LncRNA-disease association prediction method based on comparative learning

Abstract

The invention provides an LncRNA-disease association prediction method based on comparative learning. Comprising the following steps: S1, preparing an LncRNA-disease incidence matrix, calculating functional similarity, cosine similarity and Gaussian interaction spectrum kernel similarity of lncRNA, and calculating semantic similarity, cosine similarity and Gaussian interaction spectrum kernel similarity of diseases; s2, constructing an LncRNA similarity graph, a disease similarity graph and an LncRNA disease association graph; s3, obtaining an LncRNA feature matrix, an LncRNA auxiliary feature matrix, a disease feature matrix and a disease auxiliary feature matrix; s4, obtaining an LncRNA personalized fusion feature and a disease personalized fusion feature; and S5, carrying out cross-view comparative learning on the personalized fusion features of the LncRNA and the personalized fusion features of the diseases, and predicting potential LncRNA and disease association. According to the method, information fused into bioinformatics is considered, so that the method has certain value for predicting the potential correlation between the LncRNA and the diseases, improving the accuracy of disease diagnosis and treatment and promoting the development of drug research and development.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to a LncRNA-disease association prediction method based on contrastive learning. Background Art

[0002] Non-coding RNA (ncRNA) is mainly divided into small RNA, medium RNA and long non-coding RNA (LncRNA) according to the number of nucleotides. Among them, long non-coding RNA plays a very important role in the regulation of disease expression in animals and plants. Current biological studies have shown that in the development of lung adenocarcinoma, long non-coding RNA HAGLR inhibits tumor growth, so this LncRNA is confirmed as a tumor suppressor. At the same time, in the study of esophageal cancer, LncRNA (ADAMTS9-AS2) can effectively inhibit the proliferation, invasion and migration of cancer cells. It can be seen that the study of the potential association between LncRNA and disease is helpful for the prevention, detection and treatment of related diseases in humans. Traditional biological experimental methods have high accuracy in discovering potential correlations, but the process is complicated and time-consuming. Therefore, it is of great significance to develop an efficient and convenient calculation method based on the current rapid development of computer technology to detect the association between LncRNA and disease. For example, the authorized patents CN113160880B, CN115249538B, CN112289373B, etc.

[0003] The current method for detecting the association between LncRNA and disease is to extract low-dimensional features of LncRNA and disease through graph convolutional neural networks or stacked autoencoders, splice LncRNA and disease nodes, and use classifiers such as random forest and xgboost for prediction. There is also a method of using matrix decomposition. By decomposing the original input matrix, a low-dimensional matrix that can approximate the matrix is ​​found, and then the missing associated data in the original matrix is ​​filled.

[0004] However, existing methods have certain shortcomings. The performance of graph neural network-based methods in predicting unknown data under the framework of supervised learning is reduced because the known association matrix is ​​too sparse. In addition, many graph neural network-based methods only consider the relationship between LncRNA and disease for encoding, but do not incorporate rich biological semantic information into the representation of node embedding, nor consider personalized enhancement of the embedding representation of specific LncRNA and specific diseases. Summary of the invention

[0005] According to the above technical problem that the performance of the existing method for predicting unknown data will decrease, a LncRNA-disease association prediction method based on contrastive learning is provided. The present invention mainly uses contrastive self-supervised learning to perform data enhancement through unlabeled data itself to solve the data sparsity problem. The combination of contrastive learning and graph neural network can enhance the robustness of node embedding in the case of sparsely associated graphs.

[0006] The technical means adopted by the present invention are as follows:

[0007] A LncRNA-disease association prediction method based on contrastive learning comprises the following steps:

[0008] S1. Prepare the lncRNA-disease association matrix and calculate the functional similarity, cosine similarity and Gaussian interaction spectrum kernel similarity of lncRNA. Calculate the semantic similarity, cosine similarity and Gaussian interaction spectrum kernel similarity of diseases;

[0009] S2, constructing LncRNA similarity graph, disease similarity graph and LncRNA disease association graph;

[0010] S3. Using GCN-based heterogeneous graph relational learning to integrate the biological similarities of LncRNA and disease dependencies to increase the embedding representation extracted from the LncRNA-disease association matrix; obtaining the LncRNA feature matrix, the LncRNA auxiliary feature matrix, the disease feature matrix, and the disease auxiliary feature matrix;

[0011] S4. Based on the LncRNA feature matrix, the LncRNA auxiliary feature matrix, the disease feature matrix and the disease auxiliary feature matrix, personalized feature fusion is performed from the auxiliary information to generate personalized embedded expressions of specific LncRNA and specific diseases; personalized fusion features of LncRNA and personalized fusion features of diseases are obtained;

[0012] S5. Perform cross-view comparative learning on lncRNA personalized fusion features and disease personalized fusion features to predict potential lncRNA and disease associations.

[0013] Furthermore, S2 specifically includes the following steps:

[0014] The average value of each position of the three similarity matrices of LncRNA functional similarity, cosine similarity and Gaussian interaction spectrum kernel similarity is taken, and the obtained matrix is ​​used as the matrix representation of the LncRNA similarity map, and the normalized matrix is ​​used as the feature of the LncRNA similarity map;

[0015] The three similarity matrices of disease functional similarity, cosine similarity, and Gaussian interaction spectral kernel similarity also take the method of taking the average value at each position. The obtained matrix is ​​used as the matrix representation of the disease similarity map, and the normalized matrix is ​​used as the feature of the disease similarity map.

[0016] Furthermore, S3 specifically includes the following steps:

[0017] Three specific encoding layers are used, each containing three ordinary GCN encoders to extract node embeddings of LncRNA similarity graph, LncRNA-disease association graph, and disease similarity graph, respectively;

[0018] The lncRNA-disease association graph was passed through the GCN encoder to obtain the lncRNA feature matrix and the disease feature matrix;

[0019] After the LncRNA similarity graph passes through the GCN encoder, the LncRNA auxiliary feature matrix is ​​generated. The LncRNA auxiliary feature matrix and the LncRNA feature matrix are average-pooled to generate the LncRNA fusion feature. The LncRNA fusion feature is passed to the next layer of encoder as the feature of the LncRNA similarity graph and the feature of the LncRNA part in the LncRNA disease association graph;

[0020] The disease similarity graph generates a disease auxiliary feature matrix after passing through the GCN encoder; the disease auxiliary feature matrix and the disease feature matrix are average-pooled to fuse and generate disease fusion features, and the disease fusion features are passed to the next layer of encoder as the features of the disease similarity graph and the features of the disease part in the disease-disease association graph;

[0021] The LncRNA features and disease features obtained in each coding layer are added up and then averaged to obtain the overall feature representation of LncRNA and disease; the LncRNA auxiliary features and disease auxiliary features obtained in each layer are added up and then averaged to obtain the overall auxiliary feature representation of LncRNA and disease.

[0022] Furthermore, S4 specifically includes the following steps:

[0023] The features of all diseases related to the target LncRNA are mean-pooled, and then spliced ​​with the features of the LncRNA and its auxiliary features, and the personalized features of the LncRNA are generated through ordinary MLP; the auxiliary features of the LncRNA and its personalized features are mean-pooled again to obtain the personalized fusion features of the LncRNA; the LncRNA features and the LncRNA personalized fusion features are mean-pooled to obtain the LncRNA features used for prediction;

[0024] The features of all LncRNAs related to the target disease are mean-pooled, and then concatenated with the features of the disease and its auxiliary features to generate personalized features of the disease through the MLP layer; the auxiliary features of the disease and its personalized features are mean-pooled again to obtain personalized fusion features of the disease; the disease and the personalized fusion features of the disease are mean-pooled to obtain the disease features for prediction.

[0025] Furthermore, S5 specifically includes the following steps:

[0026] Comparative learning: For LncRNA, the personalized fusion features of LncRNA and LncRNA features are compared and learned; for diseases, the personalized fusion features of diseases and disease features are compared and learned;

[0027] The inner product of the LncRNA features used for prediction and the disease features used for prediction is performed to obtain the final predicted association score.

[0028] Furthermore, the binary cross entropy loss function is used to calculate the loss between the true value and the predicted association value, and the contrastive learning loss (InfoNEC loss) is used to calculate the contrastive loss of the personalized fusion features of LncRNA and the LncRNA features, and the contrastive loss of the personalized fusion features of the disease and the disease features. By calculating the overall loss, i.e., the contrastive learning loss of LncRNA and the disease and the binary cross entropy loss, the internal parameters of the model are further optimized using the gradient descent method.

[0029] The present invention also provides a storage medium, which includes a stored program, wherein when the program is run, any of the above-mentioned LncRNA-disease association prediction methods based on contrastive learning is executed.

[0030] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes any of the above-mentioned LncRNA-disease association prediction methods based on comparative learning through the operation of the computer program.

[0031] Compared with the prior art, the present invention has the following advantages:

[0032] The technical solution provided by the present invention first calculates multiple similarities between LncRNA and diseases, and fuses them into similarity networks of LncRNA and diseases respectively. Subsequently, heterogeneous network information extraction and fusion are performed, and auxiliary feature information is integrated into the final feature to enhance the embedded representation. Then, personalized cross-view feature fusion is performed on specific LncRNA and specific diseases to further enhance the embedded representation. The corresponding features of LncRNA and disease are compared and learned and then predicted. The present invention not only relies on known associations, but also considers the integration of bioinformatics information, which is of certain value for predicting the potential association between LncRNA and diseases, improving the accuracy of disease diagnosis and treatment, and promoting the progress of drug research and development. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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.

[0034] Figure 1 The figure is a flow chart of the method of the present invention.

[0035] Figure 2 Schematic diagram of using GCN to extract and fuse heterogeneous information in the present invention.

[0036] Figure 3 This is a schematic diagram of personalized feature fusion from auxiliary information in the present invention. DETAILED DESCRIPTION

[0037] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme 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 only 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 should fall within the scope of protection of the present invention.

[0038] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0039] like Figure 1 As shown, the present invention provides a LncRNA-disease association prediction method based on contrastive learning, comprising the following steps:

[0040] S1. Obtain the LncRNA-disease association matrix. For diseases, first obtain the GO annotations of each disease, and then use the classic Wang method to calculate the semantic similarity of the disease; use the Gaussian interaction attribute kernel calculation formula to calculate the Gaussian similarity of the disease through the obtained LncRNA-disease association matrix, and use the cosine similarity calculation formula to calculate the cosine similarity of the disease. For lncRNA, use the semantic similarity of the disease and the obtained LncRNA-disease association matrix to calculate the functional similarity of LncRNA. Use the Gaussian interaction attribute kernel function to calculate the Gaussian similarity of LncRNA through the obtained LncRNA-disease association matrix, and use the cosine similarity calculation formula to calculate the cosine similarity of LncRNA.

[0041] S2. Construct the corresponding similarity graph based on the existing similarity matrix of LncRNA and disease. The LncRNA association matrix is ​​the underlying data of the LncRNA association graph.

[0042] The average value of each position of the three similarity matrices of LncRNA functional similarity, cosine similarity and Gaussian interaction spectrum kernel similarity is taken, and the obtained matrix is ​​used as the matrix representation of the LncRNA similarity map. The normalized matrix is ​​used as the feature of the LncRNA similarity map.

[0043] The three similarity matrices of disease functional similarity, cosine similarity, and Gaussian interaction spectrum kernel similarity are also averaged at each position, and the obtained matrices are used as matrix representations of disease similarity maps. Similarly, the normalized matrices are used as features of disease similarity maps.

[0044] S3. We use GCN-based heterogeneous graph relation learning to integrate the biological similarities of LncRNA and disease dependencies to enhance the embedding representation extracted from the LncRNA-disease association matrix, as Figure 2 shown.

[0045] Existing GCN-based methods only extract features based on a single interaction graph, and do not incorporate information from different views, so they cannot extract the features of LncRNA and disease well. Therefore, we use GCN to extract LncRNA and disease while incorporating heterogeneous information to enhance the embedding representation. Specifically, three GCN encoders are used to extract node embeddings of LncRNA similarity graph, LncRNA disease association graph, and disease similarity graph, respectively. Taking LncRNA as an example, the LncRNA similarity graph generates the LncRNA auxiliary feature matrix after passing through the GCN encoder, and the LncRNA disease association graph obtains the feature matrix of LncRNA and disease after passing through the GCN encoder. We average pool the LncRNA auxiliary feature matrix and the LncRNA feature matrix to fuse and generate the LncRNA fusion feature, and pass the LncRNA fusion feature to the next layer of encoder as the feature of the LncRNA similarity graph and the feature of the LncRNA part of the LncRNA disease association graph. The same method is used for feature extraction and feature fusion for the disease part.

[0046] In order to make full use of the representation of each layer of encoding, a multi-level information aggregation method is used. Specifically, the LncRNA features and disease features obtained in each layer are added and then averaged to obtain the overall feature representation of LncRNA and disease. Similarly, the overall auxiliary features of LncRNA features and diseases can be obtained.

[0047] S4. Perform personalized feature fusion from auxiliary information to generate personalized embedded expressions of specific LncRNAs and specific diseases, such as Figure 3 As shown;

[0048] Previous model methods are based on the fact that similar lncRNAs are often associated with similar diseases. However, some lncRNAs may be associated with diseases under the joint action of other lncRNAs. Therefore, personalized embedding representation fusion from auxiliary features is very necessary. We designed a cross-view fusion method to achieve personalized feature extraction and fusion of lncRNAs and diseases.

[0049] Taking LncRNA as an example, the features of all diseases related to a certain LncRNA are averaged and pooled, and then spliced ​​with the features of the LncRNA and its auxiliary features, and the personalized features of the LncRNA are generated through the MLP layer. The auxiliary features of the LncRNA and its personalized features are averaged and pooled again to obtain the personalized fusion features of the LncRNA. Finally, the LncRNA features and the personalized fusion features of the LncRNA are averaged and pooled to obtain the LncRNA features for prediction. Similarly, for diseases, the corresponding personalized fusion features of the disease and the disease features for prediction can be obtained according to similar methods.

[0050] S5. Cross-view comparative learning and prediction of potential LncRNA and disease associations;

[0051] In the comparative learning part, for LncRNA, the personalized fusion features of LncRNA and LncRNA features are compared and learned. For diseases, the personalized fusion features of diseases and disease features are compared and learned. The predicted LncRNA features and the predicted disease features are inner-producted to get the final predicted association score. The binary cross entropy loss function is used to calculate the loss between the true value and the predicted association value. The comparative loss of the personalized fusion features of LncRNA and LncRNA features, and the comparative loss of the personalized fusion features of diseases and disease features are calculated using the comparative learning loss (InfoNEC loss). By calculating the overall loss, i.e., the comparative learning loss of LncRNA and disease and the binary cross entropy loss, the gradient descent method is used to further optimize the parameters within the model.

[0052] Then the steps of testing the models built by S2~S5 are also included:

[0053] Based on the known associations between LncRNA and diseases, an equal amount of unknown associations are selected to balance the sample data as the true labels of the samples. After the data is extracted and fused by heterogeneous information in the model, personalized embedding features, and comparative learning, the association probability prediction value is finally obtained. The AUC value and AUPR value are calculated by the predicted label and the actual label to measure the prediction performance of the model.

[0054] Test various parameters and important modules that affect model performance. Observe the impact of the dimension of embedded features and the number of GCN layers in the model on the model test results. Develop modules to remove LncRNA or disease similarity graphs, or modules to remove personalized embedded features, or comparative learning modules, and test the results of models containing different modules.

[0055] The present invention also provides a storage medium, which includes a stored program, wherein when the program is run, a LncRNA-disease association prediction method based on contrastive learning is executed.

[0056] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes a LncRNA-disease association prediction method based on comparative learning through the computer program.

[0057] 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 LncRNA-disease association prediction method based on contrastive learning, characterized in that: The steps include: S1. Prepare the lncRNA-disease association matrix and calculate the functional similarity, cosine similarity and Gaussian interaction spectrum kernel similarity of lncRNA. Calculate the semantic similarity, cosine similarity and Gaussian interaction spectrum kernel similarity of diseases; S2, constructing LncRNA similarity graph, disease similarity graph and LncRNA disease association graph; S3, using GCN-based heterogeneous graph relational learning to integrate the biological similarities of lncRNA and disease dependencies to enhance the embedding representation extracted from the lncRNA-disease association matrix; Obtaining LncRNA feature matrix, LncRNA auxiliary feature matrix, disease feature matrix and disease auxiliary feature matrix; S4, based on the lncRNA feature matrix, the lncRNA auxiliary feature matrix, the disease feature matrix and the disease auxiliary feature matrix, personalized feature fusion is performed from the auxiliary information to generate personalized embedded expressions of specific lncRNAs and specific diseases; Obtain personalized LncRNA fusion signatures and disease personalized fusion signatures; S5. Perform cross-view comparative learning on lncRNA personalized fusion features and disease personalized fusion features to predict potential lncRNA and disease associations.

2. The LncRNA-disease association prediction method based on contrastive learning according to claim 1, characterized in that: S2 specifically includes the following steps: The average value of each position of the three similarity matrices of LncRNA functional similarity, cosine similarity and Gaussian interaction spectrum kernel similarity is taken, and the obtained matrix is ​​used as the matrix representation of the LncRNA similarity map, and the normalized matrix is ​​used as the feature of the LncRNA similarity map; The three similarity matrices of disease functional similarity, cosine similarity, and Gaussian interaction spectral kernel similarity also take the method of taking the average value at each position. The obtained matrix is ​​used as the matrix representation of the disease similarity map, and the normalized matrix is ​​used as the feature of the disease similarity map.

3. The LncRNA-disease association prediction method based on contrastive learning according to claim 1, characterized in that: S3 specifically includes the following steps: Three specific encoding layers are used, each containing three GCN encoders to extract node embeddings of the LncRNA similarity graph, LncRNA-disease association graph, and disease similarity graph, respectively; The lncRNA-disease association graph was passed through the GCN encoder to obtain the lncRNA feature matrix and the disease feature matrix; After the LncRNA similarity graph passes through the GCN encoder, the LncRNA auxiliary feature matrix is ​​generated. The LncRNA auxiliary feature matrix and the LncRNA feature matrix are average-pooled to generate the LncRNA fusion feature. The LncRNA fusion feature is passed to the next layer of encoder as the feature of the LncRNA similarity graph and the feature of the LncRNA part in the LncRNA disease association graph; The disease similarity graph generates a disease auxiliary feature matrix after passing through the GCN encoder; the disease auxiliary feature matrix and the disease feature matrix are average-pooled to fuse and generate disease fusion features, and the disease fusion features are passed to the encoder of the next encoding layer as the features of the disease similarity graph and the features of the disease part in the disease-disease association graph; The LncRNA features and disease features obtained in each coding layer are added up and then averaged to obtain the overall feature representation of LncRNA and disease; the LncRNA auxiliary features and disease auxiliary features obtained in each layer are added up and then averaged to obtain the overall auxiliary feature representation of LncRNA and disease.

4. The LncRNA-disease association prediction method based on contrastive learning according to claim 1, characterized in that: S4 specifically includes the following steps: The features of all diseases related to the target LncRNA are mean-pooled, and then spliced ​​with the features of the LncRNA and its auxiliary features, and the personalized features of the LncRNA are generated through MLP; the auxiliary features of the LncRNA and its personalized features are mean-pooled again to obtain the personalized fusion features of the LncRNA; the LncRNA features and the LncRNA personalized fusion features are mean-pooled to obtain the LncRNA features used for prediction; The features of all LncRNAs related to the target disease are mean-pooled, and then concatenated with the features of the disease and its auxiliary features to generate personalized features of the disease through the MLP layer; the auxiliary features of the disease and its personalized features are mean-pooled again to obtain personalized fusion features of the disease; the disease and the personalized fusion features of the disease are mean-pooled to obtain the disease features for prediction.

5. The LncRNA-disease association prediction method based on contrastive learning according to claim 1, characterized in that: S5 specifically includes the following steps: Comparative learning: For LncRNA, the personalized fusion features of LncRNA and LncRNA features are compared and learned; for diseases, the personalized fusion features of diseases and disease features are compared and learned; The inner product of the LncRNA features used for prediction and the disease features used for prediction is performed to obtain the final predicted association score.

6. The LncRNA-disease association prediction method based on contrastive learning according to claim 1, characterized in that: The binary cross entropy loss function is used to calculate the loss between the true value and the predicted association value. The contrastive learning loss is used to calculate the contrastive loss of the personalized fusion features of LncRNA and the LncRNA features, as well as the contrastive loss of the personalized fusion features of the disease and the disease features. By calculating the overall loss, i.e., the contrastive learning loss of LncRNA and disease and the binary cross entropy loss, the gradient descent method is used to further optimize the parameters within the model.

7. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is run, the LncRNA-disease association prediction method based on contrastive learning according to any one of claims 1 to 6 is executed.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor executes the LncRNA-disease association prediction method based on contrastive learning according to any one of claims 1 to 6 through the operation of the computer program.

Images