A rice blast resistance prediction model, a prediction method and related equipment

Abstract

The application discloses a rice blast resistance prediction model, a prediction method and related equipment, and comprises the following steps: acquiring rice blast resistance phenotype and genotype information; using a correlation analysis method to calculate the correlation between all markers in the genotype information and the rice blast resistance phenotype, screening and retaining the genotype information meeting the correlation requirement, and constructing a data set of a rice blast resistance prediction model with the corresponding rice blast resistance phenotype; constructing a deep neural network model architecture, taking the genotype information as the input and the rice blast resistance phenotype as the output, training and optimizing the deep neural network model architecture by using the data set, and obtaining a trained rice blast resistance prediction model. The results show that the prediction model can significantly improve the prediction accuracy of rice blast resistance, and realizes effective prediction of rice blast resistance by using second-generation sequencing data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent detection, in particular to a rice blast resistance prediction model, a prediction method and related equipment. BACKGROUND

[0002] Rice blast is the most prevalent and devastating disease in rice industry. It usually causes 10-50% yield loss in rice field, and even absolute loss in severe cases. However, the identification of rice blast resistance phenotype requires inoculation of rice with Magnaporthe oryzae or planting in rice blast nursery, which makes it more difficult to obtain compared to other morphological characteristics of rice. With the development of next-generation sequencing technology, the cost of sequencing has been greatly reduced, and whole genome data can also be obtained at a lower cost. Therefore, developing a method to predict rice blast resistance based on genome will be more effective to assist breeders in rice breeding improvement.

[0003] Nowadays, human beings have entered the era of breeding concept of the trinity of Internet, big data and artificial intelligence, which emphasizes the cross integration of life science, information science and breeding science. Artificial intelligence is a branch of computer science, which aims to simulate human consciousness and thinking mode, as well as the transmission mode of neurons in the brain to process information. Machine learning method is the core of artificial intelligence application, which uses computer algorithm model to analyze data, and trains the model by constantly learning the characteristics of data, so as to predict and judge new data. In the field of breeding, machine learning uses high-density molecular markers obtained by sequencing technology, and its prediction accuracy is closely related to population size, genetic relationship and linkage disequilibrium degree. Compared with the previous molecular marker assisted breeding, whole genome selection breeding can use high-density genetic markers covering the whole genome to more accurately predict the breeding value of individuals. In machine learning, deep learning shows great potential. Its nonlinear neural network model architecture can better explain the non-additive genetic variation of crops. However, the current deep learning still faces some challenges in the popularization and application of crop whole genome selection and phenotype prediction: first, the population phenotype is difficult to obtain on a large scale; second, the marker density of genotype data is too large, and most studies focus on whole genome SNP markers.

[0004] Therefore, how to provide an effective rice blast resistance prediction model, prediction method and related equipment is a technical problem to be solved by those skilled in the art. SUMMARY

[0005] The present application aims at the above research status and existing problems, and provides a rice blast resistance prediction model, a prediction method and related equipment, proposes a method combining SNP, PAV and IOT, and reduces the genotype dimension through association analysis, so as to realize the prediction of the rice blast resistance phenotype based on the whole genome. With the continuous accumulation of a large amount of data and the continuous optimization of the model, deep learning has greater application potential in the prediction of the rice blast resistance phenotype.

[0006] The present application provides a construction method of a rice blast resistance prediction model, comprising the following steps:

[0007] S1: obtaining rice blast resistance phenotype and genotype information;

[0008] S2: using an association analysis method to calculate the correlation between all markers in the genotype information and the rice blast resistance phenotype, screening and retaining the genotype information meeting the correlation requirement, and constructing a data set of the rice blast resistance prediction model together with the corresponding rice blast resistance phenotype;

[0009] S3: constructing a deep neural network model architecture, taking the genotype information as the input and the rice blast resistance phenotype as the output, training and optimizing the deep neural network model architecture by using the data set, and obtaining a trained rice blast resistance prediction model.

[0010] Preferably, the genotype information is haplotype information integrated after SNP, PAV and IOT, and the obtaining step comprises: mounting to the pan-genome according to the second-generation sequencing data of the rice population, detecting SNP, PAV and IOT and integrating, and obtaining the population genotype.

[0011] Preferably, the output of the rice blast resistance prediction model is the grade of the rice blast resistance phenotype.

[0012] Preferably, the association analysis method comprises: using a gapit algorithm to perform correlation analysis on all markers in the genotype information and the rice blast resistance phenotype, screening all markers with a p-value greater than a set value in the association analysis result, and proposing the corresponding genotype information.

[0013] Preferably, it further comprises the step of one-hot encoding the genotype information in the data set:

[0014] The genotype of SNP includes A, T, C, G, and deletion or heterozygosity, which are respectively converted into five classification identifiers, and then one-hot encoding is performed;

[0015] The genotype of PAV includes two types of presence and absence, which are respectively converted into two classification identifiers, and then one-hot encoding is performed;

[0016] The genotypes of the IOT include two types of presence and absence, which are converted into two classification labels and then one-hot encoded;

[0017] The encoded genotype information is used as an input of the deep neural network model architecture.

[0018] Preferably, the input layer shape of the rice blast resistance prediction model is [n, m, 5], wherein n is the number of rice samples, m is the number of screened genotype markers, and 5 is the dimension of the one-hot encoded genotype.

[0019] Preferably, the rice blast resistance prediction model includes eight hidden layers.

[0020] The first hidden layer is a one-dimensional convolutional layer, the number of convolutional kernels is p, the size of the convolutional kernel is 3, and the activation function is relu.

[0021] The second hidden layer is a one-dimensional convolutional layer, the number of convolutional kernels is 2*p, the size of the convolutional kernel is 3, and the activation function is relu.

[0022] The third hidden layer is a flattening layer.

[0023] The fourth hidden layer is a random inactivation layer.

[0024] The fifth hidden layer is a fully connected layer, which includes 2*p neurons, and the activation function is relu.

[0025] The sixth hidden layer is a random inactivation layer.

[0026] The seventh hidden layer is a fully connected layer, which includes p neurons, and the activation function is relu.

[0027] The eighth hidden layer is a random inactivation layer.

[0028] The application also provides a rice blast resistance prediction method based on artificial intelligence, which includes the following steps:

[0029] S1: Collecting rice samples, sequencing the rice samples, identifying SNP, PAV and IOT genotypes, integrating and constructing a genotype matrix as genotype information.

[0030] S2: Based on the correlation between all markers in the genotype information and the rice blast resistance phenotype, screening and retaining genotype information that meets the correlation requirement.

[0031] S3: Inputting the genotype information meeting the correlation requirement into a rice blast resistance prediction model to predict the rice blast resistance phenotype and outputting the model prediction result; the rice blast resistance prediction model is the trained rice blast resistance prediction model.

[0032] The application also provides an electronic device, comprising:

[0033] one or more processors;

[0034] a storage device for storing one or more programs,

[0035] When the one or more programs are executed by the one or more processors, the one or more processors implement the artificial intelligence-based rice blast resistance prediction method.

[0036] The application also provides a computer readable medium having stored thereon a computer program, which, when executed by a processor, implements the artificial intelligence-based rice blast resistance prediction method.

[0037] The rice blast resistance prediction model, prediction method and related equipment provided by the application have the following beneficial effects compared with the prior art:

[0038] By combining the second-generation sequencing technology and the deep learning algorithm, the application constructs an efficient rice blast resistance prediction model. The model uses the genomic data of the rice population, filters out SNP (single nucleotide polymorphism), PAV (large fragment deletion or presence) and IOT (transposon or reverse mutation) markers highly related to rice blast resistance through association analysis, and performs one-hot encoding processing on these markers. Finally, by training a deep neural network model, the precise prediction of rice blast resistance is realized.

[0039] The application uses the association analysis method to calculate the correlation between all markers in the genotype and the rice blast resistance phenotype, and removes the markers completely unrelated to the phenotype, thereby effectively reducing the dimensionality of the genotype data. This process not only reduces the consumption of computing resources, but also improves the training efficiency and prediction performance of the model.

[0040] With the development of second-generation sequencing technology, it has become easier and more economical to obtain genomic data. The genotype information of rice can be determined by sequencing before planting, and the trained model can be used to quickly predict the rice blast resistance grade. This method greatly simplifies the resistance identification process, saves time and resources, and is especially suitable for large-scale breeding screening and variety improvement, and can quickly respond to the breeding needs of the big data era. Through continuous accumulation and optimization of data, the prediction ability of the model will be further improved, providing stronger support for future rice breeding. BRIEF DESCRIPTION OF DRAWINGS

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely embodiments of the present invention, and those skilled in the art can obtain other drawings based on the provided drawings without creative effort.

[0042] Figure 1 This is a flowchart of the construction process for the predictive model of rice blast resistance provided in this embodiment of the invention;

[0043] Figure 2 This is a Manhattan graph used for association analysis with a training set, as provided in an embodiment of the present invention;

[0044] Figure 3 This is a neural network structure diagram of the rice blast resistance prediction model provided in this embodiment of the invention;

[0045] Figure 4 This is a comparison chart of the prediction accuracy of the rice blast resistance prediction model provided in this embodiment of the invention before and after training. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] The first aspect of this invention discloses a method for constructing a predictive model for rice blast resistance, comprising the following steps:

[0048] S1: Obtain information on rice blast resistance phenotype and genotype;

[0049] S2: Use association analysis to calculate the correlation between all markers in the genotype information and the rice blast resistance phenotype, screen and retain the genotype information that meets the correlation requirements, and construct the dataset of the rice blast resistance prediction model with the corresponding rice blast resistance phenotype.

[0050] S3: Construct a deep neural network model architecture, using genotype information as input and rice blast resistance phenotype as output. Use the dataset to train and optimize the deep neural network model architecture to obtain a well-trained predictive model for rice blast resistance.

[0051] like Figure 1 The specific execution steps of the first aspect of the present invention are shown in the figure.

[0052] In one embodiment, the genotype information is haplotype information integrated from SNP, PAV and IOT, and the obtaining step comprises: mounting to a pan-genome according to second-generation sequencing data of a rice population, detecting and integrating SNP, PAV and IOT to obtain population genotypes.

[0053] Table 1 SNP genotype table

[0054]

[0055] In the table, SNP genotypes of four rice samples are given, and ATCG represents base variation.

[0056] Table 2 PAV genotype table

[0057]

[0058] In the table, PAV genotypes of four rice samples are given, and A represents the absence of large fragments, and C represents the presence of large fragments.

[0059] Table 3 IOT genotype table

[0060]

[0061] In the table, IOT genotypes of four rice samples are given, and A represents the absence of inversion or translocation, and C represents the presence of inversion or translocation.

[0062] The same genotypes in SNP, PAV and IOT are integrated, as shown in Table 4:

[0063] Table 4 SNP, PAV and IOT genotype integration table

[0064]

[0065] In one embodiment, the data set for constructing the prediction model includes a training set, a test set and a validation set. S3 specifically comprises the following execution steps:

[0066] S31: Determine the parameters of the deep neural network, construct a deep neural network model, and train the model using the training set data;

[0067] S32: Test the model using data of the validation set, adopt ten-fold cross-validation, and optimize the parameters according to the prediction accuracy of the model;

[0068] S33: Evaluate the optimized model using the test set. If the requirements are met, the training of the model is completed, and the model is output. If the requirements are not met, continue to optimize and train using the training set and the validation set.

[0069] In this embodiment, the deep neural network parameters include model parameters and hyperparameters. Further, the model parameters include the number of convolution kernels, the weighting coefficients of linear regression, and the bias term; and the hyperparameters include the learning rate, the number of iterations, the batch size, and the number of hidden layers of the neural network.

[0070] In one embodiment, the association analysis method comprises: using the gapit algorithm to perform correlation analysis on all markers in the genotype information and the specific data of the rice blast resistance phenotype (such as the lesion degree after infection of the leaf, stem or ear, the number and size of the lesion, the disease development speed, and the final yield loss, etc.), screening all markers with a p-value greater than a set value in the association analysis result, and proposing the corresponding genotype information. For example, the marker genotype completely unrelated to the rice blast resistance phenotype is all markers with a p-value greater than 0.001 in the association analysis result. The remaining SNP, PAV and IOT markers form the genotype information input to the prediction model.

[0071] In one embodiment, it further comprises the step of one-hot encoding the genotype information in the data set:

[0072] The genotype of SNP includes A, T, C, G, and deletion or heterozygosis, which are respectively converted into five classification identifiers 1, 2, 3, 4 and 5, and then one-hot encoded;

[0073] The genotype of PAV includes two types of presence and absence, which are respectively converted into two classification identifiers 1 and 2, and then one-hot encoded;

[0074] The genotype of IOT includes two types of presence and absence, which are respectively converted into two classification identifiers 1 and 2, and then one-hot encoded;

[0075] The encoded genotype information is used as the input of the deep neural network model architecture.

[0076] The one-hot encoding of the SNP genotype is shown in Table 5:

[0077] Table 5: One-hot encoding table of SNP genotype

[0078]

[0079] In one embodiment, the output of the prediction model of rice blast resistance of rice is the grade of rice blast resistance phenotype, for example, divided into three levels, resistant to disease is 0 level, susceptible to disease is 2 level and intermediate type is 1 level; as shown in Table 6. The grade division of the rice blast resistance phenotype is obtained according to the evaluation of the specific data of the rice blast resistance phenotype. For the prediction model training process, the corresponding rice blast resistance phenotype grade label is determined by screening the genotype information related rice blast resistance phenotype specific data, and participates in the model training.

[0080] The proportion of the predicted rice blast resistance grade equal to the real rice blast resistance grade in the current validation set is used as an index for judging the prediction accuracy of the prediction model.

[0081] The steps of ten-fold cross-validation include: re-dividing the data set of S1, re-running the correlation analysis of S2, and the number and location information of the reduced SNP, PAV and IOT markers are likely to change. The accuracy of the prediction model is verified for the validation sets obtained by multiple divisions, and the final prediction accuracy of the prediction model is the mean of the accuracies of 10 validations.

[0082] The conditions for completing the training of the prediction model include: the accuracy of the model, and / or the robustness of the model, and / or the generalization ability of the model meet the requirements.

[0083] Further, the accuracy of the model refers to the proportion of the predicted rice blast resistance grade equal to the real rice blast resistance grade; the robustness of the model refers to the variance between the accuracy results of multiple runs of the model; and the generalization ability of the model refers to the model still has similar accuracy when applied to non-training set data as when applied to the training set.

[0084] In this embodiment, statistically, the higher the accuracy, the more the model meets the requirements or the higher the degree of meeting the requirements. The smaller the variance, the more the model meets the requirements or the higher the degree of meeting the requirements.

[0085] Table 6: Classification table of rice blast resistance phenotype

[0086]

[0087] In one embodiment, the prediction model includes a core architecture and a module.

[0088] Further, the core architecture is based on Tensorflow; and the module includes Python, panda and scikit_learn.

[0089] In this embodiment, the prediction model includes an input layer, a hidden layer and an output layer. The shape of the input layer of the prediction model of rice blast resistance is [n, m, 5], where n is the number of rice samples, m is the number of screened genotype markers, and 5 is the dimension after one-hot encoding of the genotype.

[0090] In this embodiment, the prediction model of rice blast resistance includes 8 hidden layers.

[0091] The first hidden layer is a one-dimensional convolutional layer with 32 convolutional kernels, a kernel size of 3 and a relu activation function.

[0092] The second hidden layer is a one-dimensional convolutional layer, the number of convolutional kernels is 64, the size of the convolutional kernel is 3, and the activation function is relu.

[0093] The third hidden layer is a flatten layer.

[0094] The fourth hidden layer is a dropout layer.

[0095] The fifth hidden layer is a fully connected layer, containing 64 neurons, and the activation function is relu.

[0096] The sixth hidden layer is a dropout layer.

[0097] The seventh hidden layer is a fully connected layer, containing 32 neurons, and the activation function is relu.

[0098] The eighth hidden layer is a dropout layer.

[0099] In this embodiment, the output layer is a fully connected layer, containing the same number of neurons as the number of rice blast disease phenotype grades, for example, if the grades are divided into three levels, the number of neurons is three, and the activation function is softmax.

[0100] In one embodiment, the loss function of the model uses sparse class cross-entropy, the optimizer uses Adam optimizer, the learning rate is set to 1e-4, and the evaluation index uses accuracy.

[0101] The following provides a specific execution example of the first aspect of the embodiment:

[0102] S100: Data acquisition and segmentation.

[0103] S101: Extract DNA from 863 international rice varieties.

[0104] S102: After the DNA is randomly broken by ultrasonic waves, a library is built, and then high-depth sequencing (greater than or equal to 20x) is performed using second-generation DNA sequencing technology. The read length of the current second-generation sequencing data is 150bp.

[0105] S103: Use alignment software to align all samples of high-depth sequencing to the linear pan-genome, respectively. Obtain SNP, PAV and IOT information of the sample, and integrate them into a genotype matrix.

[0106] S103: Plant the population in Yangjiang rice blast nursery, and investigate leaf blast and panicle blast resistance phenotype during flowering period.

[0107] S105: Correspond the genotype and phenotype to construct a data set of rice blast resistance prediction model, including training set, test set and validation set; wherein the data of 63 samples is the test set, the data of 80 samples is the validation set, and 720 samples are the training set.

[0108] S200: Dimensionality reduction of rice genotypes.

[0109] S201: Using gapit software, input the genotypes and phenotypes in the training set for association analysis.

[0110] S202: Remove marker genotypes with p-value less than 0.001 to achieve genotype dimensionality reduction.

[0111] S300: Construct a prediction model from genotype to rice blast resistance phenotype.

[0112] S301: One-hot encoding of the reduced marker genotypes. Among them, SNP contains A, T, C, G, deletion, heterozygous and other types, the deletion and heterozygous in the genotype are classified as one genotype, A, T, C, G, deletion or heterozygous are respectively converted into 1, 2, 3, 4, 5, then one-hot encoding is performed, there are 5 types, PAV genotypes include two types of presence and absence, which are converted into 1, 2 and then one-hot encoded, IOT genotypes include two types of presence and absence, which are converted into 1, 2 and then one-hot encoded.

[0113] S302: Classification of disease resistance phenotype, where the phenotype value in the range of [1, 3.4] is resistant type, recorded as 1, the phenotype in the range of (3.4, 6.6) is intermediate type, recorded as 2, and the phenotype in the range of [6.6, 9] is susceptible type, recorded as 3.

[0114] S303: Set the prediction model parameters, the hidden layer includes 8 hidden layers, the first hidden layer is a one-dimensional convolution layer, the number of convolution kernels is 32, the size of the convolution kernel is 3, and the activation function is relu; the second hidden layer is a one-dimensional convolution layer, the number of convolution kernels is 64, the size of the convolution kernel is 3, and the activation function is relu; the third hidden layer is a flatten layer; the fourth hidden layer is a dropout layer; the fifth hidden layer is a fully connected layer, containing 64 neurons, and the activation function is relu; the sixth hidden layer is a dropout layer; the seventh hidden layer is a fully connected layer, containing 32 neurons, and the activation function is relu; the eighth hidden layer is a dropout layer; the output layer is a fully connected layer, containing 3 neurons, and the activation function is softmax. Set the loss function of the model as sparse categorical cross entropy, use Adam optimizer as the optimizer, set the learning rate to 1e-4, and use accuracy as the evaluation index.

[0115] S400: Train the model and test the prediction accuracy of the model.

[0116] S401: Use the untrained model to predict rice blast resistance, input the genotype, and calculate the accuracy.

[0117] S402: Set the hyperparameter epochs to 100 and perform model training. If the accuracy evaluation indicator does not improve for 5 consecutive epochs, terminate the training early.

[0118] S403: Use the trained model to predict rice blast resistance phenotype in the validation set and calculate the accuracy.

[0119] S404: Divide the training set into 9 parts, take the first part as the validation set, and integrate the original validation set data into the remaining training set. Repeat the model training and prediction to obtain an accuracy data. Take the second, third, …, ninth part as the validation set to complete the model training and prediction. A total of 10 accuracy values are obtained, and the average of the 10 results represents the accuracy of the model in the validation set.

[0120] S500: Apply the trained model to the test set for prediction evaluation.

[0121] S501: Randomly select 63 other samples as the test set, and divide the remaining 800 data into training and validation sets for ten-fold cross-validation model accuracy. Finally, test the model accuracy on the test set.

[0122] S502: Repeat the S501 step several times to evaluate the accuracy of the constructed model in predicting rice blast resistance after training.

[0123] In the above example S201: Use gapit software to input genotype and phenotype in the training set for association analysis, the following further examples are realized:

[0124] Use gapit software to perform GWAS analysis on the 720 samples in the training set based on the mixed linear regression (MLM) model. The specific operation is: integrate and construct the genotype matrix of SNP, PAV and IOT genotype, input genotype and disease resistance phenotype, and the program will output the p-value of each marker and the contribution to the phenotype.

[0125] Since the investigated rice blast disease resistance phenotype includes panicle blast and leaf blast, two GWAS results can be obtained. Take the markers with p-value <0.001 for subsequent analysis. Since panicle blast and leaf blast are related, the markers affecting panicle blast may also affect leaf blast, and vice versa. Therefore, the markers screened from the two GWAS results are taken and used for subsequent analysis.

[0126] In S301 of the above example, the genotype initially contains various base types such as A, T, C, and G. The deep learning deletes the input numerical type data, so it is necessary to convert different genotypes into numerical types. If A, T, C, and G are simply converted into 1, 2, 3, and 4, there are problems such as 2 being greater than 1 and 4 being greater than 3. However, there is no size difference between different genotypes, so one-hot encoding is used to avoid the type difference of the input data and achieve normalization.

[0127] In S303 of the above example, for the genotype matrix, each rice sample is independent, and the similar genotype markers may have some connection due to being in a similar position on the chromosome. Therefore, one-dimensional convolution operation can be performed on the genotype matrix, and the convolution kernel performs convolution operation on different markers of the same sample. In this example, one-bit convolution operation is used in the first hidden layer and the second hidden layer. Then, a flattening layer is used, followed by a plurality of subsequent fully connected layers. In order to avoid overfitting, a plurality of random inactivation layers are added.

[0128] The neural network parameters inside the model can be compared and optimized using the control variable method. For example, the number of convolution kernels in the first hidden layer and the second hidden layer is set to 32 and 64, respectively, and the number of convolution kernels in the second layer is set to 32, 64, and 128, respectively. A neural network with 2*3=6 combinations can be constructed, and all other parameters are the same. The model accuracy on the ten-fold cross-validation and test set is used to compare the results, and the optimal parameter combination is selected. Other parameters such as increasing the number of hidden layers can be optimized by the control variable method, and the final evaluation standard is the model accuracy on the ten-fold cross-validation and test set.

[0129] The above example uses ten-fold cross-validation to obtain the average accuracy of the trained model on the validation set, which is significantly improved compared to before training. Finally, the model accuracy is verified on the test set, indicating that the constructed model can be used to predict rice blast resistance. The results show that the model of the present application can accurately predict the performance of rice genotype on rice blast resistance, which is of great significance for the improvement of rice materials and cross-regional introduction.

[0130] The second aspect of the embodiment of the present application also discloses a rice blast resistance prediction method based on artificial intelligence, comprising the following steps:

[0131] S1: collecting rice samples, sequencing the rice samples, identifying SNP, PAV, and IOT genotypes, integrating and constructing a genotype matrix as genotype information;

[0132] S2: based on the correlation between all markers in the genotype information and the rice blast resistance phenotype, screening and retaining genotype information that meets the correlation requirement;

[0133] S3: inputting the genotype information meeting the correlation requirement into a prediction model of rice blast resistance to perform phenotype prediction of rice blast resistance, and outputting a model prediction result; the prediction model of rice blast resistance is the prediction model of rice blast resistance trained in the first aspect of the embodiment.

[0134] In one embodiment, the sequencing method of S1 refers to second-generation high-throughput sequencing.

[0135] In one embodiment, S2 further comprises: one-hot encoding conversion of the screened and retained marker genotypes to obtain genotype information meeting the correlation requirement.

[0136] In one embodiment, the prediction method in S3 comprises: loading and reusing the trained model using the load function in the Tensorflow.saved_model of the prediction model, thereby completing the prediction of the rice blast resistance grade.

[0137] The third aspect of the embodiment of the present application further discloses an electronic device comprising:

[0138] one or more processors;

[0139] a storage device configured to store one or more programs,

[0140] When the one or more programs are executed by the one or more processors, the one or more processors implement the artificial intelligence-based rice blast resistance prediction method disclosed in the second aspect of the embodiment.

[0141] The fourth aspect of the embodiment of the present application further discloses a computer readable medium having a computer program stored thereon, which is executed by a processor to implement the artificial intelligence-based rice blast resistance prediction method disclosed in the second aspect of the embodiment.

[0142] The above provides a prediction model, a prediction method and related equipment for rice blast resistance, and the principles and implementation modes of the present application are described in this paper. The above examples are only used to help understand the method and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed, and the above description should not be understood as a limitation of the present application.

[0143] In this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional elements of the same name in the process, method, article, or apparatus.

Claims

1. A method for constructing a predictive model for rice blast resistance, characterized in that, Includes the following steps: S1: Obtain rice blast resistance phenotype and genotype information; the genotype information is haplotype information after integration of SNP, PAV and IOT, and the acquisition steps include: based on the second-generation sequencing data of the rice population, it is loaded onto the pangenome, SNP, PAV and IOT are detected and integrated to obtain the population genotype; S2: Use association analysis to calculate the correlation between all markers in the genotype information and the rice blast resistance phenotype, screen and retain the genotype information that meets the correlation requirements, and construct the dataset of the rice blast resistance prediction model with the corresponding rice blast resistance phenotype. S3: Construct a deep neural network model architecture and perform one-hot encoding on the genotype information in the dataset: The genotypes of SNPs include A, T, C, G, as well as deletion or heterozygote, which are converted into five classification markers and then one-hot encoded. PAV genotypes include two types: present and absent. These are converted into two classification markers and then one-hot encoded. The genotype of IOT includes two types: presence and absence. These are converted into two classification markers and then one-hot encoded. Using the encoded genotype information as input to the deep neural network model architecture and the rice blast resistance phenotype as output, the deep neural network model architecture is trained and optimized using the dataset to obtain a trained predictive model for rice blast resistance.

2. The method for constructing a predictive model for rice blast resistance according to claim 1, characterized in that, The output of the rice blast resistance prediction model is the rice blast resistance phenotype level.

3. The method for constructing a predictive model for rice blast resistance according to claim 1, characterized in that, The association analysis method includes: using the Gapit algorithm to perform correlation analysis between all markers in the genotype information and the rice blast resistance phenotype, screening all markers in the association analysis results whose p-value is greater than a set value, and extracting the corresponding genotype information.

4. The method for constructing a predictive model for rice blast resistance according to claim 1, characterized in that, The input layer shape of the rice blast resistance prediction model is [n, m, 5], where n is the number of rice samples, m is the number of genotype markers after screening, and 5 is the dimension after one-hot encoding of genotypes.

5. The method for constructing a predictive model for rice blast resistance according to claim 1, characterized in that, The predictive model for rice blast resistance includes eight hidden layers: The first hidden layer is a one-dimensional convolutional layer with p convolutional kernels, a kernel size of 3, and a ReLU activation function. The second hidden layer is a one-dimensional convolutional layer with 2*p convolutional kernels, a kernel size of 3, and the activation function is ReLU. The third hidden layer is a flattened layer; The fourth hidden layer is a randomly deactivated layer; The fifth hidden layer is a fully connected layer containing 2*p neurons, with ReLU as the activation function; The sixth hidden layer is a randomly deactivated layer; The seventh hidden layer is a fully connected layer containing p neurons, with ReLU as the activation function; The eighth hidden layer is a randomly deactivated layer.

6. A method for predicting rice blast resistance based on artificial intelligence, characterized in that, Includes the following steps: S1: Collect rice samples, sequence the rice samples, identify SNP, PAV and IOT genotypes, integrate and construct a genotype matrix as genotype information; S2: Based on the correlation between all markers in the genotype information and the rice blast resistance phenotype, screen and retain genotype information that meets the correlation requirements; S3: Input the genotype information that meets the correlation requirements into the rice blast resistance prediction model to predict the rice blast resistance phenotype, and output the model prediction results; the rice blast resistance prediction model is a trained rice blast resistance prediction model constructed according to any one of claims 1-5.

7. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the artificial intelligence-based rice blast resistance prediction method as described in claim 6.

8. A computer-readable medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the artificial intelligence-based rice blast resistance prediction method as described in claim 6.

Images