Multi-species epigenetic modification prediction method and apparatus
By combining a multi-species epigenetic modification prediction method with a feature extraction module and a multi-task learning model, the problem of cross-species epigenetic regulatory networks has been solved, enabling cross-species epigenetic modification prediction. This method can be applied to fields such as plant epigenetics research and agricultural production.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- THE INST OF BIOTECHNOLOGY OF THE CHINESE ACAD OF AGRI SCI
- Filing Date
- 2024-11-29
- Publication Date
- 2026-05-21
Smart Images
Figure CN2024135695_21052026_PF_FP_ABST
Abstract
Description
A method and apparatus for predicting epigenetic modifications in multiple species
[0001] Cross-references
[0002] This application claims priority to Chinese Patent Application No. 2024116069975, filed on November 12, 2024, entitled “A Method and Apparatus for Predicting Epigenetic Modifications in Multiple Species,” the entire disclosure of which is incorporated herein by reference. Technical Field
[0003] This invention relates to the fields of bioinformatics and plant epigenetics, and in particular to a method and apparatus for predicting epigenetic modifications in multiple species. Background Technology
[0004] Epigenetic modifications refer to gene expression regulation mechanisms that do not involve changes in DNA sequence, including but not limited to DNA methylation and histone modifications. These modifications play a crucial role in plant growth and development, as well as in responses to environmental stresses. Although the development of high-throughput sequencing technology has greatly advanced research in plant epigenetics, limitations in experimental techniques mean that our understanding of the epigenetic modification profiles of most plant species remains limited. Traditional machine learning methods are often limited to single species and struggle to capture complex cross-species epigenetic regulatory networks. Therefore, developing an epigenetic modification prediction method that can integrate multi-species data and possess cross-species predictive capabilities is essential. Summary of the Invention
[0005] This invention provides a method and apparatus for predicting epigenetic modifications across multiple species, addressing the shortcomings of existing technologies that are limited to a single species and struggle to capture complex cross-species epigenetic regulatory networks, thereby enabling cross-species epigenetic modification prediction. The technical solution proposed by this invention is as follows:
[0006] In a first aspect, the present invention provides a method for predicting epigenetic modifications in multiple species, comprising:
[0007] Obtain genome sequences and pre-established multi-species epigenetic modification prediction models; wherein, the multi-species epigenetic modification prediction model includes a feature extraction module and a multi-task learning model;
[0008] The genomic sequence is used to extract features based on the feature extraction module to obtain a feature representation;
[0009] The feature representation is input into the multi-task learning model to predict the modification level data for each type of epigenetic modification.
[0010] Optionally, the feature extraction module performs feature extraction on the genome sequence to obtain a feature representation, including:
[0011] The genome sequence is labeled to obtain the forward sequence and the reverse complementary sequence, respectively;
[0012] The forward sequence and the reverse complementary sequence are input into the feature extraction module for feature extraction to obtain the feature representation.
[0013] Optionally, the feature extraction module includes a first convolutional module, multiple residual modules, a position embedding layer, and multiple Transformer encoding modules; the step of inputting the forward sequence and the inverse complementary sequence into the feature extraction module for feature extraction to obtain the feature representation includes:
[0014] The forward sequence and the reverse complementary sequence are respectively input into the first convolutional module to obtain the first feature map;
[0015] The first feature map is input into the plurality of residual modules, and each residual module captures local and global feature information through a skip connection mechanism to obtain the second feature map;
[0016] The second feature map is input into the location embedding layer for location encoding to obtain the third feature map;
[0017] The third feature map is input into the Transformer encoding module, and global contextual relationship features are extracted through positional encoding and multi-head attention mechanism to obtain the fourth feature map;
[0018] The feature representation is obtained by superimposing the fourth feature maps corresponding to the forward sequence and the reverse complementary sequence.
[0019] Optionally, the multi-species epigenetic modification prediction model is trained in the following manner:
[0020] The genome sequences and corresponding epigenetic modification data of multiple species were obtained to form a dataset; the epigenetic modification data represents the actual modification level of the genome sequences.
[0021] The dataset is divided into a training set, a validation set, and a test set.
[0022] The pre-established deep neural network model is trained based on the training set, and the model parameters are updated using the backpropagation algorithm to minimize the loss function, thereby obtaining the trained model; wherein, the loss function is used to measure the difference between the model's predicted output and the actual level of modification.
[0023] The trained model is evaluated using a validation set under different combinations of hyperparameters, and the hyperparameters are optimized using a preset optimization method to obtain an optimized model.
[0024] The performance of the optimized model is evaluated using a test set until the model performance meets the preset requirements, thus obtaining the multi-species epigenetic modification prediction model.
[0025] Optionally, the method further includes:
[0026] The genome sequences of all species are converted into a unified probabilistic coding format.
[0027] Optionally, the multi-task learning model includes a second convolutional module, a global pooling layer, and multiple species prediction heads, each species prediction head corresponding to a prediction task of an epigenetic modification type;
[0028] The feature representation is input into the multi-task learning model to predict the modification level data for each epigenetic modification type, including:
[0029] The feature representation is input into the second convolution module for feature extraction, and the output of the convolution module is obtained.
[0030] The global pooling layer performs global pooling on the output of the convolution module, and inputs the output of the global pooling layer into the prediction head of each species to obtain the modification level data of the corresponding epigenetic modification type.
[0031] Secondly, the present invention also provides a multi-species epigenetic modification prediction device, comprising the following modules:
[0032] An acquisition module is used to acquire genome sequences and pre-established multi-species epigenetic modification prediction models; wherein, the multi-species epigenetic modification prediction model includes a feature extraction module and a multi-task learning model;
[0033] The extraction module is used to extract features from the genome sequence based on the feature extraction module to obtain a feature representation;
[0034] The prediction module is used to input the feature representation into the multi-task learning model to predict the modification level data for each type of epigenetic modification.
[0035] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multi-species epigenetic modification prediction method as described in the first aspect above.
[0036] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-species epigenetic modification prediction method as described in the first aspect above.
[0037] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the multi-species epigenetic modification prediction method as described in the first aspect above.
[0038] Based on the above technical solution, the beneficial effects of the present invention compared with the prior art are as follows:
[0039] This invention provides a method and apparatus for predicting epigenetic modifications across multiple species. By using a feature extraction module to extract features from the genome sequence, the method learns and extracts feature representations that can represent epigenetic modification information. Since the prediction model is trained on multi-species data, the feature extraction module can learn common features across species. These common features exist in the genome sequences of different species and can therefore be used for cross-species prediction. The feature representations are input into a multi-task learning model, which can simultaneously handle the prediction of multiple epigenetic modification types. The multi-task learning model predicts multiple epigenetic modification types based on the feature representations of the input genome sequence and provides modification level data for each type. This invention, by combining a feature extraction module and a multi-task learning model, achieves cross-species prediction capabilities and can more accurately predict modification level data for multiple epigenetic modification types. It can be widely applied in plant epigenetics research, agricultural production, and biodiversity conservation.
[0040] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.
[0041] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description
[0042] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0043] Figure 1 is a flowchart illustrating the multi-species epigenetic modification prediction method provided by this invention.
[0044] Figure 2 is a schematic diagram of the architecture of the multi-species epigenetic modification prediction model provided by the present invention.
[0045] Figure 3 is a schematic diagram of the structure of the multi-species epigenetic modification prediction device provided by the present invention.
[0046] Figure 4 is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0048] The multi-species epigenetic modification prediction method and apparatus of the present invention are described below with reference to Figures 1-3.
[0049] Referring to Figure 1, the multi-species epigenetic modification prediction method includes the following:
[0050] Step S101: Obtain the genome sequence and a pre-established multi-species epigenetic modification prediction model.
[0051] First, the genome sequence to be predicted is obtained from databases or experimental data, which can be obtained through high-throughput sequencing technology. Simultaneously, a pre-established multi-species epigenetic modification prediction model is acquired. This model is trained on a large amount of multi-species epigenetic modification data using machine learning methods, particularly deep learning. The model consists of a feature extraction module and a multi-task learning model to effectively capture the characteristic differences between different plant species. The feature extraction module extracts useful feature information from the genome sequence, while the multi-task learning model handles the prediction tasks of multiple epigenetic modification types simultaneously. The feature extraction module learns general epigenetic features, while the multi-task learning model captures species-specific features, enabling the model to distinguish between multi-species data in detail. Through this architecture, the model can comprehensively analyze the input data to obtain the predicted modification levels of various epigenetic modifications, including DNA methylation and histone modifications. The prediction results can provide accurate candidate modification sites and target regions for further epigenetic experimental research.
[0052] Training the aforementioned multi-species epigenetic modification prediction model requires integrating epigenetic modification data from multiple species. By pre-training on large-scale, diverse datasets, the model can learn common features across species, thus gaining the ability to make cross-species predictions.
[0053] Step S102: Extract features from the genome sequence based on the feature extraction module to obtain a feature representation.
[0054] A feature extraction module is used to extract features from genomic sequences to obtain feature representations that can represent epigenetic modifications within the genomic sequence. This module can include deep learning models such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), or Transformers to learn and extract useful feature information from the genomic sequence. Feature extraction transforms the raw genomic sequence into higher-level feature representations that more accurately reflect epigenetic modifications. Because the model is trained on multi-species data, the feature extraction module can learn common features across species. These common features exist in the genomic sequences of different species and can therefore be used for cross-species predictions.
[0055] Step S103: Input the feature representation into the multi-task learning model to predict the modification level data for each epigenetic modification type.
[0056] By inputting feature representations into a multi-task learning model, this model can simultaneously handle prediction tasks for multiple epigenetic modification types. The multi-task learning model trains multiple species prediction heads simultaneously using the feature representations output by the feature extraction module, with each species prediction head corresponding to a prediction task for a specific epigenetic modification type. In this way, the model can share information across multiple tasks, improving prediction accuracy. Multi-task learning can leverage the correlations between different tasks to improve the model's generalization ability and prediction accuracy. Furthermore, because the model is pre-trained on multi-species data, it possesses cross-species prediction capabilities.
[0057] The aforementioned epigenetic modifications refer to gene expression regulation mechanisms that do not involve changes in DNA sequence, including but not limited to DNA methylation and histone modifications. Correspondingly, the types of epigenetic modifications include DNA methylation, DNA modification, DNA-binding proteins, histone markers, transcription factors, chromatin accessibility, and RNA modification. The modification level data for each type of epigenetic modification specifically refers to the degree or state of that type of modification in a particular tissue, cell, or organism, as described in existing technologies. For example, the modification level data for DNA methylation typically refers to the degree of addition of methyl groups on the DNA molecule, i.e., the methylation level. The modification level data for histone modifications refers to the degree of modification of specific amino acid residues on histones, such as acetylation and methylation. The modification level data for RNA modifications refers to the degree of addition of specific chemical groups on RNA molecules, such as N6-methyladenine (m6A).
[0058] This invention provides a multi-species epigenetic modification prediction method. A feature extraction module extracts features from the genome sequence, learning and extracting feature representations that represent epigenetic modification information. Since the prediction model is trained on multi-species data, the feature extraction module can learn common features across species. These common features exist in the genome sequences of different species and can therefore be used for cross-species prediction. The feature representations are input into a multi-task learning model, which can simultaneously handle the prediction of multiple epigenetic modification types. The multi-task learning model predicts multiple epigenetic modification types based on the feature representations of the input genome sequence and provides modification level data for each type. This invention, by combining a feature extraction module and a multi-task learning model, achieves cross-species prediction capabilities and can more accurately predict modification level data for multiple epigenetic modification types. It can be widely applied in plant epigenetics research, agricultural production, and biodiversity conservation.
[0059] In an optional embodiment, forward and reverse information from the genome sequence can be utilized to capture more biological features. Step S102 above, which involves feature extraction from the genome sequence using a feature extraction module to obtain a feature representation, includes:
[0060] S1021. The genome sequence is labeled to obtain the forward sequence and the reverse complementary sequence.
[0061] The genome sequence is labeled using the genome labeling module (module a), specifically by marking specific nucleotide or base sequences to facilitate subsequent feature extraction and model understanding. The encoder module generates the forward sequence (i.e., the original sequence itself) and the reverse complementary sequence based on the original genome sequence. The reverse complementary sequence is obtained by replacing bases in the original sequence according to the AT / CG complementarity rules and reversing the sequence order.
[0062] S1022. Input the forward sequence and the reverse complementary sequence into the feature extraction module to extract features and obtain the feature representation.
[0063] The forward and reverse complementary sequences are input into the feature extraction module to extract useful feature information from the sequence. The feature extraction module captures pattern, structural, and variation information in the genome sequence and converts it into high-dimensional feature representations. These feature representations can more accurately reflect the epigenetic modification information in the genome sequence.
[0064] This invention, by introducing inverse complementary sequences, captures both forward and inverse information from genomic sequences, leading to a more comprehensive understanding of the sequence's biological characteristics. This helps improve the model's predictive accuracy for epigenetic modifications. Furthermore, the introduction of inverse complementary sequences increases dataset diversity, enabling the model to learn more features and patterns during training. This enhances the model's generalization ability, allowing it to better adapt to different species and types of genomic sequences.
[0065] In an optional embodiment, referring to Figure 2, the present invention designs a feature extraction module, including a convolutional module (hereinafter referred to as the first convolutional module), multiple residual blocks, a position embedding layer, and multiple Transformer encoder blocks. The convolutional module captures key patterns from the genomic sequence, and then applies batch normalization and activation functions to enhance feature expression capabilities. The feature extraction module extracts features from the input sequence through the convolutional module and Transformer encoder blocks, thereby more effectively capturing complex epigenetic regulatory patterns in the plant genome and significantly improving prediction accuracy.
[0066] Step S1022 above, which involves inputting the forward sequence and the reverse complementary sequence into the feature extraction module for feature extraction to obtain the feature representation, includes:
[0067] S201. Input the forward sequence and the reverse complementary sequence into the first convolution module respectively to obtain the first feature map.
[0068] DNA sequences are retrieved from a genome database, and their forward and reverse complementary sequences are generated. These two sets of sequences represent the original and reverse complementary directions of the DNA sequence, respectively, helping to capture more comprehensive biological information. The forward and reverse complementary sequences are then input into the first convolutional module.
[0069] The first convolutional module performs convolution operations on the sequence by sliding convolution kernels across it, capturing key patterns from the genome sequence, and then applies batch normalization and activation functions to enhance feature expression capabilities, generating the first feature map.
[0070] S202. The first feature map is input into the plurality of residual modules. Each residual module captures local and global feature information through a skip connection mechanism to obtain the second feature map.
[0071] Referring to Figure 2, the residual module c consists of multiple convolutional layers and batch normalization layers, combined with residual connections, the Swish activation function, and Dropout operations to achieve information transfer and feature preservation in the deep network. The residual module mitigates the vanishing gradient problem through cross-layer information transfer, enhancing the training effect of the deep network. The first feature map is input into multiple residual modules, which capture local and global feature information through a skip connection mechanism. Each residual module further extracts and refines features to generate a second feature map. The second feature map contains richer local and global feature information.
[0072] S203. Input the second feature map into the position embedding layer for position encoding to obtain the third feature map.
[0073] A positional embedding layer is a mechanism for capturing positional information within a sequence. The second feature map is input into the positional embedding layer for positional encoding. Through positional encoding, each feature is assigned its positional information within the sequence, which is crucial for understanding the contextual relationships within the sequence. A third feature map is then generated, containing the positionally encoded feature information.
[0074] S204. Input the third feature map into the Transformer encoding module d, and extract global contextual relationship features through positional encoding and multi-head attention mechanism to obtain the fourth feature map.
[0075] Referring to Figure 2, the Transformer encoding module d is a deep learning model based on an attention mechanism, capable of capturing global contextual relationship features. The Transformer encoding module includes positional encoding, multi-head attention, feedforward layers, layer normalization layers, and dropout layers. These components work together to capture dependencies and interactions at different positions in the genome sequence from a global perspective, helping to extract more biologically meaningful features.
[0076] The third feature map is input into the Transformer encoding module. Through positional encoding and multi-head attention mechanisms, the Transformer encoding module can extract global features and information from the DNA sequence, generating a fourth feature map that includes global contextual features. The output of the Transformer encoding module corresponding to the above forward sequence is a forward tensor, and the output corresponding to the above reverse complementary sequence is a reverse tensor.
[0077] S205. The fourth feature maps corresponding to the forward sequence and the reverse complementary sequence are superimposed to obtain the feature representation.
[0078] By superimposing the features of the forward and reverse tensors corresponding to the forward and reverse complementary sequences, a comprehensive feature representation containing information about the forward and reverse complementary sequences can be obtained.
[0079] This invention, through the combined use of convolutional modules, residual modules, positional embedding layers, and Transformer encoding modules, can extract local, global, and contextual features from DNA sequences. These features help improve the model's prediction accuracy for epigenetic modification prediction tasks. Furthermore, by simultaneously processing forward and reverse complementary sequences, more comprehensive biological features and information within the DNA sequence can be captured. This contributes to a more accurate understanding of the biological functions and regulatory mechanisms of DNA sequences. Introducing reverse complementary sequences increases the diversity of the dataset, helping the model learn more features and patterns. This improves the model's generalization ability across different datasets and scenarios.
[0080] In an optional embodiment, the multi-species epigenetic modification prediction model is trained in the following manner:
[0081] S301. Obtain the genome sequences and corresponding epigenetic modification data of multiple species to obtain a dataset; wherein, the epigenetic modification data is the actual modification level of the genome sequence.
[0082] Several plant species were selected as research subjects, including model plants such as Arabidopsis thaliana, rice (Oryza sativa), and maize (Zea mays), as well as other crops that play an important role in agriculture.
[0083] Genome sequences are obtained from public databases. Genome sequence data is typically stored in FASTA format and contains sequences of chromosomes or genomic regions of the species. Epigenetic modification data is stored in BED, WIG, or custom formats and includes information such as modification location, type, and level. Low-complexity regions (such as repetitive or simple sequences) in the genome sequence are removed using bioinformatics tools or custom scripts. These regions may interfere with the prediction of epigenetic modifications.
[0084] Epigenetic modification data are obtained from specific research literature or laboratory databases. This data may include modification levels such as DNA methylation and histone modifications. For epigenetic modification data, it is also necessary to remove noisy data or unreliable modification sites to ensure data accuracy.
[0085] The genome sequences of all species are converted into a uniform probabilistic coding format. This coding method typically involves converting DNA sequences (A, T, C, G) into numerical vectors, or employing more complex representations such as one-hot encoding or k-mer encoding, to meet the input requirements of deep neural network models. For example, NumPy can be used to convert processed genome sequences into a probabilistic coding format for subsequent input to deep neural network models. The probabilistic coding process is as follows:
[0086] Create a NumPy array of shape (sequence length, 4) to store the probability vector for each base. Iterate through each base in the genome sequence, setting the corresponding probability value in the probability vector based on the base type (A, T, C, G). (For example, for base A, the vector can be set to [1,0,0,0] or a more complex probability distribution). This can be encoded as: 'A':[1,0,0,0],'C':[0,1,0,0],'G':[0,0,1,0],'T':[0,0,0,1],'M':[0.5,0.5,0,0],'R':[0.5,0,0.5,0],'W':[0.5,0,0,0.5],'S':[0,0.5,0.5,0],'Y' :[0,0.5,0,0.5],'K′:[0,0,0.5,0.5],'V′:[1 / 3,1 / 3,1 / 3,0],'H′:[1 / 3,1 / 3,0,1 / 3],'D':[1 / 3,0,1 / 3,1 / 3],'B':[0,1 / 3,1 / 3,1 / 3],'N':[0.25,0.25,0.25,0.25].
[0087] The processed genome sequences and epigenetic modification data are integrated to form a comprehensive dataset containing multiple species and various epigenetic modification types. The dataset should include the genome sequence of each species, the corresponding epigenetic modification data (including modification location, type, and modification level), and any relevant metadata (such as species name, tissue type, treatment conditions, etc.).
[0088] S302. Divide the dataset into a training set, a validation set, and a test set to train, validate, and test the deep learning model in subsequent steps. When partitioning the data, ensure the independence of the training, validation, and test sets to avoid model overfitting.
[0089] S303. Based on the training set, the pre-established deep neural network model is trained, and the model parameters are updated using the backpropagation algorithm to minimize the loss function, thereby obtaining the trained model; wherein, the loss function is used to measure the difference between the model's predicted output and the actual modification level.
[0090] A deep neural network model is pre-designed, capable of processing genomic sequence input and outputting corresponding epigenetic modification prediction results. This model represents the initial form of the multi-species epigenetic modification prediction model.
[0091] The model is trained using a training set that includes genomic information from multiple model plants and important crops, ensuring that the model learns features with universal adaptability. The model parameters are continuously adjusted using the backpropagation algorithm to minimize the loss function. The loss function measures the difference between the model's predicted output and the actual modification level, and can employ methods such as mean squared error (MSE) or cross-entropy loss.
[0092] This invention uses mean squared error (MSE) as the loss function to measure the difference between the model's predicted output and the actual modification level. The MSE loss function can intuitively reflect the deviation between the predicted value and the true value, which helps guide the optimization of model parameters. The AdamW optimizer is used for parameter updates to effectively adjust the model weights during training, promoting rapid model convergence.
[0093] The training process is as follows: Input the training data into the model, calculate the predicted output through forward propagation, and calculate the loss value according to the MSE loss function. Use the AdamW optimizer to backpropagate and update the model parameters based on the loss value. Repeat the above steps until a preset number of training epochs is reached or the validation set loss no longer improves.
[0094] S304. Use the validation set to evaluate the trained model under different combinations of hyperparameters, and optimize the hyperparameters using a preset optimization method to obtain the optimized model.
[0095] The performance of the trained model under different hyperparameter combinations is evaluated using a validation set. Hyperparameters include learning rate, regularization coefficient, kernel size, batch size, etc. A reasonable range of hyperparameters is set, and all possible combinations are explored. The performance of each combination is evaluated through cross-validation. The optimal hyperparameter combination is found using a preset optimization method (such as grid search, random search, Bayesian optimization, etc.). The hyperparameter combination with the best performance is selected as the configuration of the final model.
[0096] Monitor the changes in validation set loss during training. Stop training when the validation set loss stops improving over a period of time to prevent overfitting. In addition to the weight decay built into the AdamW optimizer, strategies such as dropout layers or L1 regularization can be introduced into the model architecture to further reduce the risk of overfitting.
[0097] S305. Use the test set to evaluate the model performance of the optimized model until the model performance reaches the preset requirements, and obtain the multi-species epigenetic modification prediction model.
[0098] The model's performance is validated using a reserved, independent test set. The test set should contain different plant species and epigenetic modification types than the training and validation sets to assess the model's predictive ability on unknown data. Evaluation metrics include accuracy, recall, F1 score, and AUC-ROC curve. If the model performance meets preset requirements (e.g., accuracy, AUC value, etc. reach certain thresholds), the aforementioned multi-species epigenetic modification prediction model is obtained. The model's generalization ability is evaluated through testing on different plant species and epigenetic modification types. This ensures the model can demonstrate good predictive performance in different application scenarios.
[0099] This invention leverages the powerful feature extraction and learning capabilities of deep neural network models to accurately capture the complex relationships between genomic sequences and epigenetic modifications, thereby improving prediction accuracy. By training on data from multiple species, the model can learn common features across different species, enhancing its generalization ability and enabling its application to unknown or newly discovered species. This model provides a powerful tool for epigenetic research. By predicting the levels of epigenetic modifications in different species, it can provide precise candidate modification sites and target regions for further epigenetic experimental studies.
[0100] In an optional embodiment, referring to FIG2, the above multi-task learning model includes a convolutional module (hereinafter referred to as the second convolutional module), a global pooling layer, and multiple species prediction heads, each species prediction head corresponding to a prediction task of an epigenetic modification type;
[0101] Step S103 above, which involves inputting the feature representation into the multi-task learning model to predict the modification level data for each epigenetic modification type, includes:
[0102] S1031. Input the feature representation into the second convolution module for feature extraction to obtain the output of the convolution module.
[0103] The feature representation of the genome sequence is input into the second convolutional module of the multi-task learning model. Referring to Figure 2, each convolutional module b in this invention includes a convolutional layer, a batch normalization layer, a Swish activation function, and a Dropout layer. The convolutional layer is the core component of the second convolutional module; it uses multiple convolutional kernels to extract local features from the input data (i.e., the aforementioned feature representation). Each convolutional kernel slides across the input data, calculates the convolution result, and generates a feature map. These feature maps capture the local features of the input data at different scales and locations. Batch normalization and the Swish activation function are then applied to enhance feature representation capabilities.
[0104] After processing through the above hierarchical structure, the second convolutional module can extract complex feature representations from the input data and pass these features to the subsequent global pooling layer for further processing.
[0105] The Swish activation function described above is applied to the output of the batch normalization layer. The Swish function exhibits linear growth when the input value is large, while tending towards zero when the input value is small. This characteristic helps the model learn more complex feature representations. Dropout is a regularization technique used to prevent overfitting during training. In the Dropout layer, each neuron has a certain probability (e.g., 0.5) of being randomly dropped during each forward propagation, meaning its output is set to zero.
[0106] S1032. The global pooling layer performs global pooling on the output of the convolution module, and inputs the output of the global pooling layer into the prediction head of each species to obtain the modification level data of the corresponding epigenetic modification type.
[0107] Following the second convolutional module, a global pooling layer performs global pooling on the output of the convolutional modules. The output of the global pooling layer is then fed into multiple species prediction heads. Each species prediction head corresponds to a specific species and epigenetic modification type prediction task. These prediction heads are fully connected layers that learn how to predict the modification level of the corresponding species and modification type based on the output of the global pooling layer. Each species prediction head outputs a prediction value representing the modification level data for the corresponding epigenetic modification type. This data can be continuous numerical values (such as methylation level, histone modification intensity, etc.) or categorical labels (such as the presence or absence of a certain modification).
[0108] This invention employs a second convolutional module for feature extraction, enabling the model to learn complex feature representations from the input data. These feature representations are shared across prediction tasks involving multiple species and epigenetic modification types, helping the model learn more general and robust features. The global pooling layer, by pooling the entire feature map, preserves the most important global feature information. This helps the model better utilize this feature information in subsequent prediction tasks, improving prediction accuracy and stability.
[0109] Multi-task learning models achieve information sharing and collaborative optimization among different tasks by simultaneously processing multiple prediction tasks. This collaborative optimization helps the model better learn the relationships between different species and epigenetic modification types, further improving prediction accuracy and generalization ability. Because the model can handle multiple prediction tasks simultaneously, it can significantly improve prediction efficiency in practical applications. Furthermore, the model is highly scalable, allowing for easy addition of prediction tasks for new species or epigenetic modification types without retraining the entire model.
[0110] After obtaining modification level data for each type of epigenetic modification, the predicted modification level data can be compared with a preset threshold to determine the modification site.
[0111] The aforementioned thresholds are important criteria for determining the presence of modified sites. Threshold settings can be based on experimental validation data, literature reports, or model performance evaluation results. The selection of thresholds should ensure accurate identification of truly present modified sites while avoiding excessive false positives.
[0112] A multi-species epigenetic modification prediction model is used to predict the epigenetic modification level of an input genome sequence, obtaining data on the level of epigenetic modifications at each locus or region. These modification levels can be continuous numerical values, representing the relative magnitude of the modification intensity.
[0113] The predicted modification level data for each locus is compared with a preset threshold. If the predicted value is higher than the threshold, the locus is considered to have corresponding epigenetic modification; if the predicted value is lower than the threshold, the locus is considered to have no significant modification or a low degree of modification. Based on the comparison results, the modified loci are labeled. The labeling information can include the modification type (e.g., DNA methylation, histone modification, etc.), the modification intensity, and the locus location (the specific location on the genome). This labeling information provides important evidence for subsequent bioinformatics analysis, experimental validation, and biological discovery.
[0114] The proposed multi-species plant epigenetic modification prediction method has good cross-species applicability and high-precision prediction capabilities, and can be widely applied in plant epigenetics research, agricultural production, and biodiversity conservation. Through a multi-species learning mechanism with shared weights, this method also demonstrates excellent predictive performance for species with scarce data.
[0115] The multi-species epigenetic modification prediction device provided by the present invention is described below. The multi-species epigenetic modification prediction device described below can be referred to in correspondence with the multi-species epigenetic modification prediction method described above.
[0116] The multi-species epigenetic modification prediction device provided by the present invention, as shown in FIG3, includes:
[0117] The acquisition module 401 is used to acquire genome sequences and pre-established multi-species epigenetic modification prediction models; wherein, the multi-species epigenetic modification prediction model includes a feature extraction module and a multi-task learning model;
[0118] Extraction module 402 is used to extract features from the genome sequence based on the feature extraction module to obtain a feature representation;
[0119] The prediction module 403 is used to input the feature representation into the multi-task learning model to predict the modification level data for each type of epigenetic modification.
[0120] Figure 4 illustrates a schematic diagram of the physical structure of an electronic device. As shown in Figure 4, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540. The processor 510, communication interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions from the memory 530 to execute the multi-species epigenetic modification prediction method.
[0121] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0122] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute the multi-species epigenetic modification prediction method provided by the above methods.
[0123] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the multi-species epigenetic modification prediction methods provided by the methods described above.
[0124] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0125] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Industrial applicability
[0127] This invention provides a method and apparatus for predicting epigenetic modifications in multiple species. The method includes: acquiring a genome sequence and a pre-established multi-species epigenetic modification prediction model; wherein the multi-species epigenetic modification prediction model includes a feature extraction module and a multi-task learning model; extracting features from the genome sequence based on the feature extraction module to obtain feature representations; inputting the feature representations into the multi-task learning model to predict the modification level data for each type of epigenetic modification. This invention, by combining the feature extraction module and the multi-task learning model, achieves cross-species prediction capabilities and can more accurately predict the modification level data for multiple types of epigenetic modifications. It can be widely applied in plant epigenetics research, agricultural production, and biodiversity conservation, and has good economic value and application prospects.
Claims
1. A multi-species epigenetic modification prediction method, characterized in that, include: Obtain genome sequences and pre-established multi-species epigenetic modification prediction models; wherein, the multi-species epigenetic modification prediction model includes a feature extraction module and a multi-task learning model; The genomic sequence is used to extract features based on the feature extraction module to obtain a feature representation; The feature representation is input into the multi-task learning model to predict the modification level data for each type of epigenetic modification.
2. The multi-species epigenetic modification prediction method of claim 1, wherein, The feature extraction module extracts features from the genome sequence to obtain a feature representation, including: The genome sequence is labeled to obtain the forward sequence and the reverse complementary sequence, respectively; The forward sequence and the reverse complementary sequence are input into the feature extraction module for feature extraction to obtain the feature representation.
3. The multi-species epigenetic modification prediction method of claim 2, wherein, The feature extraction module includes a first convolutional module, multiple residual modules, a position embedding layer, and multiple Transformer encoding modules; the step of inputting the forward sequence and the inverse complementary sequence into the feature extraction module for feature extraction to obtain the feature representation includes: The forward sequence and the reverse complementary sequence are respectively input into the first convolutional module to obtain the first feature map; The first feature map is input into the plurality of residual modules, and each residual module captures local and global feature information through a skip connection mechanism to obtain the second feature map; The second feature map is input into the location embedding layer for location encoding to obtain the third feature map; The third feature map is input into the Transformer encoding module, and global contextual relationship features are extracted through positional encoding and multi-head attention mechanism to obtain the fourth feature map; The feature representation is obtained by superimposing the fourth feature maps corresponding to the forward sequence and the reverse complementary sequence.
4. The multi-species epigenetic modification prediction method of claim 1, wherein, The multi-species epigenetic modification prediction model was trained in the following manner: The genome sequences and corresponding epigenetic modification data of multiple species were obtained to form a dataset; the epigenetic modification data represents the actual modification level of the genome sequences. The dataset is divided into a training set, a validation set, and a test set; The pre-established deep neural network model is trained based on the training set, and the model parameters are updated using the backpropagation algorithm to minimize the loss function, thereby obtaining the trained model; wherein, the loss function is used to measure the difference between the model's predicted output and the actual level of modification. The trained model is evaluated using a validation set under different combinations of hyperparameters, and the hyperparameters are optimized using a preset optimization method to obtain the optimized model. The performance of the optimized model is evaluated using a test set until the model performance meets the preset requirements, thus obtaining the multi-species epigenetic modification prediction model.
5. The multi-species epigenetic modification prediction method of claim 4, wherein, The method further includes: The genome sequences of all species are converted into a unified probabilistic coding format.
6. The multi-species epigenetic modification prediction method of claim 1, wherein, The multi-task learning model includes a second convolutional module, a global pooling layer, and multiple species prediction heads, each species prediction head corresponding to a prediction task of an epigenetic modification type. The feature representation is input into the multi-task learning model to predict the modification level data for each epigenetic modification type, including: The feature representation is input into the second convolution module for feature extraction, and the output of the convolution module is obtained. The global pooling layer performs global pooling on the output of the convolution module, and inputs the output of the global pooling layer into the prediction head of each species to obtain the modification level data of the corresponding epigenetic modification type.
7. A multi-species epigenetic modification prediction method, characterized in that, include: An acquisition module is used to acquire genome sequences and pre-established multi-species epigenetic modification prediction models; wherein, the multi-species epigenetic modification prediction model includes a feature extraction module and a multi-task learning model; The extraction module is used to extract features from the genome sequence based on the feature extraction module to obtain a feature representation; The prediction module is used to input the feature representation into the multi-task learning model to predict the modification level data for each type of epigenetic modification.
8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the multi-species epigenetic modification prediction method as described in any one of claims 1 to 6. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the multi-species epigenetic modification prediction method as described in any one of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the multi-species epigenetic modification prediction method as described in any one of claims 1 to 6.