Method for predicting DNA methylation of brain tissue by using DNA methylation of peripheral tissue

Abstract

The invention discloses a method for predicting DNA methylation of brain tissues by using DNA methylation of peripheral tissues. The method comprises the following steps: acquiring DNA methylation data of the peripheral tissues of the brain, wherein the DNA methylation data comprises blood tissues, oral tissues and saliva tissues; a convolution attention mechanism module is used for extracting a cross-tissue feature data matrix of brain peripheral tissue DNA methylation data and dividing the cross-tissue feature data matrix into a training set and a test set; a ResNet-18 model is introduced, a global pooling layer of the ResNet-18 network model is replaced with a mixed pooling layer, and an improved ResNet-18 model is obtained; inputting the training set into an improved ResNet-18 model for training to obtain a brain tissue DNA methylation prediction model, and optimizing model parameters by using ten-fold cross validation to obtain an optimized brain tissue DNA methylation prediction model; inputting the test set into the optimized brain tissue DNA methylation prediction model to obtain predicted brain tissue DNA methylation data; according to the method, the DNA methylation data of the brain tissue can be obtained without the brain tissue, the brain peripheral tissue sample is easy to obtain, and the obtained data result is stable and accurate.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and in particular to a method for predicting brain tissue DNA methylation by using peripheral tissue DNA methylation. Background Art

[0002] DNA methylation levels are closely related to the pathogenesis and progression of brain diseases. Since it is difficult to obtain brain tissue samples and the number is limited, using DNA methylation of easily accessible peripheral tissues to predict DNA methylation of brain tissue has important clinical application value. With the advancement of sequencing technology, researchers have obtained a large amount of DNA methylation data of three peripheral brain tissues: blood, oral cavity, and saliva. Experiments have shown that there is a significant correlation between DNA methylation of these three peripheral tissues and DNA methylation of brain tissue. Therefore, it is theoretically based to use multi-tissue integrated data to predict brain tissue DNA methylation. Given the differences in DNA methylation characteristics of different peripheral tissues, it is particularly important to effectively integrate multiple peripheral tissue data to extract key features in each tissue.

[0003] In summary, changes in DNA methylation levels in brain tissue can help reveal the pathogenesis, disease progression, and clinical treatment of brain diseases. However, brain tissue is difficult to obtain and is limited in quantity. It usually needs to be obtained through surgery or donation, which is a great limitation for researchers. Moreover, brain tissue samples are relatively scarce in quantity, especially in the collection of human samples, which is often affected by multiple factors such as age, disease status, and sample source. Therefore, due to the limited number of samples, the data set may be insufficient, affecting the generalization ability of the model. Limited samples cannot fully represent the diversity of different individuals, different environments, or different genetic backgrounds, which in turn affects the accuracy and predictive ability of the model. Using brain tissue DNA data alone to obtain brain tissue DNA methylation data is not accurate enough and cannot meet actual usage needs. Summary of the invention

[0004] The present invention provides a method for predicting brain tissue DNA methylation using peripheral tissue DNA methylation, so as to overcome the technical problem that the accuracy of obtaining brain tissue DNA methylation data using brain tissue DNA data is insufficient and cannot meet the actual use requirements.

[0005] In order to achieve the above object, the technical solution of the present invention is:

[0006] A method for predicting brain tissue DNA methylation using peripheral tissue DNA methylation, comprising:

[0007] S1: Acquire DNA methylation data of brain peripheral tissues, wherein the brain peripheral tissues include blood tissues, oral tissues and saliva tissues;

[0008] S2: using a convolutional attention mechanism module to perform feature extraction on the brain peripheral tissue DNA methylation data, extracting a cross-tissue feature data matrix from the peripheral tissue DNA methylation data, and dividing the cross-tissue feature data matrix into a training set and a test set;

[0009] S3: Introduce the ResNet-18 model, and use the hybrid pooling layer to replace the global pooling layer in the ResNet-18 network model to obtain an improved ResNet-18 model;

[0010] S4: inputting the training set into the improved ResNet-18 model for training to obtain a brain tissue DNA methylation prediction model, and optimizing the parameters of the brain tissue DNA methylation prediction model using ten-fold cross validation to obtain an optimized brain tissue DNA methylation prediction model;

[0011] S5: Inputting the test set into the optimized brain tissue DNA methylation prediction model to obtain predicted brain tissue DNA methylation data.

[0012] Furthermore, the ResNet-18 model is introduced, and the global pooling layer in the ResNet-18 network model is replaced with a hybrid pooling layer to obtain an improved ResNet-18 model, including:

[0013] The ResNet-18 model includes an initial convolutional layer, a maximum pooling layer, 8 residual modules, a global pooling layer and a fully connected layer connected in sequence;

[0014] The global pooling layer is replaced by a mixed pooling layer to obtain an improved ResNet-18 model.

[0015] Furthermore, S4 used ten-fold cross validation to optimize the parameters of the brain tissue DNA methylation prediction model, and obtained the optimized brain tissue DNA methylation prediction model, including:

[0016] S41, dividing the training set into 10 subsets on average, randomly selecting one of the subsets as an optimized validation set, and the other 9 subsets as optimized training sets, inputting the brain tissue DNA methylation prediction model, and obtaining a first optimized brain tissue DNA methylation prediction model;

[0017] S42, reselecting a different subset as an optimized validation set, and the other 9 subsets as optimized training sets, inputting the first optimized brain tissue DNA methylation prediction model, and obtaining a second optimized brain tissue DNA methylation prediction model;

[0018] S43. Repeat the above process of selecting a different subset as the optimized validation set and the other 9 subsets as the optimized training sets until the 10th training. The last selected subset is used as the optimized validation set and the other 9 subsets are used as the optimized training sets. The ninth optimized brain tissue DNA methylation prediction model is input to obtain the final optimized brain tissue DNA methylation prediction model.

[0019] Furthermore, the test set is input into the optimized brain tissue DNA methylation prediction model to obtain predicted brain tissue DNA methylation data, including:

[0020] S51, input the test set into the initial convolution layer of the optimized brain tissue DNA methylation prediction model and perform batch normalization processing to obtain the first feature matrix, as shown in formula (1),

[0021] X′=ReLU(BN(Conv2D(X,ω1))) (1)

[0022] Among them, X′ represents the first feature matrix obtained, X represents the cross-tissue feature data matrix extracted from the peripheral tissue DNA methylation data after the convolutional attention mechanism module, ω1 represents the convolution kernel weight matrix of the initial convolution layer, BN is the batch normalization operation, Conv2D is the two-dimensional convolution operation, and ReLU is the activation function;

[0023] S52, input the first feature matrix into the maximum pooling layer to obtain the second feature matrix, as shown in formula (2),

[0024] X1=MaxPool2D(X′,ω2) (2)

[0025] Among them, ω2 represents the convolution kernel weight matrix of the maximum pooling layer, MaxPool2D represents the two-dimensional maximum pooling operation, and X1 represents the generated second feature matrix;

[0026] S53, input X1 into the first residual module to obtain the third feature matrix, then input the third feature matrix into the second residual module to obtain the fourth feature matrix, and so on, input the feature matrix output by the previous residual module into the next residual module to obtain the final predicted brain tissue DNA methylation data.

[0027] Furthermore, the feature matrix output by the previous residual module is input to the next residual module to obtain the final predicted brain tissue DNA methylation data, including:

[0028] S531, X α Input into the αth residual module to obtain the feature map in the αth residual module, as shown in formula (3) and formula (4),

[0029] F(X α )=BN(Conv2D(ReLU(BN(COnv2d(X α ))))) (3)

[0030] Output α =F(X α )+X α (4)

[0031] In formula (3), COnv2D is a two-dimensional convolution operation, BN is a batch normalization operation, ReLU is an activation function, and F(X α ) is the feature map in the αth residual block, X α is the input feature matrix of any residual block; Conv2d represents the two-dimensional convolution operation in the residual module;

[0032] Output in formula (4) α is the output of the αth residual block;

[0033] S532, the output feature matrix of the last residual block is input into the hybrid pooling layer and the fully connected layer to obtain the final prediction value, as shown in formulas (5) and (6),

[0034] X mix =a·Maxpool(X last )+(1-a)·Avgpool(X last ) (5)

[0035] Y = Softmax(FC(X mix )) (6)

[0036] In formula (5), MaxPool is the maximum average pooling operation, AvgPool is the average pooling operation, and X last The output feature matrix of the last residual block, a is the hyperparameter for adjusting the maximum pooling and average pooling, X mix is the output feature vector of the mixed pooling layer; in formula (6), Softmax is the activation function, FC is the fully connected layer, and Y represents the final brain DNA methylation prediction value.

[0037] Furthermore, a convolutional attention mechanism module is used to perform feature extraction on the brain peripheral tissue DNA data to extract a cross-tissue feature data matrix in the peripheral tissue DNA methylation data, including:

[0038] S21. Define the blood methylation data matrix as B i,j , the oral methylation data matrix is ​​O i,j , the salivary methylation data matrix is ​​S i,j, where i represents the sample, i=1,2,...,n, j represents the feature, j=1,2,...,m, and the peripheral tissue DNA methylation data matrix is ​​defined as D i,j , where D∈B,O,S;

[0039] S22. Data matrix D for DNA methylation of peripheral tissues i,j Perform global average pooling and global maximum pooling respectively to obtain a one-dimensional feature vector obtained by global average pooling and a one-dimensional feature vector obtained by global maximum pooling, as shown in formulas (7) and (8).

[0040] F avg =AvgPool(D i,j ) (7)

[0041] F max =MaxPool(D i,j ) (8)

[0042] Among them, F avg and F max The dimensions of are all n×1×c, that is, each channel of the sample has one value; F avg represents global average pooling, F max represents the global maximum pooling; n represents the total number of samples, and c represents the number of channels;

[0043] S23, input both one-dimensional feature vectors into the shared fully connected layers W0 and W1 to obtain two weight vectors, as shown in formulas (9) and (10),

[0044] M avg =σ(W1·ReLU(W0·F avg )) (9)

[0045] M max =σ(W1·ReLU(W0·F max )) (10)

[0046] Among them, σ represents the Sigmoid function, W0 and W1 are the two weight matrices of the fully connected layer; M avg represents the weight vector of the one-dimensional feature vector obtained by global average pooling, M max Represents the one-dimensional feature vector obtained by global maximum pooling;

[0047] S24, add the two weight vectors to obtain the final channel weight matrix, as shown in formula (11),

[0048] M c =σ(M avg +M max ) (11)

[0049] S25, the channel weight matrix M c Application to peripheral tissue DNA methylation data matrix D i,j On each channel of , the channel-weighted output matrix is ​​obtained, as shown in formula (12),

[0050] Z=M c ·D i,j (12)

[0051] S26. Perform feature aggregation on the channel weighted output matrix to obtain two two-dimensional feature vectors, as shown in formulas (13) and (14).

[0052] F′ avg =AvgPool(Z) (13)

[0053] F′ max =MaxPool(Z) (14)

[0054] S27, the two two-dimensional feature vectors are convolved to obtain the spatial attention weight matrix, as shown in formula (15),

[0055] M s =σ(Conv([F′ avg ; F′ max ])) (15)

[0056] Among them, Conv represents the convolution operation;

[0057] S28, the spatial attention weight matrix M s The final output matrix is ​​obtained by using the channel weighted output matrix, as shown in formula (16):

[0058] X=M s ·Z (16)

[0059] Among them, X represents the final output matrix, that is, the cross-tissue feature data matrix in the peripheral tissue DNA methylation data.

[0060] Beneficial effect: The present invention provides a method for predicting brain tissue DNA methylation by using peripheral tissue DNA methylation, by acquiring brain peripheral tissue DNA methylation data, finding DNA methylation data related to brain tissue in the brain peripheral tissue DNA methylation data through a convolutional attention mechanism module, and inputting the data into an improved ResNet-18 model, thereby obtaining brain tissue DNA methylation data that can be obtained based on the brain peripheral tissue DNA methylation data, and obtaining brain tissue DNA methylation data without obtaining brain tissue, the sample acquisition is simple, and the data results are stable and accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0062] Figure 1 A flowchart of a method for predicting brain tissue DNA methylation using peripheral tissue DNA methylation provided by the present invention;

[0063] Figure 2 This is a structural diagram of the improved ResNet-18 provided by the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of 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.

[0065] This embodiment provides a method for predicting brain tissue DNA methylation using peripheral tissue DNA methylation. Figure 1 As shown, including:

[0066] S1: Acquire DNA methylation data of brain peripheral tissues, wherein the brain peripheral tissues include blood tissues, oral tissues and saliva tissues;

[0067] S2: using a convolutional attention mechanism module to perform feature extraction on the brain peripheral tissue DNA methylation data, extracting a cross-tissue feature data matrix from the peripheral tissue DNA methylation data, and dividing the cross-tissue feature data matrix into a training set and a test set;

[0068] S3: Introduce the ResNet-18 model, and use the hybrid pooling layer to replace the global pooling layer in the ResNet-18 network model to obtain an improved ResNet-18 model;

[0069] S4: inputting the training set into the improved ResNet-18 model for training to obtain a brain tissue DNA methylation prediction model, and optimizing the parameters of the brain tissue DNA methylation prediction model using ten-fold cross validation to obtain an optimized brain tissue DNA methylation prediction model;

[0070] S5: Inputting the test set into the optimized brain tissue DNA methylation prediction model to obtain predicted brain tissue DNA methylation data.

[0071] Specifically, the DNA methylation data of brain peripheral tissues are obtained, and the brain peripheral tissues include blood tissues, oral tissues and saliva tissues. The brain DNA methylation data can be inferred based on the peripheral tissue DNA methylation data. The results are accurate and the samples are easy to obtain. Secondly, the convolutional attention mechanism module is used to extract features from the brain peripheral tissue DNA methylation data, extract the cross-tissue feature data matrix in the peripheral tissue DNA methylation data, and divide the cross-tissue feature data into a training set and a test set, which can enhance the expression ability of important features, automatically select features and improve the generalization ability of the model. The ResNet-18 model is introduced, and the mixed pooling layer is used to replace the ResNet-18 The global pooling layer in the network model obtains an improved ResNet-18 model, which can simultaneously extract local significant features and global information, thereby improving the expression ability of the model; the training set is input into the improved ResNet-18 model for training to obtain a brain tissue DNA methylation prediction model, and the parameters of the brain tissue DNA methylation prediction model are optimized using ten-fold cross validation to obtain an optimized brain tissue DNA methylation prediction model. By optimizing the model through ten-fold cross validation, a more accurate optimization model can be obtained, and more accurate brain DNA methylation data can be obtained; finally, the test set is input into the optimized brain tissue DNA methylation prediction model to obtain predicted brain tissue DNA methylation data.

[0072] In a specific embodiment, the scheme for obtaining DNA data of brain peripheral tissues, wherein the brain peripheral tissues include blood tissues, oral tissues and saliva tissues, is:

[0073] The brain peripheral tissue DNA methylation data obtained in this embodiment is derived from GSE111165. By obtaining the brain peripheral tissue DNA methylation data, the brain DNA methylation data can be inferred based on the peripheral tissue DNA methylation data. The obtained results are accurate and the samples are easy to obtain.

[0074] In a specific embodiment, the convolutional attention mechanism module is used to perform feature extraction on the brain peripheral tissue DNA methylation data, extract the cross-tissue feature data matrix in the peripheral tissue DNA methylation data, and divide the cross-tissue feature data matrix into a training set and a test set. The scheme is:

[0075] S21. Define the blood methylation data matrix as B i,j , the oral methylation data matrix is ​​O i,j , the salivary methylation data matrix is ​​S i,j, where i represents the sample, i=1,2,...,n, j represents the feature, j=1,2,...,m, and the peripheral tissue DNA methylation data matrix is ​​defined as D i,j , where D∈B,O,S;

[0076] S22. Data matrix D for DNA methylation of peripheral tissues i,j Perform global average pooling and global maximum pooling respectively to obtain a one-dimensional feature vector obtained by global average pooling and a one-dimensional feature vector obtained by global maximum pooling, as shown in formulas (17) and (18).

[0077] F avg =AvgPool(D i,j ) (17)

[0078] F max =MaxPool(D i,j ) (18)

[0079] Among them, F avg and F max The dimension is n×1×c, that is, each channel of the sample has one value; F avg represents global average pooling, F max represents the global maximum pooling; D i,j The size is n×m×c, where n represents the total number of samples, m represents the total number of features, and c represents the number of channels;

[0080] Global average pooling calculates the average value of each channel, captures the spatial information of the channel as a whole, and reflects the relatively stable or global features of the channel; global maximum pooling calculates the maximum value of each channel and captures the most significant local features of the channel. It can highlight the most representative and important feature points in the local area;

[0081] S23, input both one-dimensional feature vectors into the shared fully connected layers W0 and W1 to obtain two weight vectors, as shown in formulas (19) and (20),

[0082] M avg =σ(X1·ReLU(W0·F avg )) (19)

[0083] M max =σ(W1·ReLU(W0·F max )) (20)

[0084] Among them, σ represents the Sigmoid function, W0 and W1 are the two weight matrices of the fully connected layer; M avg represents the weight vector of the one-dimensional feature vector obtained by global average pooling, M maxRepresents the one-dimensional feature vector obtained by global maximum pooling;

[0085] S24, add the two weight vectors to obtain the final channel weight matrix, as shown in formula (21),

[0086] M c =σ(M avg +M max ) (twenty one)

[0087] S25, the channel weight matrix M c Application to peripheral tissue DNA methylation data matrix D i,j On each channel of , the channel-weighted output matrix is ​​obtained, as shown in formula (22),

[0088] Z=M c ·D i,j (twenty two)

[0089] S26. Perform feature aggregation on the channel-weighted output matrix to obtain two two-dimensional feature vectors, as shown in formulas (23) and (24).

[0090] F′ avg =AvgPool(Z) (23)

[0091] F′ max =MaxPool(Z) (24)

[0092] S27, the two two-dimensional feature vectors are convolved to obtain the spatial attention weight matrix, as shown in formula (25),

[0093] M s =σ(Conv([F′ avg ; F′ max ])) (25)

[0094] Among them, Conv represents the convolution operation, and the convolution kernel size is usually 7x7;

[0095] S28, the spatial attention weight matrix M s The final output matrix is ​​obtained by using the channel weighted output matrix, as shown in formula (26):

[0096] X=M s ·Z (26)

[0097] Among them, X represents the final output matrix, that is, the cross-tissue feature data matrix in the peripheral tissue DNA methylation data;

[0098] S29. Divide the acquired data into a training set and a test set in a ratio of 9:1.

[0099] The convolutional attention mechanism used in this scheme consists of two parts: channel attention module and spatial attention module:

[0100] The channel attention module aims to weight each channel by selecting more important features and generating a channel weight matrix;

[0101] The spatial attention module aims to aggregate the weighted channel weight matrix into a channel information by focusing on more important spatial positions, generate two two-dimensional feature vectors and then perform convolution to generate a spatial weight matrix to weight each position;

[0102] This solution uses the convolutional attention mechanism module to extract peripheral tissue CpG site data that is more relevant to brain tissue from peripheral tissue DNA data, which has the following advantages:

[0103] 1. Enhance the expressiveness of important features:

[0104] For peripheral tissue DNA methylation data, the convolutional attention mechanism can improve the performance of the model in key areas and automatically learn which features in peripheral tissues are most important for predicting brain tissue DNA methylation, thereby reducing the interference of irrelevant information and improving the accuracy and robustness of predictions.

[0105] 2. Automatic feature selection:

[0106] When the DNA methylation data of peripheral tissues is more complex, the convolutional attention mechanism can automatically learn features related to brain tissue methylation from the original data, saving time and improving efficiency.

[0107] 3. Combination of local information and global information:

[0108] Convolutional neural networks are good at capturing local features, while the attention mechanism can enhance the model's expressiveness through the interaction of global information. Combining the two, when processing peripheral tissue DNA methylation data, it is possible to not only capture local features but also integrate global contextual information, thereby improving the generalization ability of the model and helping to better predict the DNA methylation level of brain tissue.

[0109] In a specific embodiment, a ResNet-18 model is introduced, and a hybrid pooling layer is used to replace the global pooling layer in the ResNet-18 network model, and a solution to obtain an improved ResNet-18 model is:

[0110] The ResNet-18 model includes an initial convolutional layer, a maximum pooling layer, 8 residual modules, a global pooling layer and a fully connected layer connected in sequence;

[0111] The size of the convolution kernel of the initial convolution layer is 7*7, and the stride is 2;

[0112] The size of the convolution kernel of the maximum pooling layer is 3*3, the step size is 2, and the padding is 1; the residual module structure is:

[0113] Each residual module contains two convolutional layers: the first and second convolutional layers are 3x3 convolutional kernels with a stride of 1;

[0114] Residual connection: The input of each convolutional layer is directly added to the output to form a residual (skip connection), which can speed up training and improve the performance of the network; the global pooling layer averages the features of each channel;

[0115] Linear layer: used to output the final classification results or regression values;

[0116] The global pooling layer is replaced by the hybrid pooling layer to obtain an improved ResNet-18 model, as shown in Figure 2 shown.

[0117] In this solution, the global pooling layer in the ResNet-18 network model is replaced by a hybrid pooling layer, which can extract local significant features and global information at the same time, thereby improving the expressiveness of the model. The model is mainly composed of eight residual blocks, each of which has two convolutional layers and a residual connection. The data is continuously passed through convolution, batch normalization, activation function and skip connection in each residual block to gradually learn and extract deep features.

[0118] In a specific embodiment, the training set is input into the improved ResNet-18 model for training to obtain a brain tissue DNA methylation prediction model, and the model parameters are optimized using ten-fold cross validation to obtain the optimized brain tissue DNA methylation prediction model:

[0119] S41, dividing the training set into 10 subsets on average, randomly selecting one of the subsets as an optimized validation set, and the other 9 subsets as optimized training sets, inputting the brain tissue DNA methylation prediction model, and obtaining a first optimized brain tissue DNA methylation prediction model;

[0120] S42, reselecting a different subset as an optimized validation set, and the other 9 subsets as optimized training sets, inputting the first optimized brain tissue DNA methylation prediction model, and obtaining a second optimized brain tissue DNA methylation prediction model;

[0121] S43. Repeat the above process of selecting a different subset as the optimized validation set and the other 9 subsets as the optimized training sets until the 10th training. The last selected subset is used as the optimized validation set and the other 9 subsets are used as the optimized training sets. The ninth optimized brain tissue DNA methylation prediction model is input to obtain the final optimized brain tissue DNA methylation prediction model.

[0122] In this embodiment, the learning rate, batch size, loss function and regularization parameter of the model are optimized, and the model is trained with the training set in each cross-validation iteration. The model learns different features on different training sets, and each time it tries to optimize the learned weights or parameters. The model is optimized by ten-fold cross-validation, and each optimization process selects a different subset as the optimization validation set, which can obtain a more accurate optimization model and more accurate brain DNA methylation data.

[0123] In a specific embodiment, the test set is input into the optimized brain tissue DNA methylation prediction model to obtain the predicted brain tissue DNA methylation data by:

[0124] S51, input the test set into the initial convolution layer of the optimized brain tissue DNA methylation prediction model and perform batch normalization processing to obtain the first feature matrix, as shown in formula (27),

[0125] X′=ReLU(BN(Conv2D(X,ω1))) (27)

[0126] Among them, X′ represents the first feature matrix obtained, X represents the cross-tissue feature data matrix extracted from the peripheral tissue DNA methylation data after the convolutional attention mechanism module, ω1 represents the convolution kernel weight matrix of the initial convolution layer, the convolution kernel size is 7×7, BN is the batch normalization operation, Conv2D is the two-dimensional convolution operation, and ReLU is the activation function;

[0127] S52, input the first feature matrix into the maximum pooling layer to obtain the second feature matrix, as shown in formula (28),

[0128] X1=MaxPool2D(X′,ω2) (28)

[0129] Among them, ω2 represents the convolution kernel weight matrix, the convolution kernel size is 3×3, the step size is 2, MaxPool2D represents the two-dimensional maximum pooling operation, and X1 represents the generated feature matrix;

[0130] S53, input X1 into the first residual module to obtain the third feature matrix, and then input the third feature matrix into the second residual module to obtain the fourth feature matrix, and so on, input the feature matrix output by the previous residual module into the next residual module to obtain the final predicted brain tissue DNA methylation data:

[0131] S531, X α The feature matrix input to the αth residual block, where X α is the input feature matrix of any residual block, representing the input into the αth residual block, α = 1, 2, ..., 8. The feature map in the αth residual block is obtained through two convolutional layers and batch normalization, as shown in formulas (29) and (30).

[0132] F(X α )=BN(Conv2D(ReLU(BN(Conv2d(X α ))))) (29)

[0133] Output α =F(X α )+X α (30)

[0134] In formula (29), Conv2D is a two-dimensional convolution operation, BN is a batch normalization operation, ReLU is an activation function, and F(X α ) is the feature map in the αth residual block; X α is the input feature matrix of any residual block; Conv2d represents the two-dimensional convolution operation in the residual module;

[0135] Output in formula (30) α is the output of the αth residual block;

[0136] S532, the output feature matrix of the last residual block is input into the hybrid pooling layer and the fully connected layer to obtain the final prediction value, as shown in formulas (31) and (32),

[0137] X mix =a·Maxpool(X last )+(1-a)·Avgpool(X last ) (31)

[0138] Y = Softmax(FC(X mix )) (32)

[0139] In formula (31), MaxPool is the maximum average pooling operation, AvgPool is the average pooling operation, and X lastThe output feature matrix of the last residual block, a is the hyperparameter for adjusting the maximum pooling and average pooling, X mix is the output feature vector of the mixed pooling layer; in formula (32), Softmax is the activation function, FC is the fully connected layer, and Y represents the final brain DNA methylation prediction value.

[0140] The performance of the optimized brain tissue DNA methylation prediction model was evaluated. In this scheme, mean absolute error (MAE) and mean square error (MSE) were used for evaluation. The results are shown in Table 1:

[0141] Table 1 Evaluation performance table

[0142] MAE MSE Blood tissue 0.111 0.042 Oral tissue 0.121 0.049 Salivary tissue 0.107 0.036 Blood, oral cavity, saliva 0.037 0.003

[0143] As can be seen from Table 1, the MAE score of using blood tissue to predict brain tissue DNA methylation is 0.111, and the MSE score is 0.042. The MAE score of using oral tissue to predict brain tissue DNA methylation is 0.121, and the MSE score is 0.049. The MAE score of using saliva tissue to predict brain tissue DNA methylation is 0.107, and the MSE score is 0.036. However, the MAE score of multiple peripheral tissues (blood, oral, and saliva tissues) to predict brain tissue DNA methylation is 0.037, and the MSE score is 0.003.

[0144] In summary, using multiple peripheral tissues to predict brain tissue DNA methylation results has the smallest error and is more accurate.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting brain tissue DNA methylation using peripheral tissue DNA methylation, characterized in that: include: S1: Acquire DNA methylation data of brain peripheral tissues, wherein the brain peripheral tissues include blood tissues, oral tissues and saliva tissues; S2: using a convolutional attention mechanism module to perform feature extraction on the brain peripheral tissue DNA methylation data, extracting a cross-tissue feature data matrix from the peripheral tissue DNA methylation data, and dividing the cross-tissue feature data matrix into a training set and a test set; S3: Introduce the ResNet-18 model, and use the hybrid pooling layer to replace the global pooling layer in the ResNet-18 network model to obtain an improved ResNet-18 model; S4: inputting the training set into the improved ResNet-18 model for training to obtain a brain tissue DNA methylation prediction model, and optimizing the parameters of the brain tissue DNA methylation prediction model using ten-fold cross validation to obtain an optimized brain tissue DNA methylation prediction model; S5: Inputting the test set into the optimized brain tissue DNA methylation prediction model to obtain predicted brain tissue DNA methylation data.

2. The method of claim 1, wherein the method comprises: The ResNet-18 model is introduced, and the global pooling layer in the ResNet-18 network model is replaced with a hybrid pooling layer to obtain an improved ResNet-18 model, including: The ResNet-18 model includes an initial convolutional layer, a maximum pooling layer, 8 residual modules, a global pooling layer and a fully connected layer connected in sequence; The global pooling layer is replaced by a mixed pooling layer to obtain an improved ResNet-18 model.

3. The method of predicting brain tissue DNA methylation using peripheral tissue DNA methylation according to claim 1, characterized in that: S4 uses ten-fold cross validation to optimize the parameters of the brain tissue DNA methylation prediction model and obtains the optimized brain tissue DNA methylation prediction model, including: S41, dividing the training set into 10 subsets on average, randomly selecting one of the subsets as an optimized validation set, and the other 9 subsets as optimized training sets, inputting the brain tissue DNA methylation prediction model, and obtaining a first optimized brain tissue DNA methylation prediction model; S42, reselecting a different subset as an optimized validation set, and the other 9 subsets as optimized training sets, inputting the first optimized brain tissue DNA methylation prediction model, and obtaining a second optimized brain tissue DNA methylation prediction model; S43. Repeat the above process of selecting a different subset as the optimized validation set and the other 9 subsets as the optimized training sets until the 10th training. The last selected subset is used as the optimized validation set and the other 9 subsets are used as the optimized training sets. The ninth optimized brain tissue DNA methylation prediction model is input to obtain the final optimized brain tissue DNA methylation prediction model.

4. The method of claim 1, wherein the method comprises: Inputting the test set into the optimized brain tissue DNA methylation prediction model to obtain predicted brain tissue DNA methylation data, including: S51, input the test set into the initial convolution layer of the optimized brain tissue DNA methylation prediction model and perform batch normalization processing to obtain the first feature matrix, as shown in formula (1), X′=ReLU(BN(Conv2D(X,ω1))) (1) Among them, X′ represents the first feature matrix obtained, X represents the cross-tissue feature data matrix extracted from the peripheral tissue DNA methylation data after the convolutional attention mechanism module, ω1 represents the convolution kernel weight matrix of the initial convolution layer, BN is the batch normalization operation, Conv2D is the two-dimensional convolution operation, and ReLU is the activation function; S52, input the first feature matrix into the maximum pooling layer to obtain the second feature matrix, as shown in formula (2), X1=MaxPool2D(X′,ω2) (2) Among them, ω2 represents the convolution kernel weight matrix of the maximum pooling layer, MaxPool2D represents the two-dimensional maximum pooling operation, and X1 represents the generated second feature matrix; S53, input X1 into the first residual module to obtain the third feature matrix, then input the third feature matrix into the second residual module to obtain the fourth feature matrix, and so on, input the feature matrix output by the previous residual module into the next residual module to obtain the final predicted brain tissue DNA methylation data.

5. The method of predicting brain tissue DNA methylation using peripheral tissue DNA methylation according to claim 4, characterized in that: The feature matrix output by the previous residual module is input to the next residual module to obtain the final predicted brain tissue DNA methylation data, including: S531, X α Input into the αth residual module to obtain the feature map in the αth residual module, as shown in formula (3) and formula (4), F(X α )=BN(Conv2D(ReLU(BN(Conv2d(X α ))))) (3) Output α =F(X α )+X α (4) In formula (3), Conv2D is a two-dimensional convolution operation, BN is a batch normalization operation, ReLU is an activation function, and F(X α ) is the feature map in the αth residual block, X α is the input feature matrix of any residual block; Conv2d represents the two-dimensional convolution operation in the residual module; Output in formula (4) α is the output of the αth residual block; S532, the output feature matrix of the last residual block is input into the hybrid pooling layer and the fully connected layer to obtain the final prediction value, as shown in formulas (5) and (6), X mix =a·Maxpool(X last )+(1-a)·Avgpool(X last ) (5) Y=Softmax(FC(X mix ))(6) In formula (5), MaxPool is the maximum average pooling operation, AvgPool is the average pooling operation, and X last The output feature matrix of the last residual block, a is the hyperparameter for adjusting the maximum pooling and average pooling, X mix is the output feature vector of the mixed pooling layer; in formula (6), Softmax is the activation function, FC is the fully connected layer, and Y represents the final brain DNA methylation prediction value.

6. The method of claim 1, wherein the method comprises: Using a convolutional attention mechanism module to extract features from the brain peripheral tissue DNA data, extracting a cross-tissue feature data matrix from the peripheral tissue DNA methylation data, including: S21. Define the blood methylation data matrix as B i , the oral methylation data matrix is, and the salivary methylation data matrix is ​​S i,j , where i represents the sample, i=1,2,...,n, j represents the feature, j=1,2,...,m, and the peripheral tissue DNA methylation data matrix is ​​defined as D i,j , where D∈B,O,S; S22. Data matrix D for DNA methylation of peripheral tissues i,j Perform global average pooling and global maximum pooling respectively to obtain a one-dimensional feature vector obtained by global average pooling and a one-dimensional feature vector obtained by global maximum pooling, as shown in formulas (7) and (8). F avg =AvgPool(D i，j ) (7) F max =MaxPool(D i，j ) (8) Among them, F avg and F max The dimensions of are all n×1×c, that is, each channel of the sample has one value; F avg represents global average pooling, F max represents the global maximum pooling; n represents the total number of samples, and c represents the number of channels; S23, input both one-dimensional feature vectors into the shared fully connected layers W0 and W1 to obtain two weight vectors, as shown in formulas (9) and (10), M avg =σ(W1·ReLU(W0·F avg )) (9) M max =σ(W1·ReLU(W0·F max )) (10) Among them, σ represents the Sigmoid function, W0 and W1 are the two weight matrices of the fully connected layer; M avg represents the weight vector of the one-dimensional feature vector obtained by global average pooling, M max Represents the one-dimensional feature vector obtained by global maximum pooling; S24, add the two weight vectors to obtain the final channel weight matrix, as shown in formula (11), M c =σ(M avg +M max ) (11) S25, the channel weight matrix M c Application to peripheral tissue DNA methylation data matrix D i，j On each channel of , the channel-weighted output matrix is ​​obtained, as shown in formula (12), Z=M c ·D i，j (12) S26. Perform feature aggregation on the channel weighted output matrix to obtain two two-dimensional feature vectors, as shown in formulas (13) and (14). F′ avg =AvgPool(Z) (13) F′ max =MaxPool(Z) (14) S27, the two two-dimensional feature vectors are convolved to obtain the spatial attention weight matrix, as shown in formula (15), M s =σ(Conv([F′ avg ;F′ max ])) (15) Among them, Conv represents the convolution operation; S28, the spatial attention weight matrix M s The final output matrix is ​​obtained by using the channel weighted output matrix, as shown in formula (16): X=M s ·Z (16) Among them, X represents the final output matrix, that is, the cross-tissue feature data matrix in the peripheral tissue DNA methylation data.

Images