Crop genotype prediction phenotype method, control device and equipment

Abstract

The embodiment of the invention provides a crop genotype prediction phenotype method, a control device and equipment, and the method comprises the steps: S1, constructing an input layer which is used for receiving genotype data; s2, a data feature extraction layer is constructed to perform feature extraction on the genotype data, the data feature extraction layer comprises a convolutional neural network and an auto-encoder, and the auto-encoder is used for generating low-dimensional global features; s3, inputting the low-dimensional global features into a feature relationship capture layer, wherein the feature relationship capture layer captures a long-distance dependency relationship of the genotype data based on a self-attention mechanism; and S4, constructing an output layer, wherein the output layer is used for mapping the output of the feature relationship capture layer into a phenotype prediction value. The method not only can deeply understand and extract data features, but also can capture a complex relationship between a genotype and a phenotype, thereby improving the expression ability and prediction accuracy of the model.

Description

Technical Field

[0001] The present application relates to the field of phenotypic prediction, and in particular to a method, control device and equipment for predicting phenotypes of crop genotypes. Background Art

[0002] As the global population grows and climate change intensifies, improving crop breeding efficiency is crucial to food security. Traditional breeding relies on field trials, which are long and costly, while genomic prediction technology accelerates the breeding process through genotype-phenotype association analysis. Although existing methods such as BLUP and genomic selection have certain effects, these methods have significant limitations, such as over-reliance on pedigree information, difficulty in adapting to complex populations, only capturing additive effects, and ignoring nonlinear relationships and epistatic effects between genotypes and phenotypes. In recent years, deep learning has provided new ideas for genome prediction, but its application still faces challenges: although models based on convolutional neural networks can extract local genotype features, it is difficult to take into account global features and long-distance dependencies; structures such as recurrent neural networks have low training efficiency and are difficult to process high-dimensional genotype data; existing models are insufficient in integrating multi-scale features and balancing computational efficiency and prediction accuracy.

[0003] Therefore, developing a method for predicting crop genotypes and phenotypes and improving the prediction accuracy of crop phenotypes are of great value for realizing intelligent breeding and shortening the breeding cycle. Summary of the invention

[0004] The present application provides a crop genotype prediction phenotype method, control device and equipment, which can not only deeply understand and extract data features, but also capture the complex relationship between genotype and phenotype, thereby improving the expression ability of the model and the prediction accuracy, and providing support for breeding work.

[0005] In a first aspect, a method for predicting crop phenotype by genotype is provided, the method comprising: S1: construct the input layer, which is used to receive genotype data and preprocess and encode the genotype data; S2: Construct a data feature extraction layer to extract features from genotype data. The data feature extraction layer includes: Convolutional neural network, which is used to extract local features of genotype data; Autoencoder: Autoencoder is used to reduce the dimension of local features and generate low-dimensional global features; S3: The low-dimensional global features are input into the feature relationship capture layer, which captures the long-distance dependency of genotype data based on the self-attention mechanism; S4: Construct the output layer, which is used to map the output of the feature relationship capture layer into phenotypic prediction values.

[0006] It should be understood that this method can process complex and high-dimensional genotype data, extract representative features from it, better capture the complex relationship between genotype and phenotype, and predict crop phenotypic values ​​based on genotype data, thereby providing support for breeding work.

[0007] In conjunction with the first aspect, in certain implementations of the first aspect, the genotype data includes a numerical matrix encoded by a single nucleotide polymorphism (SNP) marker.

[0008] It should be understood that the genotype data obtained through sequencing or genotyping technology of individual genomes can reflect the genetic variation of individuals at the DNA level. In addition to SNP, there are several common types of variation, such as insertion and deletion (Indels), structural variations (Structural Variations, SVs), etc. SNP refers to the phenomenon of a nucleotide variation at a specific position in the genome. It is widely present, relatively stable, not easy to mutate, and easy to be detected.

[0009] In combination with the first aspect, in certain implementations of the first aspect, the convolutional neural network is a 1D convolutional neural network, which uses convolution kernels of multiple different scales to extract local genotype features; the 1D convolutional neural network includes a nonlinear activation function and reduces the feature dimension through pooling operations.

[0010] It should be understood that this design can perform feature extraction and data dimensionality reduction, which is beneficial to subsequent model processing and result output.

[0011] In combination with the first aspect, in some implementations of the first aspect, the autoencoder includes an encoder and a decoder, wherein: The encoder includes a multi-layer convolutional neural network, which is used to reduce the dimension of high-dimensional data and extract key features; The decoder is a mirror image of the encoder and is used to reconstruct the input data to confirm the validity of the features identified by the encoder.

[0012] It should be understood that the low-dimensional global features include low-dimensional genetic information extracted from genotype data.

[0013] It should be understood that the convolutional neural network in the data feature extraction layer focuses on capturing local correlations in genotype data and extracting local features. The encoder (Autoencoder) is used to extract global features from high-dimensional genotype data, while reducing feature dimensions and redundant information, which is very effective in reducing data dimensions, removing noise, and capturing potential genetic structures. In addition, the Autoencoder achieves data compression and reconstruction through encoders and decoders, and can handle high-dimensional and complex genotype data, reduce model computational complexity, and achieve rapid data processing and analysis while retaining as much useful information as possible, which helps to improve the generalization ability of subsequent models.

[0014] In combination with the first aspect, in some implementations of the first aspect, the feature relationship capture layer adopts a Transformer architecture, including: Multi-head self-attention layer, which is used to calculate the association weights between SNP data sites; Position encoding module, the position encoding module is used to embed the chromosome position information of the SNP data site into the feature representation; Feedforward Neural Network,The feedforward neural network is used to perform nonlinear transformation on the output of the multi-head self-attention layer.

[0015] It should be understood that the Transformer architecture is a model based on an attention mechanism, which was originally used for natural language processing tasks, but is also applicable to other types of data. The feature relationship capture layer of this application is designed based on the Transformer architecture. This design uses the self-attention mechanism to directly capture the dependency between any two positions in the sequence without being restricted by distance, allowing the model to effectively capture the complex relationship between features from low-dimensional features. In addition, the Transformer architecture retains the position information of elements in the sequence through position encoding, helping the model understand the specific location of the SNP on the chromosome and its impact on the phenotype, which is conducive to maintaining the importance of genomic structure.

[0016] It should be understood that in the context of genotype data, long-range dependencies refer to the interactions or correlations between SNPs that are far apart in the gene sequence, meaning that even SNPs that are far apart can be directly compared and associated, thereby discovering potential long-range interactions that affect crop phenotypes.

[0017] In combination with the first aspect, in certain implementations of the first aspect, the position encoding module generates a position vector through a sine-cosine function and superimposes it with a low-dimensional global feature to retain the spatial position information of the SNP data site.

[0018] It should be understood that the design of sine-cosine position encoding and low-dimensional feature superposition generates a multidimensional position vector through a mathematical function, which can simultaneously capture the linkage effects of local sites (such as mutation associations between adjacent sites) and the long-range genetic dependencies of global chromosomal regions.

[0019] In combination with the first aspect, in some implementations of the first aspect, the connection method of the convolutional neural network and the autoencoder includes: Flatten the convolutional neural network output feature map into a vector; Map the vector to a low-dimensional latent space through an encoder; A regularization layer is introduced between the encoder and decoder to suppress overfitting.

[0020] It should be understood that flattening the feature map output by the convolutional neural network into a vector and mapping it to a low-dimensional latent space through an encoder can effectively extract and compress the multi-scale spatial features learned by the convolution layer. The construction of a low-dimensional latent space not only reduces data redundancy, but also enhances the model's ability to abstract complex patterns by retaining key features; introducing a regularization layer between the encoder and the decoder can constrain the distribution smoothness of the latent space, suppress overfitting caused by noise interference, and improve the model's robustness to changes in data of different scales.

[0021] In combination with the first aspect, in some implementations of the first aspect, the output layer includes a fully connected network, and the fully connected network includes: At least one hidden layer, the hidden layer is transformed by a nonlinear activation function; The final linear mapping layer is used to output continuous phenotype prediction values.

[0022] It should be understood that the fully connected network with hidden layers and linear mapping layers containing nonlinear activation functions in the output layer design can achieve a balance between feature abstraction and output adaptation through a hierarchical structure. The introduction of nonlinear activation functions (such as ReLU) in the hidden layer can capture complex patterns in the data and break through the bottleneck of the linear model's expressive power; and the final linear layer directly maps high-order features to continuous phenotypic values ​​through a weighted combination without an activation function, which not only avoids the limitation of the activation function on the output range, but also meets the requirements of the regression task for numerical continuity. This design improves the model fitting ability through the nonlinear layer, and then uses the linear layer to achieve smooth prediction, taking into account both prediction accuracy and output stability.

[0023] In a second aspect, a control device is provided, which includes a processor and a memory, wherein the processor is coupled to the memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions in the memory, so that the method described in any implementation method of the first aspect is executed.

[0024] In a third aspect, a device is provided, comprising the control device as described in the second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flowchart of a method for predicting phenotypes from crop genotypes provided in an embodiment of the present application.

[0026] Figure 2 A data feature extraction layer structure diagram provided in an embodiment of the present application.

[0027] Figure 3 A feature relationship capture layer structure diagram provided in an embodiment of the present application. DETAILED DESCRIPTION

[0028] The terms used in the following embodiments are only for the purpose of describing specific embodiments and are not intended to be limiting of the present application. As used in the specification and appended claims of the present application, the singular expressions "one", "a kind of", "", "above", "the" and "this" are intended to also include expressions such as "one or more", unless there is a clear contrary indication in the context. It should also be understood that in the following embodiments of the present application, "at least one", "one or more" refer to one, two or more. The term "and / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist; for example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship.

[0029] References to "one embodiment" or "some embodiments" etc. described in this specification mean that a particular feature, structure or characteristic described in conjunction with the embodiment is included in one or more embodiments of the present application. Thus, the phrases "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear at different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0030] In the context of global population growth and climate change, improving crop breeding efficiency is crucial to food security. Traditional breeding relies on field trials and is inefficient. Genomic prediction technology can accelerate the breeding process through genotype-phenotype association analysis. Existing mainstream methods such as BLUP and genomic selection have significant limitations: excessive reliance on pedigree information leads to insufficient adaptability to complex populations, and only captures additive effects while ignoring nonlinear relationships and epistatic effects between genotypes and phenotypes.

[0031] The embodiments of the present application provide a method, control device and equipment for predicting crop genotype phenotype. The method can not only deeply understand and extract data features, but also capture the complex relationship between genotype and phenotype, thereby improving the expressive power and prediction accuracy of the model, effectively shortening the breeding cycle and significantly improving the accuracy of breeding.

[0032] The technical solution of the present application will be described below with reference to the accompanying drawings.

[0033] Figure 1 A flowchart of a method for predicting phenotypes of crop genotypes provided in an embodiment of the present application. In some examples, the method includes: S1: construct the input layer, which is used to receive genotype data and preprocess and encode the genotype data; S2: Construct a data feature extraction layer to extract features from genotype data. The data feature extraction layer includes: Convolutional neural network, which is used to extract local features of genotype data; Autoencoder: Autoencoder is used to reduce the dimension of local features and generate low-dimensional global features; S3: The low-dimensional global features are input into the feature relationship capture layer, which captures the long-distance dependency of genotype data based on the self-attention mechanism; S4: Construct the output layer, which is used to map the output of the feature relationship capture layer into phenotypic prediction values.

[0034] In some examples, the genotype data includes a matrix of values ​​encoding single nucleotide polymorphism (SNP) markers.

[0035] In one possible implementation, there are 17,114 SNP sites and 10,000 varieties, forming a high-dimensional data matrix.

[0036] Figure 2 A data feature extraction layer structure diagram provided in an embodiment of the present application.

[0037] In some examples, the convolutional neural network is a 1D convolutional neural network, which uses multiple convolution kernels of different scales to extract local genotype features; the 1D convolutional neural network includes a nonlinear activation function and reduces the feature dimension through a pooling operation.

[0038] In one possible implementation, a 1D convolutional neural network uses multiple convolution kernels of different scales (such as sizes of 3, 5, and 7) to extract local patterns in genotype data, introduces nonlinearity through the ReLU activation function to enhance the expressiveness of the model, uses maximum pooling to reduce feature dimensions, reduces computational complexity, and provides a basis for subsequent global feature extraction.

[0039] Optionally, the convolutional layer is calculated as follows: , The calculation formula of the pooling layer is as follows: , in, X is the input data, W i It is i The weight matrix of the convolution kernel, ∗ represents the convolution operation, b i is the bias term, ReLU is the activation function: , Z i is the output of the i-th convolution kernel, P i is the output after pooling.

[0040] In some examples, an autoencoder includes an encoder and a decoder, wherein: The encoder includes a multi-layer convolutional neural network, which is used to reduce the dimension of high-dimensional data and extract key features; The decoder is a mirror image of the encoder and is used to reconstruct the input data to confirm the validity of the features identified by the encoder.

[0041] In one possible implementation, the encoder consists of a series of neural network layers. When processing genotype data, the encoder may contain multiple hidden layers, eventually reaching a bottleneck layer whose dimension is much smaller than the dimension of the input data, thereby achieving compression of the feature dimension. For SNP data, it can identify important features associated with the phenotype; the decoder is also composed of a series of neural network layers, and its architecture is usually a mirror image of the encoder. If the encoder uses convolutional layers for dimensionality reduction, the decoder will use deconvolution layers to gradually restore the dimension of the original data, reconstruct the original input data from the low-dimensional representation generated by the encoder, and ensure that the encoder learns valid features.

[0042] Figure 3 A structural diagram of a feature relationship capture layer provided in an embodiment of the present application. In some examples, the feature relationship capture layer adopts a Transformer architecture, including: Multi-head self-attention layer, which is used to calculate the association weights between SNP data sites; Position encoding module, the position encoding module is used to embed the chromosome position information of the SNP data site into the feature representation; Feedforward Neural Network,The feedforward neural network is used to perform nonlinear transformation on the output of the multi-head self-attention layer.

[0043] Optionally, the feature relationship capture layer is designed with multi-layer Transformer blocks, each of which contains a multi-head self-attention mechanism, a feedforward neural network, residual connections, and layer normalization.

[0044] In one possible implementation, the core components of the feature relationship capture layer include a multi-head self-attention layer and a feedforward neural network. For a single self-attention layer, the calculation formula can be expressed as: , The V value of each attention head is weighted summed according to the attention score: , Concatenate the outputs of all attention heads to get the final multi-head attention output: , in, Q , K , V Represent the query, key, and value matrices respectively, representing the dimensions of the key vector. W 0 represents the weight matrix.

[0045] In some examples, the position encoding module generates a position vector through a sine-cosine function and superimposes it with a low-dimensional global feature to retain the spatial position information of the SNP data site.

[0046] Some examples of how a convolutional neural network can be connected to an autoencoder include: Flatten the convolutional neural network output feature map into a vector; Map the vector to a low-dimensional latent space through an encoder; A regularization layer is introduced between the encoder and decoder to suppress overfitting.

[0047] In some examples, the output layer includes a fully connected network, the fully connected network including: At least one hidden layer, the hidden layer is transformed by a nonlinear activation function; The final linear mapping layer is used to output continuous phenotype prediction values.

[0048] An embodiment of the present application provides a control device, which includes a processor and a memory, wherein the processor is coupled to the memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions in the memory, so that any method in the embodiments described above is executed.

[0049] An embodiment of the present application also provides a device, which includes the control device as described above.

[0050] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. Any equivalent modifications or changes made by ordinary technicians in this field based on the contents disclosed by the present invention should be included in the protection scope recorded in the claims.

Claims

1. A method for predicting crop genotype phenotype, characterized in that: The method comprises: S1: construct an input layer, the input layer is used to receive genotype data, and the input layer is used to preprocess and encode the genotype data; S2: Constructing a data feature extraction layer to extract features from the genotype data, the data feature extraction layer comprising: A convolutional neural network, wherein the convolutional neural network is used to extract local features of the genotype data, and the convolutional neural network is a 1D convolutional neural network, and the 1D convolutional neural network uses a plurality of convolution kernels of different scales to extract local genotype features; the 1D convolutional neural network includes a nonlinear activation function and reduces feature dimensions through a pooling operation; An autoencoder, wherein the autoencoder is used to perform a dimensionality reduction operation on the local features to generate low-dimensional global features, and the autoencoder includes an encoder and a decoder, wherein: the encoder includes a multi-layer convolutional neural network, and the encoder is used to reduce the dimension of high-dimensional data and extract key features; the decoder is a mirror image architecture of the encoder, and the decoder is used to reconstruct input data to confirm the validity of the features identified by the encoder; S3: inputting the low-dimensional global features into a feature relationship capture layer, wherein the feature relationship capture layer captures the long-distance dependency of the genotype data based on a self-attention mechanism; S4: constructing an output layer, wherein the output layer is used to map the output of the feature relationship capture layer into a phenotype prediction value.

2. The method according to claim 1, characterized in that: The genotype data includes single nucleotide polymorphism (SNP), a numerical matrix encoded by SNP markers.

3. The method according to claim 1, characterized in that The feature relationship capture layer adopts the Transformer architecture, including: A multi-head self-attention layer, which is used to calculate the association weights between SNP data sites; A position encoding module, the position encoding module is used to embed the chromosome position information of the SNP data site into the feature representation; A feedforward neural network is used to perform a nonlinear transformation on the output of the multi-head self-attention layer.

4. The method according to claim 3, characterized in that The position encoding module generates a position vector through a sine-cosine function and superimposes it with the low-dimensional global feature to retain the spatial position information of the SNP data site.

5. The method according to claim 1, characterized in that The connection method between the convolutional neural network and the autoencoder includes: Flattening the convolutional neural network output feature map into a vector; Mapping the vector to a low-dimensional latent space via an encoder; A regularization layer is introduced between the encoder and decoder to suppress overfitting.

6. The method according to claim 1, characterized in that The output layer includes a fully connected network, and the fully connected network includes: at least one hidden layer, the hidden layer being transformed by a non-linear activation function; The final linear mapping layer is used to output a continuous phenotype prediction value.

7. A control device, characterized in that: The method comprises a processor and a memory, wherein the processor is coupled to the memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions in the memory, so that the method according to any one of claims 1 to 6 is executed.

8. A device, characterized in that The device comprises a control device as claimed in claim 7.

Images