A drug repositioning method and system based on biological knowledge and network topology

Abstract

The present invention proposes a drug repositioning method and system based on biological knowledge and network topology, comprising a network construction module, a feature learning module, a model training module, a drug repositioning module, and a result presentation module. The network construction module constructs biological network data into a heterogeneous information network. The feature learning module executes server computing instructions to obtain biological feature matrices and network feature matrices for drugs and diseases. The model training module obtains input parameters and trains a drug discovery model on the server. The drug repositioning module executes drug repositioning instructions after obtaining the drug repositioning model. Finally, the drug repositioning results are output and presented through the presentation module. The present invention directly acts on biological network data sets containing biological knowledge, enabling drug repositioning in heterogeneous biological networks with high accuracy and the ability to effectively discover new uses for known drugs.

Description

Technical Field

[0001] The present invention relates to the technical field of computer data processing, and in particular to a drug repositioning method and system based on biological knowledge and network topology. Background Art

[0002] Currently, existing drug repositioning methods primarily rely on bipartite graph networks to discover drug candidates, considering only the connections between drugs and diseases within biological networks. However, this approach, based solely on bipartite graph networks, not only overlooks the importance of protein molecules within biological networks to the drug-disease network but also misses the biological properties of biomolecules. This can lead to the discovery of some meaningful drug-disease associations. For example, valproic acid, which alters the life cycle of breast cancer cells by affecting histone expression, would be difficult to uncover if based solely on bipartite graph networks.

[0003] Although some drug repositioning methods that consider biological attribute information or heterogeneous networks have been proposed, the accuracy of their predicted drug-disease associations is poor. Summary of the Invention

[0004] The present invention aims to provide a drug repositioning method and system based on biological knowledge and network topology. The method comprises a network construction module, a feature learning module, a model training module, a drug repositioning module, and a results presentation module. Drug screening is achieved by comprehensively considering the biological properties of molecules in biological networks and network topology information. The present invention directly operates on biological network data sets that incorporate biological knowledge, enabling drug repositioning in heterogeneous biological networks with high accuracy and the ability to effectively discover new uses for known drugs.

[0005] The drug repositioning method based on biological knowledge and network topology described in the present invention is carried out according to the following steps:

[0006] a. Collect data on the association between known drugs and diseases, the association between drugs and proteins, the association between proteins and diseases, drug molecular structure information, and disease MeSH descriptor information;

[0007] b. constructing a heterogeneous information network based on the drug-disease association data, drug-protein association data, and protein-disease association data in step a;

[0008] c. Applying chemical informatics to calculate the drug molecular structure information in step a using the RDKit open source toolkit to obtain the drug molecular fingerprint matrix;

[0009] d. Apply semantic similarity theory to calculate the disease MeSH descriptor information in step a using the Jaccard formula to obtain the disease semantic similarity matrix;

[0010] e. Based on the drug molecular fingerprint matrix in step c and the disease semantic similarity matrix in step d, an autoencoder is applied for training to obtain a biometric matrix of drugs and diseases;

[0011] f. Based on the heterogeneous information network in step b, the DeepWalk algorithm is used to calculate the network characteristics of drugs and diseases using the graph embedding model to obtain the network characteristic matrix of drugs and diseases;

[0012] g. Based on the biological feature matrix of drugs and diseases in step e and the network feature matrix of drugs and diseases in step f, a random forest classifier is used to train a drug repositioning model;

[0013] h. Obtain the desired candidate drugs based on the drug repositioning model obtained in step g.

[0014] A drug repositioning system based on biological knowledge and network topology, the drug repositioning system comprising a network construction module, a feature learning module, a model training module, a drug discovery module, and a result display module, wherein the network construction module is connected to the feature learning module, the feature learning module is connected to the model training module, and the model training module is connected to the result display module;

[0015] The network construction module constructs the biological association network of drugs, proteins, and diseases into a heterogeneous information network, and passes the constructed heterogeneous information network to the feature learning module;

[0016] The feature learning module extracts the biological characteristics and network characteristics of drugs and diseases in the heterogeneous information network, obtains the biological characteristic matrix and network characteristic matrix of drugs and diseases, and passes these matrices to the model training module;

[0017] The model training module trains a model about drugs and diseases based on the input model parameters and the results of the feature learning module, and passes the model to the result display module;

[0018] The network building module introduces proteins as intermediaries to construct a heterogeneous information network including drugs, proteins and diseases to enhance the connectivity between drugs and diseases.

[0019] The feature learning module considers the characteristics of drugs and diseases more comprehensively, and trains the drug repositioning model by calculating the biological feature matrix and network feature matrix of drugs and diseases, so as to more accurately determine the candidate drugs for the disease.

[0020] The present invention provides a drug repositioning method based on biological knowledge and network topology. When collecting data, the method obtains drug-disease association data based on the CTD database, drug-protein association data based on the DrugBank v5.0 database, protein-disease association data based on the DisGeNET database, drug molecular structure information based on the DrugBank v5.0 database, and disease MeSH descriptor information based on the Medical Subject Headings (MeSH) synonym dictionary. Specifically:

[0021] Drug molecular structure information and disease MeSH descriptor information belong to biological knowledge;

[0022] The association data between drugs and diseases, the association data between drugs and proteins, and the association data between proteins and diseases belong to the network topology structure.

[0023] The present invention provides a drug repositioning system based on biological knowledge and network topology, the system comprising: a network construction module, which constructs the biological association network of drugs, proteins, and diseases into a heterogeneous information network, wherein;

[0024] The feature learning module inputs learning parameters and performs feature learning based on the heterogeneous information network obtained in the network construction module to obtain the biological feature matrix and network feature matrix of drugs and diseases respectively;

[0025] The model training module inputs model parameters and trains a drug discovery model based on the biological feature matrix and network feature matrix of drugs and diseases obtained by the feature learning module;

[0026] The drug repositioning module performs drug screening on the drug repositioning model trained by the model training module;

[0027] The result display module outputs the candidate drugs discovered by the drug repositioning module.

[0028] Compared with the prior art, the present invention has the following beneficial technical effects:

[0029] The present invention describes a drug repositioning method based on biological knowledge and network topology. During the algorithm design process, heterogeneous information network topology and biological property information of biomolecules are utilized to complete the drug repositioning task. By more comprehensively learning the characteristics of biomolecules, the accuracy of drug repositioning is improved, thereby solving the defects of the existing technology in biological network drug repositioning.

[0030] The drug repositioning system based on biological knowledge and network topology described in the present invention is mainly composed of a network construction module, a feature learning module, a model training module, a drug repositioning module and a result display module. First, the network construction module constructs the three biological association relationships into a heterogeneous information network. Secondly, the feature learning module performs feature learning on the obtained heterogeneous information network to obtain the biological feature matrix and network feature matrix of drugs and diseases. Then, the model training module trains the obtained biological feature matrix and network feature matrix of drugs and diseases to obtain a drug repositioning model. Subsequently, the drug repositioning module predicts the association relationship between a given disease and a candidate drug to achieve the purpose of drug repositioning. Finally, the result display module displays the results of drug repositioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a logical structure diagram of the present invention. DETAILED DESCRIPTION

[0032] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0033] Example

[0034] The drug repositioning method based on biological knowledge and network topology described in the present invention is carried out according to the following steps:

[0035] a. Collect data on the association between known drugs and diseases, the association between drugs and proteins, the association between proteins and diseases, drug molecular structure information, and disease MeSH descriptor information;

[0036] b. constructing a heterogeneous information network based on the drug-disease association data, drug-protein association data, and protein-disease association data in step a;

[0037] c. Applying chemical informatics to calculate the drug molecular structure information in step a using the RDKit open source toolkit to obtain the drug molecular fingerprint matrix;

[0038] d. Apply semantic similarity theory to calculate the disease MeSH descriptor information in step a using the Jaccard formula to obtain the disease semantic similarity matrix;

[0039] e. Based on the drug molecular fingerprint matrix in step c and the disease semantic similarity matrix in step d, an autoencoder is applied for training to obtain a biometric matrix of drugs and diseases;

[0040] f. Based on the heterogeneous information network in step b, the DeepWalk algorithm is used to calculate the network characteristics of drugs and diseases using the graph embedding model to obtain the network characteristic matrix of drugs and diseases;

[0041] g. Based on the biological feature matrix of drugs and diseases in step e and the network feature matrix of drugs and diseases in step f, a random forest classifier is used to train a drug repositioning model;

[0042] h. Obtaining the desired drug candidate based on the drug repositioning model obtained in step g;

[0043] A drug repositioning system based on biological knowledge and network topology, the drug repositioning system comprising a network construction module, a feature learning module, a model training module, a drug discovery module, and a result display module, wherein the network construction module is connected to the feature learning module, the feature learning module is connected to the model training module, and the model training module is connected to the result display module;

[0044] The network construction module constructs the biological association network of drugs, proteins, and diseases into a heterogeneous information network, and passes the constructed heterogeneous information network to the feature learning module;

[0045] The feature learning module extracts the biological characteristics and network characteristics of drugs and diseases in the heterogeneous information network, obtains the biological characteristic matrix and network characteristic matrix of drugs and diseases, and passes these matrices to the model training module;

[0046] The model training module trains a model about drugs and diseases based on the input model parameters and the results of the feature learning module, and passes the model to the result display module;

[0047] The network building module introduces proteins as intermediaries to construct a heterogeneous information network including drugs, proteins and diseases to enhance the connectivity between drugs and diseases.

[0048] The feature learning module considers the characteristics of drugs and diseases more comprehensively, and trains the drug repositioning model by calculating the biological feature matrix and network feature matrix of drugs and diseases, which can more accurately determine candidate drugs for diseases;

[0049] like Figure 1 As shown:

[0050] Network building blocks:

[0051] The first step is to build a heterogeneous information network: define the heterogeneous information network as HIN = {V, A, E}, and abstract the drugs, proteins and disease molecules in the biological network into nodes in the network respectively. and Abstract the biological knowledge of drugs and disease molecules into node attributes A in the network DR and A DI , abstracting the associations among drugs, proteins and disease molecules into edges E in the network DD , E DP and E PD ;

[0052] The second step is to store heterogeneous network information. Arrange the set V composed of all nodes in the heterogeneous information network = {V DR , V DI , V PR}、The set of node attributes A={A DR , A DI} and the set of all edges E={E DD , E DP , E PD}, and store;

[0053] Feature learning module:

[0054] The first step is to obtain the drug molecular fingerprint matrix Calculate the drug molecule fingerprint matrix based on the drug node attributes obtained in the network construction module;

[0055] The specific method for calculating the drug molecule fingerprint matrix is ​​as follows:

[0056] 2-1 Use the RDKit software package to calculate the drug molecular structure information, obtain the drug molecular fingerprint matrix, and write Python code:

[0057] from rdkit.Chem import AllChem

[0058] R DR =AllChem.GetMorganFingerprintAsBitVect(A DR ,2,nBits=K);

[0059] The second step is to obtain the disease semantic similarity matrix Calculate the disease semantic similarity matrix based on the disease node attributes obtained in the previous module.

[0060] The specific method for calculating the disease semantic similarity matrix is ​​as follows:

[0061] 2-2 Definition of Directed Acyclic Graph The Jaccard formula is used to calculate the semantic similarity of diseases. Indicates disease All ancestor nodes of Indicates disease The set of all edges of . Disease For disease The contribution of

[0062]

[0063] 2-3 By calculation The contribution of all ancestor nodes in the disease The semantic value of is:

[0064]

[0065] 2-4 Combining formulas (1-1) and (1-2), and The semantic similarity between them is:

[0066]

[0067] 2-5 According to formula (1-3), the disease semantic similarity matrix S is obtained DI :

[0068]

[0069] In the above formula,

[0070]

[0071] The third step is to obtain the bio-feature matrix of drugs and diseases. According to the above, the drug molecular fingerprint matrix R DR and disease semantic similarity matrix S DI , apply the autoencoder for training to obtain the drug and disease biological feature matrix H;

[0072] The fourth step is to obtain the network feature matrix of drugs and diseases.

[0073] The specific method for calculating the network feature matrix of drugs and diseases is as follows:

[0074] 2-7 Calculate the network characteristics of drugs and diseases through the graph embedding model DeepWalk algorithm:

[0075]

[0076] In the above formula, Φ(v i ) represents node v i (1≤i≤|V|) network characteristics, w represents the neighborhood range of the node.

[0077] 2-8 According to 2-7, the network characteristics of each node are obtained, and the network characteristic matrix of drugs and diseases is constructed. Q = Φ ({VDR , V DI})

[0078] Model training module:

[0079] The first step is to input the model parameter t;

[0080] The second step is to construct the training input feature matrix X = [H(V)Q(V)] of the random forest classifier based on the drug and disease biological feature matrix and network feature matrix obtained in the previous module;

[0081] In the third step, a random forest classifier is applied to train the input feature matrix X to obtain a drug discovery model;

[0082] Drug Discovery Module:

[0083] Use the drug discovery module to predict the association between each disease and candidate drugs. Specifically, the association between each disease and candidate drug is constructed as a test input feature matrix Y. The drug discovery model obtained above is used to test the input feature matrix Y to obtain the predicted score of the candidate drug for the disease for use in the result display module.

[0084] Results display module:

[0085] Based on the results obtained by the drug discovery module, this module treats the association between each disease and candidate drug as a row, with the disease, drug, and prediction score as elements in the row, and processes all prediction results into a text file for output display.

[0086] The above shows and describes the basic principles and main features of the present invention, as well as the characteristics of the present invention. The present invention is not limited to the above experimental examples. The above experimental examples and descriptions are merely illustrative of the principles of the present invention. Various changes and improvements are possible without departing from the principles of the present invention, and such changes and improvements fall within the scope of the invention as claimed.

Claims

1. A drug repositioning method based on biological knowledge and network topology, characterized in that: Follow these steps: a. Collect data on the association between known drugs and diseases, the association between drugs and proteins, the association between proteins and diseases, drug molecular structure information, and disease MeSH descriptor information; b. constructing a heterogeneous information network based on the drug-disease association data, drug-protein association data, and protein-disease association data in step a; c. Applying chemical informatics to calculate the drug molecular structure information in step a using the RDKit open source toolkit to obtain the drug molecular fingerprint matrix; d. Apply semantic similarity theory to calculate the disease MeSH descriptor information in step a using the Jaccard formula to obtain the disease semantic similarity matrix; e. Based on the drug molecular fingerprint matrix in step c and the disease semantic similarity matrix in step d, an autoencoder is applied for training to obtain a biometric matrix of drugs and diseases; f. Based on the heterogeneous information network in step b, the DeepWalk algorithm is used to calculate the network characteristics of drugs and diseases using the graph embedding model to obtain the network characteristic matrix of drugs and diseases; g. Based on the biological feature matrix of drugs and diseases in step e and the network feature matrix of drugs and diseases in step f, a random forest classifier is used to train a drug repositioning model; h. Obtaining the desired drug candidate based on the drug repositioning model obtained in step g; The heterogeneous information network constructed in step b is HIN = {V, A, E}, where V is the set of all nodes, including drugs, proteins, and disease molecules, A is the set of node attributes, including biological knowledge of drugs and disease molecules, and E is the set of all edges, including the associations between drugs, proteins, and disease molecules. The disease semantic similarity matrix S obtained in step d DI for: in, for and The semantic similarity between them satisfies the following formula: in, For disease The semantic value of For disease The semantic value of each disease is calculated by calculating all ancestor nodes of the corresponding disease and The contribution is obtained; For disease For disease Contribution, For disease For disease contribution.

2. A drug repositioning system based on biological knowledge and network topology, implemented based on the drug repositioning method based on biological knowledge and network topology according to claim 1, characterized in that: The drug repositioning system is composed of a network construction module, a feature learning module, a model training module, a drug discovery module and a result display module. The network construction module is connected to the feature learning module, the feature learning module is connected to the model training module, and the model training module is connected to the result display module, wherein; The network construction module constructs the biological association network of drugs, proteins, and diseases into a heterogeneous information network, and passes the constructed heterogeneous information network to the feature learning module; The feature learning module extracts the biological characteristics and network characteristics of drugs and diseases in the heterogeneous information network, obtains the biological characteristic matrix and network characteristic matrix of drugs and diseases, and passes these matrices to the model training module; The model training module trains a model about drugs and diseases based on the input model parameters and the results of the feature learning module, and passes the model to the result presentation module.

3. The drug repositioning system based on biological knowledge and network topology according to claim 2, characterized in that: The network building module introduces proteins as intermediaries to construct a heterogeneous information network including drugs, proteins and diseases to enhance the connectivity between drugs and diseases.

4. The drug repositioning method system based on biological knowledge and network topology according to claim 3 is characterized in that: The feature learning module considers the characteristics of drugs and diseases more comprehensively, and trains the drug repositioning model by calculating the biological feature matrix and network feature matrix of drugs and diseases, so as to more accurately determine the candidate drugs for the disease.

Images