Method and system for predicting incidence relation between disease and medicine based on biological regulation mechanism graph representation learning
Through the method of graph representation learning based on the biological regulation mechanism, a heterogeneous biological network is constructed and graph representation features are extracted, which solves the problem of poor accuracy in predicting drug and disease association relationships in the prior art, and achieves a more efficient drug relocation effect .
Patent Information
- Application Number
- CN202411011917.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-19
AI Technical Summary
Existing drug relocation calculation methods ignore the connection patterns within the biological network, resulting in poor accuracy in predicting drug and disease association relationships, especially in the study of regulatory mechanisms in heterogeneous biological networks.
Using a graph representation learning method based on biological regulatory mechanism, a graph representation feature of drugs, diseases and proteins is extracted by constructing heterogeneous biological networks, using metapathic pathway strategies and graph information dissemination mechanisms, and an association prediction model is trained using XGBoost classifier.
The accuracy of drug relocation is improved, and the ability to predict the association relationship between drugs and diseases is enhanced by simulating the drug regulation mechanism in organisms.
Smart Images

Figure CN118983050A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of drug repositioning, and in particular relates to a method and system for predicting the association between diseases and drugs based on biological regulation mechanism graph representation learning. Background Art
[0002] Existing drug repositioning computational methods usually integrate a variety of molecular biological knowledge such as SMILES descriptor information of drugs, MeSH semantic information of diseases, and sequence information of proteins to enhance the prediction accuracy of the model. However, these methods often ignore the importance of mining the rich connection patterns within biological networks, and this in-depth understanding of network structure information is crucial for revealing the mechanism of interaction between biological molecules. In particular, the regulatory mechanism in biological networks, such as the mechanism by which drug molecules combine with the functional groups of biological macromolecules (such as receptor proteins, enzymes, etc.) through their unique chemical structures to achieve therapeutic effects, is particularly critical in the study of heterogeneous biological networks.
[0003] Although some heterogeneous network-based drug repositioning methods have been implemented, the accuracy of their predicted drug-disease associations is poor. Summary of the invention
[0004] The purpose of the present invention is to overcome the defects of the prior art and propose a method for predicting the association between diseases and drugs based on biological regulation mechanism graph representation learning. The present invention also discloses a system for predicting the association between diseases and drugs based on biological regulation mechanism graph representation learning.
[0005] In order to achieve the above object, the present invention proposes a method for predicting the association between diseases and drugs based on biological regulation mechanism graph representation learning, comprising:
[0006] Input the disease name into the trained association prediction model to obtain candidate drugs and prediction scores for the disease; the association prediction model is a graph representation learning model;
[0007] The training process of the association relationship prediction model includes:
[0008] Establishing a training set including a correlation network between known drugs and diseases, a correlation network between known drugs and proteins, and a correlation network between proteins and diseases;
[0009] Construct heterogeneous biological networks based on training sets;
[0010] Using metapathway strategies to construct regulatory maps of drugs, diseases, and proteins;
[0011] According to the regulatory graphs of drugs, proteins and diseases, the graph information propagation mechanism is used to obtain the graph representation features of different regulatory graphs, and then the graph representation feature matrix of drugs and diseases is obtained;
[0012] Based on the graph representation feature matrix of drugs and diseases, the XGBoost classifier is used to obtain the trained association prediction model.
[0013] Preferably, the heterogeneous biological network G is: G = {V, E}; wherein,
[0014] V represents the set of all nodes of the heterogeneous biological network, including drug nodes, disease nodes, and protein molecule nodes;
[0015] E represents the set of all edges of the heterogeneous biological network, including the edges corresponding to the association between drugs and diseases, the edges corresponding to the association between drugs and protein molecules, and the edges corresponding to the association between protein molecules and diseases.
[0016] Preferably, the method of using a metapath strategy to construct a regulatory map of drugs, diseases and proteins comprises:
[0017] Step A1: define a meta-path set MP = {MP m} corresponds to m types of regulatory mechanisms;
[0018] Step A2: Define the meta-path based MP m Constructed regulatory map It consists of a set of specific instances P, P = {P1, P2, ..., P i …,P s}, m∈MP, P represents one of the predefined meta-paths in the meta-path set MP, s represents the total number of paths generated according to P, and path P i It consists of L sequential nodes, denoted as v1→…→v i →v i+1 →…v L ,v∈V, where v i to v i+1 The transition probability is:
[0019]
[0020] Among them, t represents the node type, which belongs to the defined node type set {drug, disease, protein molecule}, and the function Φ(t-1) is used to specify if the current node v i The type of the next node is t, then i+1 The type that should be, the set N(v i ,Φ(t-1)) contains all nodes related to node v i Adjacent nodes of type t-1 that meet the definition of function Φ.
[0021] Preferably, the regulatory graphs of drugs, proteins and diseases are used to obtain the graph representation features of different regulatory graphs using a graph information propagation mechanism;
[0022] Step B1: Define the control map The adjacency matrix of For the control diagram The graph representation features of the nodes in are obtained through the graph information propagation mechanism:
[0023] X l =σ(A m X l-1 )
[0024] Among them, X l Represents the representation of the node at the lth layer. When the number of neural network propagation layers is l = 0, X is used as the initial feature matrix, and one-hot encoding is used to initialize the features for each node;
[0025] Step B2: Get the weighted average of features at different levels:
[0026]
[0027] in, is a list containing the minimum number of propagation layers l for all nodes v∈V, ε is the feature threshold, α l is the weight coefficient of the contribution of the lth layer, A l Indicates that the node features are in the graph The state after l steps of propagation;
[0028] Step B3: Define a tensor T to integrate the graph representation features from different control graphs:
[0029]
[0030] Among them, X m Representation Regulation Diagram The characteristics of the nodes in , M is the total number of control graph types initially set; V represents the total number of nodes, and d represents the dimension.
[0031] Preferably, obtaining the graphical representation feature matrix of drugs and diseases comprises:
[0032] Step C1: Based on the tensor T, a nonlinear transformation function is used to aggregate the graph representation features f of each node in the control graph m :
[0033]
[0034] in, is the feature vector of node v in the m-th view control graph, is a trainable weight matrix, b m is the bias term;
[0035] Step C2: Apply the softmax function to normalize the weight coefficients between different control graphs:
[0036]
[0037] Among them, β m For control diagram The relative importance of m is the attention score, W att is a trainable weight matrix, the superscript T indicates transpose;
[0038] Step C3: According to the control graph weight β m Calculate the graph representation features of all regulatory graphs and perform weighted summation to obtain the feature matrix H of drugs and diseases:
[0039]
[0040] Preferably, the graph representation feature matrix based on drugs and diseases uses an XGBoost classifier to obtain a trained association relationship prediction model, including:
[0041] Step D1: Define a drug-disease pair dataset S, represented by S = {(s i ,y i )}, where 1≤i≤|S|, s i is the feature vector of the ith drug and disease pair, y i ∈{0,1} is the corresponding label, indicating whether there is an association between the drug and the disease, 0 means no association, 1 means association, |S| is the total number of drug-disease pairs in the dataset, S is input into the XGBoost classifier for training, and the objective function Obj is:
[0042]
[0043] Among them, the loss function Used to measure the actual label y of the i-th drug-disease pair i and predicted labels The deviation between them, K represents the total number of decision trees; Ω(f k ) is a regularization term used to constrain overly complex model structures. The specific calculation process is as follows:
[0044]
[0045] Where Le is the number of decision tree leaf nodes, γ and λ represent the complexity of the decision tree and the parameters of the weight constraint, respectively, ||ω|| is the 2-norm of the decision tree; f k (s i ) represents the feature vector of the i-th drug-disease pair, and the prediction result is calculated by the k-th decision tree.
[0046] On the other hand, the present invention provides a disease-drug association relationship prediction system based on biological regulation mechanism graph representation learning, comprising:
[0047] The prediction output module inputs the disease name into the trained association prediction model to obtain the candidate drugs and prediction scores for the disease; the association prediction model is a graph representation learning model;
[0048] The model training module is used to construct a heterogeneous biological network based on a training set including an association network of known drugs and diseases, an association network of known drugs and proteins, and an association network of proteins and diseases; extract and learn the regulatory graphs of drugs, diseases, and proteins in the heterogeneous biological network to obtain a feature matrix of drugs and diseases, and train to obtain an association relationship prediction model.
[0049] Preferably, the model training module includes:
[0050] A network construction submodule, for constructing a heterogeneous biological network based on a training set including an association network of known drugs and diseases, an association network of known drugs and proteins, and an association network of proteins and diseases;
[0051] A feature learning submodule is used to construct the regulation graphs of drugs, diseases and proteins using a meta-path strategy, and to obtain the graph representation features of different regulation graphs using a graph information propagation mechanism according to the regulation graphs of drugs, proteins and diseases, and then obtain the graph representation feature matrix of drugs and diseases, and pass the feature matrix to the model training submodule; and
[0052] The model training submodule is used to train the association relationship prediction model that meets the training requirements based on the results of the feature learning submodule.
[0053] In a third aspect, the present invention discloses a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.
[0054] In a fourth aspect, the present invention discloses a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.
[0055] Compared with the prior art, the advantages of the present invention are:
[0056] The present invention provides a method and system for predicting the relationship between diseases and drugs based on biological regulation mechanism graph representation learning. During the method design process, the drug regulation mechanism in heterogeneous biological networks is used to extract the characteristics of drugs and diseases, thereby completing the drug repositioning task. The accuracy of drug repositioning is improved by simulating the drug regulation mechanism in vivo, thereby solving the defects of the prior art in biological network drug repositioning. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] Figure 1 This is the architecture diagram of the disease-drug association prediction system based on biological regulation mechanism graph representation learning. DETAILED DESCRIPTION
[0058] The present invention provides a method for predicting the association between diseases and drugs based on biological regulation mechanism graph representation learning, comprising:
[0059] Input the disease name into the trained association prediction model to obtain candidate drugs and prediction scores for the disease; the association prediction model is a graph representation learning model;
[0060] The training process of the association relationship prediction model includes:
[0061] Establishing a training set including a correlation network between known drugs and diseases, a correlation network between known drugs and proteins, and a correlation network between proteins and diseases;
[0062] Construct heterogeneous biological networks based on training sets;
[0063] Using metapathway strategies to construct regulatory maps of drugs, diseases, and proteins;
[0064] According to the regulatory graphs of drugs, proteins and diseases, the graph information propagation mechanism is used to obtain the graph representation features of different regulatory graphs, and then the graph representation feature matrix of drugs and diseases is obtained;
[0065] Based on the graph representation feature matrix of drugs and diseases, the XGBoost classifier is used to obtain the trained association prediction model.
[0066] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.
[0067] Example 1
[0068] Embodiment 1 of the present invention provides a method for predicting the relationship between diseases and drugs based on biological regulation mechanism graph representation learning, which is performed in the following steps:
[0069] Step 1: Collect the association networks of known drugs and diseases, the association networks of drugs and proteins, and the association networks of proteins and diseases;
[0070] Step 2: construct a heterogeneous biological network based on the association network of drugs and diseases, the association network of drugs and proteins, the association network of proteins and diseases in step 1, and the biological knowledge of drugs, proteins, and diseases;
[0071] Define a heterogeneous biological network as G = {V, E};
[0072] Where V represents the set of all nodes in the heterogeneous biological network, V = {V DR ,V DI ,V PR}; V DR Represents drugs as nodes in the network. represents the i-th drug, N represents the number of drugs; V DI Representing diseases as nodes in a network represents the jth disease, M represents the number of diseases; V PR Representing protein molecules as nodes in the network represents the lth protein molecule, and P represents the number of protein molecules;
[0073] E represents the set of all edges in the heterogeneous biological network, E = {E DD ,E DP ,E PD}; E DD Indicates that the relationship between drugs and diseases is abstracted as edges in the network; E DP Indicates that the relationship between drugs and protein molecules is abstracted as edges in the network, E PD The relationship between protein molecules and diseases is abstracted as edges in the network.
[0074] Step 3: Based on the heterogeneous biological network in step 2, the meta-path strategy is applied to construct the regulatory graph of drugs, diseases and proteins;
[0075] The specific method for calculating the regulation graph of drugs, diseases and proteins is as follows:
[0076] Step 3-1: Define a series of meta-path sets MP = {MP m} to illustrate different regulatory mechanisms. For example, the metapaths MP1 = {drug → protein → disease} and MP2 = {drug → protein → drug → disease} show examples of regulation in this set. Among them, drug, protein, disease represent drugs, protein molecules, and diseases, respectively.
[0077] Step 3-2: Define the metapath-based MP m The constructed control map is It consists of a set of specific instances, denoted as P_{m}={P1,P2,…,P i …,P s Here, m∈MP, P_{m} represents one of the predefined meta-paths in the set MP, and s represents the m The total number of paths generated. Path P i It consists of L sequential nodes, denoted as v1→…→v i →v i+1 →…v L (v∈V), where v i to v i+1 The transition probability is:
[0078]
[0079] Where t represents the node type, which belongs to the defined node type set {drug, disease, protein}, and the function Φ(t-1) is used to specify if the current node v i The type of the next node is t, then i+1 The type that should be. The set N(v i ,Φ(t-1))| contains all nodes related to node v i Adjacent nodes of type t-1 that meet the definition of function Φ.
[0080] Step 3-3: Obtain the meta-path set MP from the above formula m , applying the meta-path strategy to obtain the regulatory graphs of drugs, diseases, and proteins
[0081] Defining a Control Map The adjacency matrix of For the control diagram The graph representation features of the nodes in can be obtained through the graph information propagation mechanism:
[0082] X l =σ(A m X l-1 )
[0083] Where X l Represents the representation of the node in the lth layer. When the number of neural network propagation layers is l=0, X is used as the initial feature matrix. Here, one-hot encoding is used to initialize the features for each node.
[0084] The above formula is used to obtain the weighted average of features at different levels:
[0085]
[0086] in A is a list containing the minimum number of propagation layers l for all nodes v∈V, ε is the feature threshold, and α is the weight coefficient for adjusting the contribution of different layers. l Indicates that the node features are in the graph The state after l steps of propagation.
[0087] Define a tensor T to integrate graph representation features from different control graphs:
[0088]
[0089] Where X m Representation Regulation Diagram The characteristics of the nodes in , M is the total number of control graph types initially set.
[0090] Step 3-4: Based on the tensor T, apply a nonlinear transformation function to aggregate the graph representation features of the nodes in each control graph:
[0091]
[0092] in is the feature vector of node v in the m-th view control graph, is a trainable weight matrix, b m is the bias term.
[0093] Apply the softmax function to normalize the weight coefficients between different control graphs:
[0094]
[0095] where β m For control diagram The relative importance of m is the attention score, W att is a trainable weight matrix.
[0096] According to the control graph weight β m Calculate the graph representation features of all regulatory graphs and perform weighted summation to obtain the final feature matrix of drugs and diseases
[0097]
[0098] Step 4: Based on the feature matrix of drugs and diseases in step 3 Apply XGBoost classifier to train the association prediction model:
[0099] Define a drug-disease pair dataset, represented as S = {(s i ,yi )}, where 1≤i≤|S|, s i is from The derived feature vector for the ith drug-disease pair, y i ∈{0,1} is the corresponding label, indicating whether there is an association between the drug and the disease, and |S| is the total number of drug-disease pairs in the dataset. S is input into the XGBoost classifier to complete the training of the following objective function:
[0100]
[0101] The loss function is used to measure the actual label y of the i-th drug-disease pair i and predicted labels The deviation between them, K represents the total number of decision trees. In addition, Ω(f k ) is a regularization term, the purpose of which is to constrain overly complex model structures. The specific calculation process is as follows:
[0102]
[0103] Where Le is the number of decision tree leaf nodes, γ and λ represent the complexity of the decision tree and the parameters of the weight constraint, respectively, ||ω|| is the 2-norm of the decision tree; f k (s i ) represents the feature vector of the i-th drug-disease pair, and the prediction result is calculated by the k-th decision tree.
[0104] Step 4-1: Enter the parameter K of the XGBoost classifier;
[0105] Step 4-2: Based on the feature matrix H of drugs and diseases, apply the XGBoost classifier to train the input feature matrix H to obtain the association prediction model.
[0106] Step 5: Obtain the required candidate drugs and prediction scores based on the association prediction model obtained in step 4;
[0107] The association between each disease and candidate drug is constructed as a test input feature matrix Y, and the drug discovery model obtained above is used to test the input feature matrix Y to obtain the prediction score of the candidate drug for the disease. The higher the prediction score of the candidate drug, the more suitable the candidate drug is for the input disease.
[0108] The association relationship prediction model is a graph neural network model.
[0109] Example 2
[0110] like Figure 1As shown, the present invention also provides a disease-drug association relationship prediction system based on biological regulation mechanism graph representation learning, which is implemented based on the method of Example 1. The system includes a prediction output module and a model training module.
[0111] A prediction output module is used to input the disease name into a trained association prediction model to obtain candidate drugs and prediction scores for the disease; the association prediction model is a graph representation learning model;
[0112] The model training module is used to construct the biological association network and biological knowledge of drugs, proteins, and diseases into a heterogeneous biological network, extract the multi-level subgraphs of drugs and diseases in the heterogeneous biological network, obtain the graph representation feature matrix of drugs and diseases, and train the association relationship prediction model.
[0113] The prediction output module includes: a drug discovery submodule and a result output submodule.
[0114] The drug discovery submodule is used to generate prediction results based on the prediction data input into the association prediction model, and pass these results to the result output submodule.
[0115] The model training module is connected to the drug discovery submodule, and the drug discovery submodule is connected to the result output submodule.
[0116] The model training module includes a network construction submodule, a feature learning submodule and a model training submodule.
[0117] The network construction submodule is used to construct the biological association network of drugs, proteins, and diseases into a heterogeneous biological network, and pass the constructed heterogeneous biological network to the feature learning submodule;
[0118] The network construction submodule introduces proteins as intermediaries to construct a heterogeneous biological network including drugs, proteins and diseases to enhance the connectivity between drugs and diseases.
[0119] The feature learning submodule is used to extract and learn the regulatory graphs of drugs, diseases, and proteins in heterogeneous biological networks, obtain the feature matrices of drugs and diseases, and pass the feature matrices to the model training submodule;
[0120] The feature learning submodule simulates the drug regulation mechanism in the organism more effectively. By extracting and learning the regulatory graphs of drugs, diseases, and proteins to train the association prediction model, it can more accurately identify candidate drugs for a given disease.
[0121] The model training submodule is used to train the association relationship prediction model based on the results of the feature learning submodule.
[0122] Example 3
[0123] The present invention may also provide a computer device, comprising: at least one processor, a memory, at least one network interface and a user interface. The various components in the device are coupled together through a bus system. It is understood that the bus system is used to achieve connection and communication between these components. In addition to the data bus, the bus system also includes a power bus, a control bus and a status signal bus.
[0124] The user interface may include a display, a keyboard or a pointing device, such as a mouse, a trackball, a touch pad or a touch screen.
[0125] It is understood that the memory in the disclosed embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus random access memory (DRRAM). The memories described herein are intended to include, but are not limited to, these and any other suitable types of memories.
[0126] In some embodiments, the memory stores the following elements, executable modules or data structures, or a subset thereof, or an extended set thereof: an operating system and applications.
[0127] The operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process hardware-based tasks. The application includes various application programs, such as a media player (Media Player), a browser (Browser), etc., which are used to implement various application services. The program for implementing the method of the embodiment of the present disclosure can be included in the application.
[0128] In the above embodiment, the processor may also call a program or instruction stored in the memory, specifically, a program or instruction stored in an application program, and is used to:
[0129] Execute the steps of the above method.
[0130] The above method can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The above-disclosed methods, steps and logic block diagrams can be implemented or executed. The general processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the above-disclosed method can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor are combined to execute. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.
[0131] It is understood that the embodiments described in the present invention can be implemented by hardware, software, firmware, middleware, microcode or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), general processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in the present application or a combination thereof.
[0132] For software implementation, the technology of the present invention can be implemented by executing the functional modules (such as procedures, functions, etc.) of the present invention. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or outside the processor.
[0133] Example 4
[0134] The present invention may also provide a non-volatile storage medium for storing a computer program. When the computer program is executed by a processor, each step in the above method embodiment can be implemented.
[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that any modification or equivalent replacement of the technical solutions of the present invention does not depart from the spirit and scope of the technical solutions of the present invention and should be included in the scope of the claims of the present invention.
Claims
1. A method for predicting the relationship between diseases and drugs based on biological regulation mechanism graph representation learning, comprising: Input the disease name into the trained association prediction model to obtain candidate drugs and prediction scores for the disease; The association relationship prediction model is a graph representation learning model; The training process of the association relationship prediction model includes: Establishing a training set including a correlation network between known drugs and diseases, a correlation network between known drugs and proteins, and a correlation network between proteins and diseases; Construct heterogeneous biological networks based on training sets; Using metapathway strategies to construct regulatory maps of drugs, diseases, and proteins; According to the regulatory graphs of drugs, proteins and diseases, the graph information propagation mechanism is used to obtain the graph representation features of different regulatory graphs, and then the graph representation feature matrix of drugs and diseases is obtained; Based on the graph representation feature matrix of drugs and diseases, the XGBoost classifier is used to obtain the trained association prediction model.
2. The method for predicting the relationship between diseases and drugs based on biological regulation mechanism graph representation learning according to claim 1, characterized in that: The heterogeneous biological network G is: G = {V, E}; wherein, V represents the set of all nodes of the heterogeneous biological network, including drug nodes, disease nodes, and protein molecule nodes; E represents the set of all edges of the heterogeneous biological network, including the edges corresponding to the association between drugs and diseases, the edges corresponding to the association between drugs and protein molecules, and the edges corresponding to the association between protein molecules and diseases.
3. The method for predicting the relationship between diseases and drugs based on biological regulation mechanism graph representation learning according to claim 2, characterized in that: The meta-path strategy is used to construct regulatory maps of drugs, diseases and proteins; include: Step A1: define a meta-path set MP = {MP m } corresponds to m types of regulatory mechanisms; Step A2: Define the meta-path based MP m Constructed regulatory map It consists of a set of specific instances P_{m}, P_{m}={P1,P2,…,P i ,P s }, m∈MP, P_{m} represents one of the predefined meta-paths in the meta-path set MP, s represents the total number of paths generated according to P_{m}, and the path P i It consists of L sequential nodes, denoted as v1→…→v i →v i+1 →…v L ,v∈V, where v i to v i+1 The transition probability is: Among them, t represents the node type, which belongs to the defined node type set {drug, disease, protein molecule}, and the function Φ(t-1) is used to specify if the current node v i The type of the next node is t, then i+1 The type that should be, the set N(v i ,Φ(t-1)) contains all nodes related to node v i Adjacent nodes of type t-1 that meet the definition of function Φ.
4. The method for predicting the relationship between diseases and drugs based on biological regulation mechanism graph representation learning according to claim 3, characterized in that: According to the regulation graphs of drugs, proteins and diseases, the graph information propagation mechanism is used to obtain the graph representation features of different regulation graphs; Step B1: Define the control map The adjacency matrix of For the control diagram The graph representation features of the nodes in are obtained through the graph information propagation mechanism: X l =σ(A m X l-1 ) Among them, X l Represents the representation of the node at the lth layer. When the number of neural network propagation layers is l=0, X is used as the initial feature matrix, and one-hot encoding is used to initialize the features for each node; Step B2: Get the weighted average of features at different levels: in, is a list containing the minimum number of propagation layers l for all nodes v∈V, ε is the feature threshold, α l is the weight coefficient of the contribution of the lth layer, A l Indicates that node features are in the graph The state after l steps of propagation; Step B3: Define a tensor T to integrate the graph representation features from different control graphs: Among them, X m Representation Regulation Diagram The characteristics of the nodes in , M is the total number of control graph types initially set; V represents the total number of nodes, and d represents the dimension.
5. The method for predicting the relationship between diseases and drugs based on biological regulation mechanism graph representation learning according to claim 4, characterized in that: The method of obtaining the graph representation feature matrix of drugs and diseases includes: Step C1: Based on the tensor T, a nonlinear transformation function is used to aggregate the graph representation features f of each node in the control graph m : in, is the feature vector of node v in the m-th view control graph, is a trainable weight matrix, b m is the bias term; Step C2: Apply the softmax function to normalize the weight coefficients between different control graphs: Among them, β m For control diagram The relative importance of m is the attention score, W att is a trainable weight matrix, the superscript T indicates transpose; Step C3: According to the control graph weight β m Calculate the graph representation features of all regulatory graphs and perform weighted summation to obtain the feature matrix H of drugs and diseases:
6. The method for predicting the relationship between diseases and drugs based on biological regulation mechanism graph representation learning according to claim 5, characterized in that: The graph representation feature matrix based on drugs and diseases uses XGBoost classifier to obtain a trained association relationship prediction model, including: Step D1: Define a drug-disease pair dataset S, represented by S = {(s i ,y i )}, where 1≤i≤|S|, s i is the feature vector of the ith drug and disease pair, y i ∈{0,1} is the corresponding label, indicating whether there is an association between the drug and the disease, 0 means no association, 1 means association, |S| is the total number of drug-disease pairs in the dataset, S is input into the XGBoost classifier for training, and the objective function Obj is: Among them, the loss function Used to measure the actual label y of the i-th drug-disease pair i and predicted labels The deviation between them, K represents the total number of decision trees; Ω(f k ) is a regularization term used to constrain overly complex model structures. The specific calculation process is as follows: Where Le is the number of decision tree leaf nodes, γ and λ represent the complexity of the decision tree and the parameters of the weight constraint, respectively, ||ω|| is the 2-norm of the decision tree; f k (s i ) represents the feature vector of the i-th drug-disease pair, and the prediction result is calculated by the k-th decision tree.
7. A disease-drug association prediction system based on biological regulation mechanism graph representation learning, characterized in that: include: The prediction output module inputs the disease name into the trained association prediction model to obtain the candidate drugs and prediction scores for the disease; the association prediction model is a graph representation learning model; and The model training module is used to construct a heterogeneous biological network based on a training set including an association network of known drugs and diseases, an association network of known drugs and proteins, and an association network of proteins and diseases; extract and learn the regulatory graphs of drugs, diseases, and proteins in the heterogeneous biological network to obtain a feature matrix of drugs and diseases, and train to obtain an association relationship prediction model.
8. The disease-drug association prediction system based on biological regulation mechanism graph representation learning according to claim 7 is characterized in that: The model training module includes: A network construction submodule, for constructing a heterogeneous biological network based on a training set including an association network of known drugs and diseases, an association network of known drugs and proteins, and an association network of proteins and diseases; A feature learning submodule is used to construct the regulation graphs of drugs, diseases and proteins using a meta-path strategy, and to obtain the graph representation features of different regulation graphs using a graph information propagation mechanism according to the regulation graphs of drugs, proteins and diseases, and then obtain the graph representation feature matrix of drugs and diseases, and pass the feature matrix to the model training submodule; and The model training submodule is used to train the association relationship prediction model that meets the training requirements based on the results of the feature learning submodule.
9. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method of claim 1.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to claim 1 are implemented.
Citation Information
Cited By
Method for constructing network relationship among environmental exposure, biological process and diseases
CN119829805A
