Microorganism-disease association prediction method based on meta-path

Abstract

The invention provides a microbe-disease association prediction method based on a meta-path. The method comprises the following steps: S1, calculating similarities of a disease, a microbe and a drug respectively; s2, constructing a heterogeneous graph containing microorganism nodes, disease nodes and drug nodes based on the isomorphic networks of microorganisms, diseases and drugs; s3, feature representation of the microbial nodes and the disease nodes in the same meta-path is aggregated through attention in the meta-path; s4, obtaining final representations of the microbial nodes and the disease nodes; and S5, predicting the association probability of the microorganism and disease pair, and distinguishing whether there is an association relationship or no association relationship. According to the method, not only are feature representations from different similarity networks fused by adopting the MOSFEL, but also structure and semantic information of nodes are introduced through a meta path, and the information is adaptively aggregated through local attention and global attention, so that the method has certain value for actual disease diagnosis and treatment and subsequent 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 meta-path-based microbe-disease association prediction method. Background Art

[0002] With the development of high-throughput sequencing technology and microbiome, many studies have shown that there is an inseparable relationship between microorganisms and human diseases. Microorganisms affect human diseases and physical health by participating in important life processes such as metabolism and apoptosis, such as cardiovascular disease, autoinflammatory disease, and cancer. At present, the relationship between microorganisms and diseases is mainly divided into four categories. One is that changes in microorganisms are the direct cause of the disease; another is that there is no direct correlation between microorganisms and diseases, but there is a common third-party factor; the third is that the disease can affect the changes in microorganisms so that they can be used as biomarkers for disease diagnosis; the last category is that the disease can affect the changes in microorganisms, which in turn further affects the severity of the disease. Therefore, confirming the relationship between microorganisms and diseases helps to deepen the understanding of disease mechanisms and improve the efficiency of disease diagnosis and treatment. Such as the authorized patent CN109920478B.

[0003] Prediction of microbial-disease association helps to deeply understand the disease mechanism and promote drug development. However, traditional biological experimental methods consume a lot of time and financial resources and cannot meet the needs of high-throughput data. The latest solution is to use similarity calculation methods to calculate the similarity relationship between microorganisms and diseases to construct a similarity network, and then predict the association between microorganisms and diseases through a similarity network model. In addition, some methods use graph neural networks to learn the characteristic representation of microorganisms and diseases from microbial isomorphic networks, disease isomorphic networks, or microbial-disease heterogeneous networks for predicting associations. Among the methods for predicting microbial-disease associations, methods based on graph neural networks can well predict the correlation between microorganisms and diseases. The similarity information of microorganisms and diseases plays a very important role as the initial feature of nodes, but they only obtain comprehensive similarity through simple processing of multiple similarities, and cannot obtain high-quality similarity; in addition, they only use graph neural networks to aggregate the features of nodes, and lack consideration of the features of edges in the graph. Summary of the invention

[0004] According to the above technical problem that the existing methods only obtain comprehensive similarity through simple processing of multiple similarities and cannot obtain high-quality similarity, a meta-path-based microorganism-disease association prediction method is provided.

[0005] The technical means adopted by the present invention are as follows:

[0006] A method for predicting microorganism-disease association based on metapathway comprises the following steps:

[0007] S1. Calculate the similarity of diseases, microorganisms and drugs respectively;

[0008] S2. Based on the calculation of the similarities of diseases, microorganisms and drugs, multiple similarity views are used to construct isomorphic networks of microorganisms, diseases and drugs to enrich the node similarity information, and a heterogeneous graph containing microorganism nodes, disease nodes and drug nodes is constructed based on the isomorphic networks of microorganisms, diseases and drugs;

[0009] S3, encoding the meta-path instances corresponding to the four meta-paths of microbe-disease-microbe, microbe-drug-microbe, disease-microbe-disease, and disease-drug-disease in the heterogeneous graph, and aggregating the feature representations of microbe nodes and disease nodes under the same meta-path through attention within the meta-path;

[0010] S4. Based on the attention mechanism between meta-paths, the feature representations of microbial nodes and disease nodes under different meta-paths are aggregated through local attention and global attention to obtain the final representations of microbial nodes and disease nodes.

[0011] S5. Predict the probability of association between microorganisms and diseases, and distinguish whether there is an association or no association.

[0012] Furthermore, in S1:

[0013] Disease similarity includes semantic similarity of diseases, Gaussian interaction spectrum kernel similarity, cosine similarity, and sigmoid kernel function similarity;

[0014] Microbial similarity includes microbial functional similarity, Gaussian interaction spectrum kernel similarity, cosine similarity, and sigmoid kernel function similarity;

[0015] Drug similarity includes drug functional similarity, Gaussian interaction spectrum kernel similarity, cosine similarity, and sigmoid kernel function similarity.

[0016] Furthermore, S2 specifically includes the following steps:

[0017] S21. Taking microorganisms, diseases and drugs as nodes and different similarity views as the adjacency matrix of the isomorphic networks of microorganisms, diseases and drugs respectively, construct a functional similarity network, a Gaussian kernel similarity network, a cosine similarity network and a sigmoid kernel function similarity network containing only microorganism nodes, and construct a semantic similarity network, a Gaussian kernel similarity network, a cosine similarity network and a sigmoid kernel function similarity network containing only disease nodes, as well as a functional similarity network, a Gaussian kernel similarity network, a cosine similarity network and a sigmoid kernel function similarity network of drug nodes;

[0018] S22. Utilize multi-order similarity fusion learning, combine low-order global learning and high-order feature learning, and construct a multi-order similarity fusion network through optimization technology to obtain richer and more reliable structural information and high-order similarity information between nodes, and construct a microorganism-drug-disease heterogeneous network.

[0019] Furthermore, S3 specifically includes the following steps:

[0020] S31. When learning node feature representation, the features of the edges in the graph are introduced by means of meta-paths, so that the model can learn high-quality node feature representation; for heterogeneous graphs, the following four meta-paths are introduced: microbe-disease-microbe, microbe-drug-microbe, disease-microbe-disease, and disease-drug-disease; for the meta-path instance of each meta-path, the encoding of the meta-path instance is obtained by averaging the vectors of the nodes on the meta-path instance;

[0021] S32. Encode each meta-path instance to learn the structure and semantic information of the target node; calculate the attention coefficient of each meta-path instance, that is, calculate the similarity between the meta-path encoding vectors, normalize it through the softmax function to obtain the attention score of each meta-path instance, and weightedly aggregate all meta-path instances under each meta-path according to the attention score.

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

[0023] Metapathways include microbe-drug-microbe, microbe-disease-microbe, disease-drug-disease, and disease-microbe-disease;

[0024] Through weighted aggregation of the local attention mechanism based on CNN, for the input query, key, and value vectors, a convolutional neural network is used to extract local features, and then the similarity between the query vector and the key vector is calculated. After obtaining the attention coefficient, the value vector is weightedly aggregated to obtain the local feature representation of the node;

[0025] The final representation of microbial nodes and disease nodes is obtained through aggregation through a global attention mechanism based on a fully connected layer.

[0026] Further, S5 includes the following steps:

[0027] The features of the final microbial node and disease node are represented, and the association score of the microbial and disease pair is output through the random forest classifier. The loss between the true value and the predicted value is calculated through the binary cross entropy loss function to further optimize the parameters within the model. Finally, the microbial and disease pair with an association score greater than 0.5 is considered to be associated, otherwise it is considered to be unassociated.

[0028] The present invention also provides a storage medium, which includes a stored program, wherein when the program is run, any of the above-mentioned meta-path-based microorganism-disease association prediction methods is executed.

[0029] 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 one of the above-mentioned meta-path-based microorganism-disease association prediction methods through the computer program.

[0030] Compared with the prior art, the present invention has the following advantages:

[0031] The technical solution provided by the present invention first constructs multiple similarity networks for microorganisms and diseases, performs multi-order similarity network fusion, then introduces meta-paths to capture the structural information and semantic information of nodes, and aggregates the features of different meta-path instances under different meta-paths through the attention mechanism, and obtains the final feature representation through the random forest classifier to predict the association score. The present invention not only uses MOSFL to fuse feature representations from different similarity networks, but also introduces the structural and semantic information of nodes through meta-paths, and adaptively aggregates this information through local attention and global attention, which is very meaningful for exploring potential associations between microorganisms and diseases, and has certain value for actual disease diagnosis and treatment and subsequent drug research and development. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] 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.

[0033] Figure 1 The figure is a flow chart of the overall method of the present invention.

[0034] Figure 2 A flow chart for constructing a heterogeneous network according to the present invention.

[0035] Figure 3 Encode and weight the aggregate graph for the meta-path instances of the present invention.

[0036] Figure 4 This is the attention aggregation and association prediction graph between meta-paths of 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 method for predicting microorganism-disease association based on metapathway, comprising the following steps:

[0040] S1. Calculate the semantic similarity of diseases, Gaussian interaction spectrum kernel similarity, cosine similarity, and sigmoid kernel function similarity. Calculate the functional similarity of microorganisms, Gaussian interaction spectrum kernel similarity, cosine similarity, and sigmoid kernel function similarity. Calculate the functional similarity of drugs, Gaussian interaction spectrum kernel similarity, cosine similarity, and sigmoid kernel function similarity.

[0041] S2, such as Figure 2 As shown, multiple similarity views are used to construct isomorphic networks of microorganisms, diseases, and drugs, enrich node similarity information, and construct a heterogeneous graph containing microorganism nodes, disease nodes, and drug nodes.

[0042] S21. For microorganisms, diseases and drugs, different similarity data are calculated from different perspectives based on different biological data. The construction of isomorphic networks, with microorganisms, diseases and drugs as nodes, and different similarity views as the adjacency matrices of their respective isomorphic networks, aggregates richer similarity information, which is very important for enhancing the expression of microorganism and disease node features. Based on this, a functional similarity network, a Gaussian kernel similarity network, a cosine similarity network and a sigmoid kernel function similarity network containing only microorganism nodes are constructed. At the same time, a semantic similarity network, a Gaussian kernel similarity network, a cosine similarity network and a sigmoid kernel function similarity network containing only disease nodes are constructed, as well as a functional similarity network, a Gaussian kernel similarity network, a cosine similarity network and a sigmoid kernel function similarity network of drug nodes.

[0043] After the construction of multiple similarity networks of S22, microorganisms, diseases and drugs is completed, multi-order similarity fusion learning (MOSFL) is used to combine low-order global learning and high-order feature learning, and multi-order similarity fusion networks are constructed through the decomposable properties of the scaled form ADMM algorithm and optimization techniques such as augmented Lagrangian functions to obtain richer and more reliable structural information and high-order similarity information between nodes. And a heterogeneous network of microorganisms, drugs and diseases is constructed.

[0044] S3, such as Figure 3 As shown, the meta-path instance is encoded, and the feature representations of microbial nodes and disease nodes under the same meta-path are aggregated through intra-meta-path attention.

[0045] Existing prediction methods based on graph neural networks generally obtain node feature representation by aggregating the information of neighboring nodes of the node, which only considers the features of the nodes in the graph and lacks consideration of the features of the edges in the graph. Therefore, when learning node feature representation, the features of the edges in the graph are introduced through the meta-path method, so that the model can learn high-quality node feature representation.

[0046] In the intra-meta-path attention mechanism, attention is aggregated on the features of different meta-path instances under the same meta-path for each microbial node and disease node. First, each meta-path instance is encoded to learn the structure and semantic information of the target node. Then the attention coefficient of each meta-path instance is calculated. In order to make the attention coefficient of each meta-path instance comparable, the attention score of each meta-path instance is obtained by normalization through the softmax function. Finally, all meta-path instances under each meta-path are weighted and aggregated according to the attention score.

[0047] S4, such as Figure 4As shown, based on the attention mechanism between meta-paths, the feature representations of microbial nodes and disease nodes under different meta-paths are aggregated through local attention and global attention to obtain the final representation of microbial nodes and disease nodes.

[0048] For the four meta-paths (microbe-drug-microbe, microbe-disease-microbe, disease-drug-disease, disease-microbe-disease), we first perform weighted aggregation based on the local attention mechanism of CNN. For the input query, key, and value vectors, we use the convolutional neural network to extract local features, then calculate the similarity between the query vector and the key vector, and then obtain the attention coefficient and perform weighted aggregation on the value vector to obtain the local feature representation of the node. Then, we aggregate the final representation of the microbe node and the disease node through the global attention mechanism based on the fully connected layer.

[0049] S5. Predict the probability of association between microorganisms and diseases, and distinguish whether there is an association or no association. The features of the final microorganism node and the disease node are represented, and the association score of the microorganism and disease pair is output through the random forest classifier. The loss between the true value and the predicted value is calculated through the cross entropy loss function, and the parameters within the model are further optimized. The loss between the true value and the predicted value is calculated through the binary cross entropy loss function, and the parameters within the model are further optimized. Finally, the association score between the microorganism and the disease pair is greater than 0.5, which is considered to be associated, otherwise it is considered to be unassociated.

[0050] It also includes the testing steps for the above-built model:

[0051] Step 1: Based on the known associations between microorganisms and diseases, an equal amount of unknown associations are selected through K-means clustering to balance the sample data as the true label of the sample. The data is fused through the similarity features in the model, the features of the meta-path instance are aggregated within the meta-path, and the features under different meta-paths are aggregated between the meta-paths, and the random forest classifier is used to obtain the final association probability prediction value. The loss value between the true sample label value and the model prediction value is recorded, as well as the indicator value for evaluating the classification ability of the model.

[0052] Step 2: Test various parameters and important modules that affect model performance. Observe the impact of the number of attention heads in the attention mechanism and the characteristic dimensions of network nodes on the model test results. Develop model variants such as removing similarity network fusion modules or removing global attention modules between meta-paths, or removing local attention modules, and test model results containing different modules.

[0053] Step 3: Test the model’s ability to discover potential associations between microorganisms and diseases. For several diseases with high incidence rates in real society, predict and observe the potential microorganisms associated with them.

[0054] 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 microorganism-disease association based on metapathway, characterized in that: The steps include: S1. Calculate the similarity of diseases, microorganisms and drugs respectively; S2. Based on the calculation of the similarities of diseases, microorganisms and drugs, multiple similarity views are used to construct isomorphic networks of microorganisms, diseases and drugs to enrich the node similarity information, and a heterogeneous graph containing microorganism nodes, disease nodes and drug nodes is constructed based on the isomorphic networks of microorganisms, diseases and drugs; S3, encoding the meta-path instances corresponding to the four meta-paths of microbe-disease-microbe, microbe-drug-microbe, disease-microbe-disease, and disease-drug-disease in the heterogeneous graph, and aggregating the feature representations of microbe nodes and disease nodes under the same meta-path through attention within the meta-path; S4. Based on the attention mechanism between meta-paths, the feature representations of microbial nodes and disease nodes under different meta-paths are aggregated through local attention and global attention to obtain the final representations of microbial nodes and disease nodes. S5. Predict the probability of association between microorganisms and diseases, and distinguish whether there is an association or no association.

2. The method for predicting microorganism-disease association based on metapath according to claim 1, characterized in that: In S1: Disease similarity includes semantic similarity of diseases, Gaussian interaction spectrum kernel similarity, cosine similarity, and sigmoid kernel function similarity; Microbial similarity includes microbial functional similarity, Gaussian interaction spectrum kernel similarity, cosine similarity, and sigmoid kernel function similarity; Drug similarity includes drug functional similarity, Gaussian interaction spectrum kernel similarity, cosine similarity, and sigmoid kernel function similarity.

3. The method for predicting microorganism-disease association based on metapath according to claim 1, characterized in that: S2 specifically includes the following steps: S21. Taking microorganisms, diseases and drugs as nodes and different similarity views as the adjacency matrix of the isomorphic networks of microorganisms, diseases and drugs respectively, construct a functional similarity network, a Gaussian kernel similarity network, a cosine similarity network and a sigmoid kernel function similarity network containing only microorganism nodes, and construct a semantic similarity network, a Gaussian kernel similarity network, a cosine similarity network and a sigmoid kernel function similarity network containing only disease nodes, as well as a functional similarity network, a Gaussian kernel similarity network, a cosine similarity network and a sigmoid kernel function similarity network of drug nodes; S22. Utilize multi-order similarity fusion learning, combine low-order global learning and high-order feature learning, and construct a multi-order similarity fusion network through optimization technology to obtain richer and more reliable structural information and high-order similarity information between nodes, and construct a microorganism-drug-disease heterogeneous network.

4. The method for predicting microorganism-disease association based on metapath according to claim 1, characterized in that: S3 specifically includes the following steps: S31. When learning node feature representation, the features of the edges in the graph are introduced by means of meta-paths, so that the model can learn high-quality node feature representation; for heterogeneous graphs, the following four meta-paths are introduced: microbe-disease-microbe, microbe-drug-microbe, disease-microbe-disease, and disease-drug-disease; for the meta-path instance of each meta-path, the encoding of the meta-path instance is obtained by averaging the vectors of the nodes on the meta-path instance; S32. Encode each meta-path instance to learn the structure and semantic information of the target node; calculate the attention coefficient of each meta-path instance, that is, calculate the similarity between the meta-path encoding vectors, normalize it through the softmax function to obtain the attention score of each meta-path instance, and weightedly aggregate all meta-path instances under each meta-path according to the attention score.

5. The method for predicting microorganism-disease association based on metapath according to claim 1, characterized in that: S4 specifically includes the following steps: Metapathways include microbe-drug-microbe, microbe-disease-microbe, disease-drug-disease, and disease-microbe-disease; Through weighted aggregation of the local attention mechanism based on CNN, for the input query, key, and value vectors, a convolutional neural network is used to extract local features, and then the similarity between the query vector and the key vector is calculated. After obtaining the attention coefficient, the value vector is weightedly aggregated to obtain the local feature representation of the node; The final representation of microbial nodes and disease nodes is obtained through aggregation through a global attention mechanism based on a fully connected layer.

6. The method for predicting microorganism-disease association based on metapath according to claim 1, characterized in that: S5 includes the following steps: The features of the final microbial node and disease node are represented, and the association score of the microbial and disease pair is output through the random forest classifier. The loss between the true value and the predicted value is calculated through the binary cross entropy loss function to further optimize the parameters within the model. Finally, the microbial and disease pair with an association score greater than 0.5 is considered to be associated, otherwise it is considered to be unassociated.

7. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is run, the meta-path-based microorganism-disease association prediction method described in 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 meta-pathway-based microorganism-disease association prediction method according to any one of claims 1 to 6 by running the computer program.

Images