype values whose values are missing.
[0074] Optionally, during the genotype value preprocessing process, the minimum allele frequency and the missing rate of each genotype value can be compared with the first frequency and the second frequency respectively, so as to obtain sample genotype values with a minimum allele frequency lower than the first frequency and a missing rate higher than the second frequency and mark them as abnormal sample genotype values to be removed. The so-called first frequency and second frequency can be configured according to business needs, such as the first frequency is 0.05 and the second frequency is 0.1, that is, the sample genotype values with a minimum allele frequency lower than 0.05 and a missing rate higher than 0.1 are marked as abnormal sample genotype values.
[0075]
[0076] The sample genotype values with missing values can be filled by the average value method under the molecular marker. As shown in Table 3, it is an m×n matrix formed after the genotype data shown in Table 1 are converted and preprocessed. For example, the value of SNP5 of the sample corn material with ID Z001E0008 is missing. By averaging all the values under the molecular marker SNP5, the average value is 0.04. The average value 0.04 is filled into the value of SNP5 of the sample corn material Z001E0008, and the value filling result of SNP5 of the sample corn material Z001E0008 can be obtained.
[0077] The method provided in this embodiment eliminates the interference of outliers on model training by removing outliers, thereby improving the accuracy of prediction; by filling in missing values, the data set can be made more complete, thereby improving the accuracy of prediction, thereby enhancing the robustness of the model and the accuracy of prediction.
[0078] In some embodiments, the step of preprocessing the phenotypic data of 100-grain weight of each of the sample corn materials comprises the following steps: performing data standardization on the phenotypic data of 100-grain weight of each of the sample corn materials; obtaining abnormal phenotypic data from the phenotypic data of the 100-grain weight of multiple sample corn materials, and eliminating the abnormal phenotypic data; the abnormal phenotypic data is the phenotypic data of 100-grain weight that is outside the range of plus or minus three times the standard deviation of the mean; obtaining missing phenotypic data from the phenotypic data of the 100-grain weight of multiple sample corn materials, and filling the missing phenotypic data; the missing phenotypic data is the phenotypic data of 100-grain weight that is missing a numerical value.
[0079] Optionally, data standardization can be implemented by using maximum and minimum normalization, inverse tangent function normalization, etc. to ensure the consistency of data dimensions, thereby reducing the overfitting phenomenon of the model and improving the generalization ability of the model. The following describes data standardization using maximum and minimum normalization as an example.
[0080] The calculation formula for the normalization of the maximum and minimum values is: ; In the formula, is the normalized 100-grain weight phenotypic data; and They are respectively the minimum and maximum values of the 100-grain weight phenotypic data of all sample corn materials.
[0081] For outlier elimination, the average value of the 100-grain weight phenotypic data of all sample corn materials can be obtained, and the range of plus or minus three times the standard deviation of the average value is used as the reference range. The 100-grain weight phenotypic data outside the range of plus or minus three times the standard deviation (2σ) of the average value is marked as abnormal phenotypic data, which is eliminated and regarded as missing phenotypic data for filling. The filling processing method of missing phenotypic data can refer to the filling processing method of missing genotype values.
[0082] It should be noted that when abnormal phenotypic data are eliminated, the corresponding genotype values also need to be eliminated synchronously and then filled as missing genotype values.
[0083] As shown in Table 4, it is an m×1 matrix formed after preprocessing the 100-grain weight phenotypic data shown in Table 2.
[0084]
[0085] In some embodiments, the deep spatiotemporal network is iteratively trained according to the sample data set to obtain a final prediction model, including: based on a preset split ratio, randomly splitting the sample data set into a training data set and a verification data set; based on the training data set, iteratively training the deep spatiotemporal network to obtain a candidate model; based on the verification data set, verifying the candidate model to obtain a predicted value of the 100-grain weight of each sample corn material in the verification data set; determining the prediction performance of the candidate model based on the similarity between the predicted value of the 100-grain weight of each sample corn material in the verification data set and the 100-grain weight label; if the prediction performance does not meet the preset performance conditions, returning to the step of randomly splitting the sample data set into a training data set and a verification data set based on the preset split ratio until a candidate model whose prediction performance meets the preset performance conditions is obtained; and constructing the prediction model based on the candidate model whose prediction performance meets the preset performance conditions.
[0086] Optionally, the model training step in step 230 further includes: loading the sample data set obtained in step 210; randomly dividing the sample data set into a training data set and a validation data set according to a preset split ratio, so as to reduce the potential deviation between the data through random splitting and improve the generalization ability of the model. The so-called preset split ratio can be configured according to actual training needs, such as 80%.
[0087] Subsequently, the deep spatiotemporal network constructed in step 220 is initialized with parameter settings, including but not limited to the model structure, maximum training rounds, batch size, loss function, optimizer parameters, etc. For example, the input feature size input_size=1, the hidden layer feature size hidden_size=32, the number of hidden layers num_layers=1, the output feature size output_size=1, the batch size batch_size=20, the sequence length seqlen=21389, and the loss function loss_function: the mean square error loss function is used to measure the difference between the predicted value and the true value. Stochastic gradient descent (SGD) is used as the optimizer, the learning rate is set to 0.00005, the momentum is set to 0.90, and the weight decay is set to 0.0. The maximum number of training rounds (epochs) is set to 30. During model training, the nn.Flatten() function is used to flatten the output of the LSTM layer. The model contains a linear layer and uses ReLU as the activation function. In order to optimize the network, the SGD optimizer is used, and the mean square error (MSE) between the network output and the observed value is used as the training loss function. The loss function is used to evaluate the accuracy of the prediction model.
[0088] Among them, the mean square error The calculation formula is as follows: ; In the formula, is the number of sample corn materials, For the 100-grain weight label, is the predicted value of 100-grain weight, is the residual.
[0089] Subsequently, based on the training data set, the deep spatiotemporal network is iteratively trained to obtain a candidate model; based on the verification data set, the candidate model is verified, and the candidate model is evaluated according to the verification results, thereby obtaining a prediction model.
[0090] Optionally, during model evaluation, the similarity between the predicted value of 100-grain weight and the 100-grain weight label is calculated as the prediction performance, and the prediction performance of the candidate model is evaluated. If the similarity between the predicted value of 100-grain weight and the 100-grain weight label is less than a preset threshold, it is characterized that the prediction performance does not meet the preset performance conditions; at this time, the sample data set is randomly re-divided into a training data set and a verification data set based on a preset split ratio, and the model is iteratively trained and evaluated using the re-divided training data set and verification data set until the similarity is greater than or equal to the preset threshold, which means that the prediction performance meets the preset performance conditions; at this time, the candidate model whose prediction performance meets the preset performance conditions is used as the prediction model.
[0091] Optionally, the similarity calculation may be based on the following formula: ; In the formula, is the similarity, n is the number of sample corn materials, is the predicted value of 100-grain weight of the i-th sample corn material, represents the 100-grain weight label of the i-th sample corn material, It represents the average value of the predicted 100-grain weight of all sample corn materials in the validation data set. Represents the average of the 100-kernel weight labels of all sample corn materials in the validation dataset.
[0092] The method provided in this embodiment can make the trained prediction model not only have higher prediction performance but also have stronger generalization ability through double iterative training.
[0093] In some embodiments, the iterative training of the deep spatiotemporal network according to the training data set to obtain a candidate model includes: for the current round of training, dividing the training data set into multiple batches of sub-training data sets; performing multiple batches of iterative training on the deep spatiotemporal network after the previous round of training based on the multiple batches of sub-training data sets to obtain the deep spatiotemporal network after the current training; when the training rounds of the current round of training have not reached the maximum rounds and the deep spatiotemporal network after the current training has not reached convergence, returning to the step of dividing the training data set into multiple batches of sub-training data sets, and iterating the next round of training until the training rounds reach the maximum rounds and / or the trained deep spatiotemporal network converges; constructing the candidate model according to the deep spatiotemporal network after each round of training.
[0094] Optionally, during the model training process, a double-layer loop training method can be used to implement the training of the deep spatiotemporal network, that is, an outer loop of the overall training data set and an inner loop of the batch sub-training data set are used for training.
[0095] For the outer loop: control the number of training rounds. In each round of training, check whether the maximum number of rounds has been reached. If not, continue with the next round of iterative training.
[0096] Inner loop: In each round, the training dataset is divided into multiple sub-training datasets (batches) according to the preset batch size. Multiple batches of deep spatiotemporal networks trained in the previous round are trained using multiple sub-training datasets. In each batch iteration, the network updates its weights and biases based on the sub-training datasets of the current batch to minimize the difference between the predicted value and the true value.
[0097] If the training round of the current training has not reached the maximum round, and the deep spatiotemporal network after the current training has not reached convergence, such as the validation loss value of multiple consecutive rounds no longer decreases significantly, return to the step of dividing the training data set into multiple batches of sub-training data sets and start the next round of training. If the training round reaches the maximum round or the network converges, the training process is terminated.
[0098] After all rounds of training are completed, we select the model with the best performance as the candidate model based on the performance of the deep spatiotemporal network after each round of training.
[0099] The following describes the model training steps provided in this embodiment with specific examples.
[0100] As shown in Table 5, the model performance of each iterative training is described by setting the maximum number of training rounds to 30 and using the prediction accuracy R as the model performance indicator.
[0101] At the same time, if Figure 3 As shown in the figure, a scatter plot of the predicted results and the true results on the validation dataset is drawn; where test_Y is the true result on the validation dataset, and test_pre is the predicted result on the validation dataset.
[0102]
[0103] The method provided in this embodiment successfully constructed a prediction model for 100-grain weight of corn materials based on a deep spatiotemporal network through the above double-layer loop training process. This model continuously optimizes its parameters during the training process to better fit the patterns in the training data, and finally shows good prediction performance on the validation data set.
[0104] The device for predicting the 100-kernel weight of corn provided by the present invention is described below. The device for predicting the 100-kernel weight of corn described below and the method for predicting the 100-kernel weight of corn described above can be referenced to each other.
[0105] like Figure 4As shown, the device for predicting the 100-grain weight of corn includes an acquisition unit 410, a conversion unit 420 and a prediction unit 430.
[0106] The acquisition unit 410 is used to acquire multi-dimensional genotype data of the corn material to be predicted.
[0107] The conversion unit 420 is used to convert the genotype data in each dimension into a genotype value according to the genotype frequency corresponding to the genotype data in each dimension.
[0108] The prediction unit 430 is used to input the multidimensional genotype values into the prediction model to obtain the predicted value of the 100-grain weight of the corn material to be predicted; wherein, the prediction model is obtained by iteratively training the multidimensional sample genotype data of sample corn materials of multiple different strains and the 100-grain weight phenotypic data of the sample corn materials on the basis of a deep spatiotemporal network; the deep spatiotemporal network is constructed by fusing a deep residual network and a bidirectional long short-term memory recurrent neural network.
[0109] The device provided by the embodiment of the present invention obtains a prediction model through training based on a deep spatiotemporal network, and can fully learn and capture the nonlinear relationship between the multidimensional genotype data of corn materials of different strains, as well as the nonlinear relationship between the multidimensional genotype data and the 100-grain weight data. It can be applied to the prediction of 100-grain weight of a wider range of traits and species, so as to effectively solve the limitations in the nonlinear relationship between complex molecular markers and the interaction between genotype and environment, optimize the prediction performance of corn 100-grain weight, thereby significantly improving the efficiency and accuracy of corn 100-grain weight prediction, and further improving the efficiency and accuracy of corn breeding selection.
[0110] The device provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.
[0111] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute a method for predicting the 100-grain weight of corn, the method comprising: obtaining multidimensional genotype data of a corn material to be predicted; converting the genotype data of each dimension into a genotype value according to the genotype frequency corresponding to the genotype data of each dimension; inputting the multidimensional genotype value into a prediction model to obtain a predicted value of the 100-grain weight of the corn material to be predicted; wherein the prediction model is obtained by iteratively training the multidimensional sample genotype data of sample corn materials of multiple different strains and the 100-grain weight phenotypic data of the sample corn materials on the basis of a deep spatiotemporal network; the deep spatiotemporal network is constructed by fusing a deep residual network and a bidirectional long short-term memory recurrent neural network.
[0112] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0113] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for predicting the 100-grain weight of corn provided by the above-mentioned methods, and the method includes: obtaining multidimensional genotype data of the corn material to be predicted; converting the genotype data of each dimension into a genotype value according to the genotype frequency corresponding to the genotype data of each dimension; inputting the multidimensional genotype value into a prediction model to obtain a predicted value of the 100-grain weight of the corn material to be predicted; wherein, the prediction model is obtained by iteratively training multidimensional sample genotype data of sample corn materials of multiple different strains and the 100-grain weight phenotypic data of the sample corn materials on the basis of a deep spatiotemporal network; the deep spatiotemporal network is constructed by fusing a deep residual network and a bidirectional long short-term memory recurrent neural network.
[0114] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the method for predicting the 100-grain weight of corn provided by the above-mentioned methods, the method comprising: obtaining multidimensional genotype data of the corn material to be predicted; converting the genotype data of each dimension into a genotype value according to the genotype frequency corresponding to the genotype data of each dimension; inputting the multidimensional genotype value into a prediction model to obtain a predicted value of the 100-grain weight of the corn material to be predicted; wherein the prediction model is obtained by iteratively training multidimensional sample genotype data of sample corn materials of multiple different strains and the 100-grain weight phenotypic data of the sample corn materials on the basis of a deep spatiotemporal network; the deep spatiotemporal network is constructed by fusing a deep residual network and a bidirectional long short-term memory recurrent neural network.
[0115] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0116] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for predicting 100-kernel weight of corn, characterized in that: include: Acquire multidimensional genotype data of corn materials to be predicted; According to the genotype frequency corresponding to the genotype data in each dimension, the genotype data in each dimension is converted into a genotype value; Inputting the multi-dimensional genotype values into a prediction model to obtain a predicted value of the 100-grain weight of the corn material to be predicted; Among them, the prediction model is obtained by iterative training based on the deep spatiotemporal network through multidimensional sample genotype data of sample corn materials of multiple different varieties and the 100-grain weight phenotypic data of the sample corn materials; the deep spatiotemporal network is constructed by fusing a deep residual network and a bidirectional long short-term memory recurrent neural network.
2. The method for predicting 100-kernel weight of corn according to claim 1, characterized in that: The training process of the prediction model includes the following steps: Based on genotyping sequencing, genotyping and typing are performed on each of the sample corn materials to obtain multi-dimensional sample genotyping data of each of the sample corn materials; Convert the sample genotype data of each dimension into a corresponding numerical form to obtain the sample genotype value of each dimension; Marking the hundred-grain weight of each corn material sample according to the multi-dimensional sample genotype data to obtain the hundred-grain weight phenotypic data of each corn material sample; Preprocessing the sample genotype values of each dimension of each sample corn material and the 100-grain weight phenotypic data; the preprocessing includes outlier removal, missing value filling and data standardization; The sample genotype values of each dimension after preprocessing are used as samples, and the 100-grain weight phenotype data after preprocessing are used as label values to construct a sample data set; According to the sample data set, the deep spatiotemporal network is iteratively trained to obtain a final prediction model.
3. The method for predicting 100-kernel weight of corn according to claim 2, characterized in that: The step of preprocessing the sample genotype values of each dimension of each sample corn material comprises the following steps: In the multi-dimensional sample genotype values of each sample corn material, abnormal sample genotype values are obtained, and the abnormal sample genotype values are eliminated; the abnormal sample genotype values are sample genotype values whose minimum allele frequency is lower than the first frequency and whose missing rate is higher than the second frequency; In the multi-dimensional sample genotype values of each sample corn material, missing sample genotype values are obtained, and the missing sample genotype values are filled; the missing sample genotype values are sample genotype values with missing values.
4. The method for predicting 100-kernel weight of corn according to claim 2, characterized in that: The step of preprocessing the 100-grain weight phenotype data of each sample corn material comprises the following steps: Performing data standardization on the 100-grain weight phenotype data of each sample corn material; Acquire abnormal phenotypic data from the hundred-grain weight phenotypic data of the plurality of sample corn materials, and remove the abnormal phenotypic data; the abnormal phenotypic data is the hundred-grain weight phenotypic data outside the range of plus or minus three times the standard deviation of the mean value; In the hundred-grain weight phenotypic data of the plurality of sample corn materials, missing phenotypic data are obtained, and the missing phenotypic data are filled; the missing phenotypic data are the hundred-grain weight phenotypic data with missing values.
5. The method for predicting 100-kernel weight of corn according to claim 2, characterized in that: The iterative training of the deep spatiotemporal network according to the sample data set to obtain a final prediction model includes: Based on a preset split ratio, the sample data set is randomly split into a training data set and a validation data set; Iteratively training the deep spatiotemporal network according to the training data set to obtain a candidate model; According to the validation data set, the candidate model is validated to obtain the predicted value of 100-grain weight of each sample corn material in the validation data set; Determining the prediction performance of the candidate model according to the similarity between the predicted value of the 100-grain weight of each sample corn material in the validation data set and the 100-grain weight label; If the prediction performance does not meet the preset performance condition, return to the step of randomly dividing the sample data set into a training data set and a validation data set based on the preset split ratio until a candidate model whose prediction performance meets the preset performance condition is obtained; The prediction model is constructed according to the candidate models whose prediction performance meets the preset performance conditions.
6. The method for predicting 100-kernel weight of corn according to claim 5, characterized in that: The iterative training of the deep spatiotemporal network according to the training data set to obtain a candidate model includes: For the current round of training, the training data set is divided into a plurality of batches of sub-training data sets; Based on multiple batches of sub-training data sets, the deep spatiotemporal network after the previous round of training is iteratively trained in multiple batches to obtain the deep spatiotemporal network after the current training; If the training round of the current training round has not reached the maximum round, and the deep spatiotemporal network after the current training has not reached convergence, return to the step of dividing the training data set into a plurality of batches of sub-training data sets, and iterate the next round of training until the training round reaches the maximum round and / or the deep spatiotemporal network after training converges; The candidate model is constructed according to the deep spatiotemporal network after each round of training.
7. The method for predicting 100-kernel weight of corn according to any one of claims 1 to 6, characterized in that: The deep spatiotemporal network includes an input layer, a residual network layer, a forward long short-term memory network layer, a reverse long short-term memory network layer, at least one fully connected layer and an output layer; The residual network layer includes multiple convolutional layers, batch normalization layers, pooling layers and multiple residual blocks.
8. A device for predicting the weight of 100 kernels of corn, characterized in that: include: An acquisition unit, used for acquiring multi-dimensional genotype data of the corn material to be predicted; A conversion unit, used for converting the genotype data of each dimension into a genotype value according to the genotype frequency corresponding to the genotype data of each dimension; A prediction unit, used for inputting the multi-dimensional genotype values into a prediction model to obtain a predicted value of 100-grain weight of the corn material to be predicted; Among them, the prediction model is obtained by iterative training based on the deep spatiotemporal network through multidimensional sample genotype data of sample corn materials of multiple different varieties and the 100-grain weight phenotypic data of the sample corn materials; the deep spatiotemporal network is constructed by fusing a deep residual network and a bidirectional long short-term memory recurrent neural network.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for predicting the 100-kernel weight of corn as described in any one of claims 1 to 7 is implemented.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting 100-kernel weight of corn as described in any one of claims 1 to 7 is implemented.