A method for predicting the association between rice genotype and phenotype based on machine learning

Abstract

The present invention relates to the fields of bioinformatics and machine learning, and in particular to a method for predicting the association between rice genotype and phenotype based on machine learning. The content includes: collecting genotype data and phenotype data of rice samples, using a dynamic feature extraction algorithm to generate genotype features, thereby obtaining a genotype feature set; S2: constructing a genotype and phenotype association prediction model, and predicting the phenotype based on the genotype feature set. It solves the problem that genotype data is difficult to accurately capture feature information due to uneven data distribution or periodic changes at different gene sites, and existing feature extraction methods are difficult to identify complex features; in terms of agricultural genotype and phenotype prediction, especially in the fields of rice breeding, disease resistance enhancement and yield optimization, there is a lack of effective methods specifically for bioinformatics feature extraction and association prediction; the performance is insufficient when dealing with feature redundancy, resulting in excessive reliance on certain unimportant or redundant features.

Description

Technical Field

[0001] The present invention relates to the fields of bioinformatics and machine learning, and in particular to a method for predicting the association between rice genotype and phenotype based on machine learning. Background Art

[0002] In modern agricultural science, rice is one of the world's major food crops, and its yield and quality directly affect human food security and nutrition supply. Traditional breeding methods rely on phenotypic observation and experience accumulation, which is time-consuming and inefficient. The association study between rice genotype and phenotype helps to reveal the genetic basis, thus providing a scientific basis for breeding. With the development of genomics and bioinformatics, as well as the rapid progress of genome sequencing technology, a large amount of genomic data has been accumulated. Combining machine learning methods to explore the association between genotype and phenotype not only improves the efficiency of research, but also provides a scientific basis for the genetic improvement of rice, helping scientists to accurately predict and screen traits during genome selection and breeding. The association prediction method between genotype and phenotype will surely play a more important role in rice breeding.

[0003] In summary, the existing methods for predicting the association between rice genotype and phenotype have the following technical problems: genotype data are difficult to accurately capture feature information due to uneven data distribution or periodic changes in different gene loci, especially in agricultural genotype research, rice genotype data are sparse, and different gene loci are closely related to the periodic changes in rice growth, and existing feature extraction methods are difficult to identify complex features; in the prediction of agricultural genotype and phenotype, especially in the fields of rice breeding, disease resistance enhancement and yield optimization, there is a lack of effective methods specifically for bioinformatics feature extraction and association prediction, there are limitations in capturing complex genotype patterns and phenotype prediction, and it is difficult to cope with the complex interaction characteristics between rice genes; the performance is insufficient in dealing with feature redundancy, resulting in excessive reliance on some unimportant or redundant features. Summary of the invention

[0004] The present invention provides a method for predicting the association between rice genotype and phenotype based on machine learning, so as to solve the problem that genotype data is difficult to accurately capture characteristic information due to uneven data distribution or periodic changes in different gene sites, especially in agricultural genotype research, the genotype data of rice shows sparseness, and at the same time, different gene sites are closely related to the periodic changes of rice growth, and the existing feature extraction method is difficult to identify complex features; in the prediction of agricultural genotype and phenotype, especially in the fields of rice breeding, disease resistance enhancement and yield optimization, there is a lack of effective methods specifically for bioinformatics feature extraction and association prediction, there are limitations in capturing complex genotype patterns and phenotype prediction, and it is difficult to cope with complex interactive features between rice genes; the performance is insufficient in processing feature redundancy, resulting in a technical problem of excessive reliance on certain unimportant or redundant features.

[0005] A method for predicting the association between rice genotype and phenotype based on machine learning, comprising the following steps:

[0006] S1: Collect genotype data and phenotype data of rice samples, use dynamic feature extraction algorithm to generate genotype features, and thus obtain a genotype feature set;

[0007] S2: Construct a genotype-phenotype association prediction model to predict the phenotype based on the genotype feature set.

[0008] Preferably, the S1 specifically includes:

[0009] The calculation formula of genotype characteristics is:

[0010]

[0011] in, Indicates The rice samples were The value of the genotype characteristic of each gene locus; Indicates The weight of the genotype characteristics of each gene locus; Indicates The rice samples were Genotype data of each gene locus; represents the hyperbolic tangent function; Indicates The mean value of each gene locus in all rice samples; Indicates The standard deviation of each gene locus in all samples; Represents the sine function.

[0012] Preferably, the S2 specifically includes:

[0013] The genotype-phenotype association prediction model predicts the phenotype by learning and extracting genotype features; the genotype-phenotype association prediction model includes an input layer, an interactive feature learning layer, a residual connection layer, a dynamic feature selection layer and an output layer.

[0014] Preferably, the S2 specifically includes:

[0015] The genotype feature set is input as input features into the input layer; the interactive feature learning layer introduces the gene interactive feature learning mechanism in bioinformatics, and multiplies each element in the genotype feature bit by bit through the element-level product operation. At the same time, a matrix is ​​generated through the outer product operation, and each element in the matrix represents the nonlinear interaction between different features; by introducing the weight matrix of the interactive feature learning layer, the results of the element-level product and the outer product are converted to the output feature space, and the output result is adjusted using the bias vector; and the nonlinear relationship is learned through the nonlinear activation function.

[0016] Preferably, the S2 specifically includes:

[0017] The residual connection layer combines the output results of the interactive feature learning layer with the genotype feature set, and introduces a gating mechanism to control the transmission of input features through the activation function; and outputs the genotype feature set of the rice sample after passing through the residual connection layer.

[0018] Preferably, the S2 specifically includes:

[0019] The dynamic feature selection layer dynamically adjusts the weight of the genotype features according to the contribution of the genotype features of the rice samples to the prediction results after passing through the residual connection layer. The adjustment is based on two aspects: one is the contribution of each genotype feature to the loss function in the previous round; the other is the similarity penalty mechanism; and outputs the genotype feature set of the rice samples after dynamic feature selection.

[0020] Preferably, the S2 specifically includes:

[0021] The similarity penalty mechanism means that when the performance of two genotype characteristics is very similar, the weight of one of the genotype characteristics will be reduced. Through gradual screening and weighting, the important characteristics will be amplified, and ultimately the genotype-phenotype association prediction model will focus on the characteristics that are most critical to the prediction results.

[0022] Preferably, the S2 specifically includes:

[0023] Design an optimized loss function based on feature similarity and gene contribution, judge the prediction effect, and update the weights of each genotype feature through back propagation according to the loss function value; the calculation formula of the optimized loss function based on feature similarity and gene contribution is:

[0024]

[0025] in, Indicates The loss function value at the round iteration; Represents the average factor in the loss function; Express The rice samples were summed; Express The phenotypic data are summed; Indicates The importance weight of each phenotypic data; Indicates The first Phenotypic data; The predicted The first Phenotypic data; represents the regularization term.

[0026] Preferably, the S2 specifically includes:

[0027] After interactive feature learning, residual connection and dynamic feature selection, the output layer linearly transforms the genotype feature set of the processed rice samples and maps the results of the linear transformation through the activation function;

[0028] The prediction formula for the phenotype dataset is:

[0029]

[0030] in, The predicted A phenotypic dataset of rice samples; express Type activation function, used to transform the result after linear transformation Mapped in between; Represents the first A collection of genotype characteristics of rice samples; represents the weight matrix of the output layer; Represents the bias vector of the output layer.

[0031] The beneficial effects of the technical solution of the present invention are:

[0032] 1. Through the dynamic feature extraction algorithm based on bioinformatics optimization, the sparsity, periodicity and complex nonlinear relationships in the genotype data are effectively captured. Specifically, the genotype data is standardized, logarithmic transformed, and a combination of hyperbolic tangent function and sine function is adopted to make the extracted features accurate and comprehensive, thereby enhancing the ability to capture important information in rice genotype data, thereby improving the accuracy of predicting rice phenotypes (such as yield, disease resistance, etc.), helping to accelerate the agricultural breeding process, and providing accurate rice gene information, thereby effectively improving the yield and disease resistance of crops.

[0033] 2. In the interactive feature learning layer, element-wise multiplication and outer product operations are used to effectively capture the linear and nonlinear interactive information between genotype features, enhancing the learning ability of the genotype-phenotype association prediction model for the complex relationship between genes, thereby providing strong technical support for rice breeding optimization.

[0034] 3. The problem of gradient vanishing or gradient exploding is overcome through the residual connection layer. The introduction of the residual structure ensures that the original feature information can be retained in the complex learning process. Even after the features undergo multiple layers of transformation, the original information can still be retained, so that the prediction of the association between rice genotype and phenotype can still obtain accurate results under complex data conditions.

[0035] 4. In the dynamic feature selection layer, by adjusting the contribution and similarity penalty mechanism of different features, the feature weights are gradually optimized, the influence of redundant features on the prediction results is reduced, and the overfitting of the genotype-phenotype association prediction model is effectively prevented, thereby improving the generalization ability of the genotype-phenotype association prediction model, and being able to adapt to the phenotype prediction needs under different rice planting conditions and environments, significantly enhancing the practicality and stability of the genotype-phenotype association prediction model, and providing robust decision support for agricultural breeding and crop improvement.

[0036] 5. The genotype-phenotype association prediction model can not only be used for rice breeding optimization, disease resistance enhancement and yield prediction, but can also be extended to genotype analysis of other crops. It uses machine learning methods to explore the complex relationship between genotype and phenotype. It has strong portability and provides technical support for genetic analysis and breeding of other agricultural crops.

[0037] 6. The combination of dynamic feature extraction algorithm optimized based on bioinformatics and machine learning model can effectively improve the ability to predict the association between genotype and phenotype of rice and other crops, and is practical, scalable and stable. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a method for predicting the association between rice genotype and phenotype based on machine learning described in the present invention. DETAILED DESCRIPTION

[0039] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0041] The specific scheme of the method for predicting the association between rice genotype and phenotype based on machine learning provided by the present invention is described in detail below with reference to the accompanying drawings.

[0042] Refer to the attached Figure 1 , which shows a flow chart of a method for predicting the association between rice genotype and phenotype based on machine learning provided by an embodiment of the present invention, the method comprising the following steps:

[0043] S1, collecting genotype data and phenotype data of rice samples, and using a dynamic feature extraction algorithm to generate genotype features, thereby obtaining a genotype feature set;

[0044] Collected from public databases and agricultural field experiments The genotype data and phenotype data of rice samples can be used to guide agricultural breeding, enhance disease resistance and optimize yield. The genotype data and phenotype data are expressed as follows:

[0045]

[0046] in, Indicates A genotype dataset of rice samples, , Indicates the number of genotype data, Indicates The rice samples were Genotype data of gene loci, such as (Single Nucleotide Polymorphism), ; Indicates A phenotypic dataset of rice samples, , Indicates the number of phenotypic data. Indicates The first Phenotypic data, such as yield and disease resistance, are directly used for numerical phenotypic data in order to facilitate systematic analysis of crop phenotypes, while categorical phenotypic data are processed using one-hot encoding. The one-hot encoding method is a well-known technical means for those skilled in the art and will not be described in detail here. ;

[0047] In order to further enhance the feature extraction capability in bioinformatics, a dynamic feature extraction algorithm based on bioinformatics optimization is adopted, which not only captures the sparsity and periodicity characteristics in genotype data, but also focuses on mining potential complex relationships from genotype data;

[0048] The dynamic feature extraction algorithm is used to extract genotype features, aiming to capture the sparsity and periodicity characteristics in the genotype data while ensuring that the genotype data is standardized;

[0049] By extracting the mean and standard deviation of each gene locus, each gene locus is standardized to ensure that the genotype data will not affect the calculation results due to different scales in subsequent steps;

[0050] The natural logarithm function can reduce the scale of genotype data and enhance the influence of sparse features by adding 1 to the absolute value of each genotype data and then taking the logarithm to capture the sparsity in the genotype data. By adding 1, mathematical errors when the input value is zero are avoided;

[0051] Nonlinear transformation using hyperbolic tangent function is used to capture subtle changes and complex relationships in genotype data;

[0052] Capturing the periodic characteristics in genotype data through the sine function can effectively reveal the periodic changes of rice genotypes in different growth stages, which is of great significance for predicting yield and disease resistance;

[0053] The calculation formula of genotype characteristics is:

[0054]

[0055] in, Indicates The rice samples were The value of the genotype characteristics of each gene locus is used to predict agricultural related traits; Indicates The weight of the genotype characteristics of each gene locus; Indicates The rice samples were Genotype data of each gene locus; Represents the hyperbolic tangent function, which is used to enhance nonlinear expression capabilities and avoid the gradient vanishing problem; Indicates The mean of each locus in all rice samples is used to standardize genotype data and eliminate scale differences between different loci. Indicates The standard deviation of each gene locus in all samples is used for standardization to eliminate the variance differences of genotype data at different gene loci; Represents the natural logarithm function, which calculates the absolute value of the genotype data and adds 1 to capture the sparsity in the genotype data. The natural logarithm function can reduce the scale of the genotype data and enhance the impact of sparse features. By adding 1, mathematical errors when the input value is zero are avoided; represents the sine function, which is used to capture the periodic characteristics of genotype data;

[0056] The extracted genotype feature set As the input of the rice genotype-phenotype association prediction model, the genotype feature set is as follows:

[0057]

[0058] in, Indicates A collection of genotype signatures for rice samples.

[0059] S2, construct a genotype-phenotype association prediction model to predict the phenotype based on the genotype feature set;

[0060] In order to achieve the prediction of genotype-phenotype association, especially for rice breeding optimization, disease resistance and yield prediction, a genotype-phenotype association prediction model was constructed to predict phenotype based on genotype feature sets to improve the ability to capture biological information;

[0061] The genotype-phenotype association prediction model aims to predict the phenotype by learning and extracting genotype features. The genotype-phenotype association prediction model includes an input layer, an interactive feature learning layer, a residual connection layer, a dynamic feature selection layer and an output layer. Through the combination and operation of each other, effective feature information can be extracted from the genotype feature set, and finally accurate phenotype prediction can be performed;

[0062] Input Layer:

[0063] Enter the A collection of genotype features of rice samples , will soon As input features, each rice sample is represented by a set of genotype features, which are the basic data from which the genotype-phenotype association prediction model further extracts effective features;

[0064] Interactive feature learning layer:

[0065] The gene interaction feature learning mechanism in bioinformatics is introduced to enhance the capture of complex relationships between genes. Through the element-level product operation, each element in the genotype feature is multiplied bit by bit to capture the linear interaction between features. At the same time, the outer product operation is used to generate a matrix. Each element in the matrix represents the nonlinear interaction between different features. Therefore, through the above operation, the interaction information of linear and nonlinear features can be learned at the same time, which enhances the learning ability of the genotype-phenotype association prediction model for complex biological data.

[0066] Furthermore, the weight matrix of the interactive feature learning layer is introduced to convert the results of element-wise multiplication and outer product into the output feature space, capture the interactive information, and use the bias vector to further adjust the output result to ensure that even if the input data value is extremely small, the output can still be expressed. Nonlinearity is introduced through a nonlinear activation function, so that the genotype-phenotype association prediction model can learn complex feature patterns.

[0067] The output of the interactive feature learning layer is calculated as follows:

[0068]

[0069] in, It means that after the interactive feature learning layer The genotype feature set of rice samples, that is, the first Output features of rice samples; Represents the activation function, which is used to introduce nonlinear characteristics and learn complex nonlinear relationships; Represents the weight matrix of the interactive feature learning layer; Indicates A collection of genotype characteristics of rice samples; Represents element-wise product, which takes the input features Each feature element of is bitwise multiplied to capture the linear interaction between features; Represents the outer product. The outer product of two vectors generates a matrix that captures the nonlinear interactions between different features. Represents the bias vector, which is used to correct the output result;

[0070] In agricultural applications, capturing linear and nonlinear relationships can help improve the analysis of the contribution of different combinations of rice gene loci to agronomic traits;

[0071] Residual connection layer:

[0072] The introduction of the residual connection layer is to stabilize the learning process and avoid the problem of gradient disappearance or gradient explosion. The output results of the interactive feature learning layer are combined with the genotype feature set to ensure that in the complex learning process, even if the feature information changes, the original feature information can still be maintained through the residual structure;

[0073] A gating mechanism is introduced to control the transmission of input features through the activation function, ensuring that only features filtered by the activation function can pass, effectively reducing the impact of redundant features on the genotype-phenotype association prediction model, while improving the robustness of the genotype-phenotype association prediction model;

[0074] The output of the residual connection layer is calculated as follows:

[0075]

[0076] in, It means after the residual connection layer The genotype feature set of rice samples, that is, the first Output features of rice samples, , It means after the residual connection layer The ?th genotype feature of rice samples, Represents the residual connection layer after Total number of genotypic characteristics of rice samples; Indicates A collection of genotype characteristics of rice samples; Indicates that in the residual connection layer, it is used to linearly transform the output features of the interactive feature learning layer and extract the weight matrix of the features; It indicates that the activation function is used to introduce nonlinear characteristics and learn complex nonlinear relationships; express Type activation function, used for gating mechanism, control The output value is between; Represents the weight matrix in the gating mechanism, which is used for input features Perform linear transformation to control input features degree of passing;

[0077] Dynamic Feature Selection Layer:

[0078] The dynamic feature selection layer dynamically adjusts the weight of the genotype feature according to the contribution of the genotype feature of the rice sample after the residual connection layer to the final prediction result. In the initial stage, each genotype feature will be given the same weight, that is, it is assumed that each genotype feature has the same importance to the prediction result. As the training of the genotype-phenotype association prediction model proceeds, the weight of the genotype feature is continuously adjusted. The basis for the adjustment is two aspects: one is the contribution of each genotype feature to the loss function in the previous round; the other is the similarity penalty mechanism. The similarity penalty mechanism means that when the performance of two genotype features is very similar, the weight of one feature will be reduced to avoid repeated calculations and feature redundancy. Through gradual screening and weighting, important features are amplified, and finally the genotype-phenotype association prediction model focuses on the features that are most critical to the prediction results.

[0079] In order to improve the application effect of the genotype-phenotype association prediction model in the field of bioinformatics, an optimized loss function based on feature similarity and gene contribution was designed. The gap between the predicted phenotypic data and the real phenotypic data was measured and summed to obtain the overall loss function value. The larger the loss function value, the worse the prediction effect. In order to reduce the loss, the weights of each genotype feature were updated through back propagation. In order to prevent the genotype-phenotype association prediction model from overfitting, a regularization term was introduced to ensure that when analyzing the contribution of multiple gene loci to agricultural phenotypes, the genotype-phenotype association prediction model is not interfered by some features and can be generalized and applied to rice phenotype prediction under different growth conditions and environments.

[0080] The output of the dynamic feature selection layer is calculated as follows:

[0081]

[0082] in, represents the first A collection of genotype characteristics of rice samples; Represents the residual connection after processing The genotype characteristics are weighted summed; It means after the residual connection layer The ?th genotype feature of the rice sample, that is, the ?th genotype feature after the residual connection layer The first output features; Indicated in The weight of the ?th genotype feature in the round iteration. In the initial state, the weights of all genotype features Set to a uniform value, i.e. , used to reflect the The importance of each genotype feature is calculated as follows:

[0083]

[0084] in, Indicated in The weight of the ?th genotype feature in the round iteration; Represents an exponential function, which is used to limit the amplitude of genotype feature adjustment to prevent the update from being too fast or too slow; Represents the learning rate, which is used to control the update speed of the genotype feature weights; Represents the similarity penalty coefficient, which is used to reduce redundant features; It means after the residual connection layer The first output features and The similarity measurement between the output features reflects the redundancy between the genotype features. The similarity measurement method of cosine similarity can be used, which is a technical means well known to those skilled in the art and will not be described in detail here; Indicates The genotype characteristics in The contribution to the loss function in the round iteration. The loss function is calculated as:

[0085]

[0086] in, Indicates The loss function value at the round iteration; Represents the average factor in the loss function, The losses of the rice samples are averaged to ensure that each rice sample has the same impact on the loss function; Express The rice samples were summed; Express The phenotypic data are summed; Indicates The importance weight of each phenotypic data is used to control the weight of different phenotypic data in the loss function; Indicates The first Phenotypic data; The predicted The first Phenotypic data; represents a regularization term to prevent overfitting, which is a technical means well known to those skilled in the art and will not be described in detail here;

[0087] The dynamic feature selection layer is based on the residual connection layer. Dynamically adjust the weight of the contribution of the output features of each rice sample to the predicted phenotype, focus on the genotype features that have important contributions to agricultural phenotypes such as yield and disease resistance, reduce feature redundancy through a similarity penalty mechanism, and optimize the performance of the genotype-phenotype association prediction model;

[0088] Output layer:

[0089] To further ensure the stability of the genotype-phenotype association prediction model and the accuracy of bioinformatics feature capture, after interactive feature learning, residual connection and dynamic feature selection, the output layer will perform a linear transformation on the genotype feature set of the processed rice samples, and map the results of the linear transformation to a reasonable range through an activation function;

[0090] The prediction formula for the phenotype dataset is:

[0091]

[0092] in, The predicted A phenotypic dataset of rice samples, , The predicted The first Phenotypic data; express Type activation function, used to transform the result after linear transformation Mapped in between; Represents the weight matrix of the output layer, which is used to connect the neurons of the dynamic feature selection layer and the output layer to form a linear transformation; Represents the bias vector of the output layer, which is used to provide an additional linear adjustment to ensure that even when all input features are zero, the output neurons can still produce non-zero outputs to help fit the data;

[0093] The genotype-phenotype association prediction model can also be used for breeding analysis of other crops, expanding the application scope of agricultural genotype-phenotype association prediction. It can help researchers understand the relationship between genotype and phenotype, and determine in advance the genotype combination that produces excellent phenotypic characteristics, which will help accelerate the crop improvement process, increase the yield and quality of agricultural products, and contribute to the sustainable development of global agriculture.

[0094] In summary, a method for predicting the association between rice genotype and phenotype based on machine learning was completed.

[0095] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0096] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such 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, and should be included in the protection scope of the present invention.

Claims

1. A method for predicting the association between rice genotype and phenotype based on machine learning, characterized in that: The following steps are involved: S1: Collect genotype data and phenotype data of rice samples, and use dynamic feature extraction algorithm to generate genotype features. The calculation formula of genotype features is: ； in, Indicates The rice samples were The value of the genotype characteristic of each gene locus; Indicates The weight of the genotype characteristics of each gene locus; Indicates The rice samples were Genotype data of each gene locus; represents the hyperbolic tangent function; Indicates The mean value of each gene locus in all rice samples; Indicates The standard deviation of each gene locus in all samples; represents the sine function; Thus, a set of genotype characteristics is obtained; S2: Construct a genotype-phenotype association prediction model to predict the phenotype based on the genotype feature set. The genotype-phenotype association prediction model predicts the phenotype by learning and extracting genotype features. The genotype-phenotype association prediction model includes an input layer, an interactive feature learning layer, a residual connection layer, a dynamic feature selection layer, and an output layer. The genotype feature set is input as input features into the input layer; the interactive feature learning layer introduces the gene interactive feature learning mechanism in bioinformatics, and multiplies each element in the genotype feature bit by bit through the element-level product operation. At the same time, a matrix is ​​generated through the outer product operation, and each element in the matrix represents the nonlinear interaction between different features; by introducing the weight matrix of the interactive feature learning layer, the results of the element-level product and the outer product are converted to the output feature space, and the output result is adjusted using the bias vector; the nonlinear relationship is learned through the nonlinear activation function; The residual connection layer combines the output results of the interactive feature learning layer with the genotype feature set, and introduces a gating mechanism to control the transmission of input features through the activation function; and outputs the genotype feature set of the rice sample after passing through the residual connection layer; The dynamic feature selection layer dynamically adjusts the weight of the genotype features according to the contribution of the genotype features of the rice samples to the prediction results after passing through the residual connection layer. The adjustment is based on two aspects: one is the contribution of each genotype feature to the loss function in the previous round; the other is the similarity penalty mechanism; and outputs the genotype feature set of the rice samples after dynamic feature selection.

2. The method for predicting the association between rice genotype and phenotype based on machine learning according to claim 1, characterized in that: The S2 specifically includes: The similarity penalty mechanism means that when the performance of two genotype characteristics is very similar, the weight of one of the genotype characteristics will be reduced. Through gradual screening and weighting, the important characteristics will be amplified, and ultimately the genotype-phenotype association prediction model will focus on the characteristics that are most critical to the prediction results.

3. The method for predicting the association between rice genotype and phenotype based on machine learning according to claim 1, characterized in that: The S2 specifically includes: Design an optimized loss function based on feature similarity and gene contribution, judge the prediction effect, and update the weights of each genotype feature through back propagation according to the loss function value; the calculation formula of the optimized loss function based on feature similarity and gene contribution is: ； in, Indicates The loss function value at the round iteration; Represents the average factor in the loss function; Express The rice samples were summed; Express The phenotypic data are summed; Indicates The importance weight of each phenotypic data; Indicates The first Phenotypic data; The predicted The first Phenotypic data; represents the regularization term.

4. The method for predicting the association between rice genotype and phenotype based on machine learning according to claim 1, characterized in that: The S2 specifically includes: After interactive feature learning, residual connection and dynamic feature selection, the output layer linearly transforms the genotype feature set of the processed rice samples and maps the results of the linear transformation through the activation function; The prediction formula for the phenotype dataset is: ； in, The predicted A phenotypic dataset of rice samples; express Type activation function, used to transform the result after linear transformation Mapped in between; Represents the first A collection of genotype characteristics of rice samples; represents the weight matrix of the output layer; Represents the bias vector of the output layer.

Images