Anticancer drug collaborative prediction method based on mapping knowledge domain embedding, GNN and Bi-LSTM

Abstract

The invention provides an anti-cancer drug collaborative prediction method based on knowledge graph embedding, GNN and Bi-LSTM, and belongs to the technical field of anti-cancer drug collaborative prediction. In order to solve the problem of poor collaborative prediction effect of anti-cancer drugs, the technical scheme adopted by the invention comprises the following steps: respectively obtaining corresponding relations of drug, protein, cell line and tissue interaction from a database; constructing a knowledge graph according to the corresponding relationship, and grouping all the drug combinations based on the collaborative scores of different drug combinations; based on a drug combination grouping result and the constructed knowledge graph, extracting a neighbor entity set interacting with the drug or the cell line, constructing an attention layer of the knowledge graph, and iteratively updating the initial entity embedding representation of each entity in the neighbor entity set; taking an entity neighborhood structure generated by iterative updating as input, extracting a high-order structure and a semantic relationship through a graph neural network, and learning medicine and neighborhood representation thereof; the method is applied to collaborative prediction of anti-cancer drugs.

Description

Technical Field

[0001] The present invention provides an anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM, belonging to the technical field of anticancer drug synergy prediction. Background Art

[0002] With the development of bioinformatics and deep learning technology, the research and development of anti-cancer drugs has entered a new stage. Currently, the traditional anti-cancer drug development method relies on a large number of laboratory research and clinical trials, which is time-consuming, costly, and has a relatively low success rate. According to research, in order to improve the efficiency and success rate of anti-cancer drug development, multiple drugs can be used in combination to reduce drug toxicity. This method is an effective anti-cancer drug development strategy.

[0003] However, there are still some limitations in predicting the effects of drug combinations. For example, in anti-cancer drug research, traditional machine learning methods often ignore the complexity of drug action and the network characteristics of biological systems, which will limit the model's in-depth understanding of the mechanism of drug synergy and result in large prediction errors for drug synergy. Summary of the invention

[0004] In order to overcome the deficiencies in the prior art, the present invention aims to solve the following technical problem: to provide an anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: an anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM, including the following prediction steps:

[0006] Step S1: respectively obtaining from the database: corresponding connections between drug and protein interaction, corresponding connections between cell line and protein interaction, corresponding connections between protein and protein interaction, and corresponding connections between cell line and tissue interaction;

[0007] Step S2: construct a knowledge graph based on the above correspondence, and group all drug combinations based on the synergy scores of different drug combinations;

[0008] Step S3: Based on the drug combination grouping results and the constructed knowledge graph, extract the neighbor entity set that interacts with the drug or cell line, build the attention layer of the knowledge graph, and iteratively update the initial entity embedding representation of each entity in the neighbor entity set;

[0009] Step S4: Taking the entity neighborhood structure generated by iterative updating as input, extracting high-order structures and semantic relationships through graph neural networks, and learning the representation of drugs and their neighborhoods;

[0010] Step S5: construct an attention-based pooling module to learn the interactive information between drug pairs, and aggregate the embedded representation into a single vector by strengthening the representation of drug pairs;

[0011] Step S6: Use Bi-LSTM network to perform regression analysis on drug and cell embedding, and after analysis, connect them to a multi-layer fully connected network to predict the synergy score;

[0012] Step S7: Use the MSELoss function to calculate the error, use the Adam optimizer to perform multiple rounds of training on the model, and output the final anticancer drug combination solution.

[0013] The specific method of constructing the knowledge graph and grouping according to the collaborative scores in step S2 is:

[0014] Step S21: Screening out groups of drug combinations and cell lines with potential predictive associations;

[0015] Step S22: Each sample consists of two anticancer drugs, a cancer cell, and a related synergy score. The synergy score in DrugCombDB is defined as a ZIP value, and the synergy score in Oncology-Screen is defined as a Loewe additive value.

[0016] Group them according to the score relationship and arrange all drug combinations in descending order of synergy scores:

[0017] The drug combinations corresponding to the synergistic scores in the first quarter are defined as synergistic, and the drug combinations corresponding to the synergistic scores in the second quarter are defined as antagonistic;

[0018] Step S23: The knowledge graph displays the information of the association between drugs and proteins, cell lines and proteins, cell lines and tissues, and proteins and proteins. When a piece of information is transmitted to the target protein and cell line through the knowledge graph, the embedding of the target protein and cell line will increase the neighbor information of other relationships.

[0019] Step S24: Represent the knowledge graph in the form of a triple T = (h, t, r) of multiple entities and relationships. The specific method is:

[0020] Define the knowledge graph of drug i-drug j-cell line, expressed as:

[0021] G = (ε, R);

[0022] Where ε is drug, protein, cell line, and tissue, and R is the four corresponding relationships obtained in step S1;

[0023] Define h, t∈ε in the triple to represent the head entity and the tail entity respectively;

[0024] Define r∈R in a triple to represent the relationship between two entities.

[0025] The specific method for iteratively updating the initial entity embedding representation of each entity in the neighbor entity set in step S3 is:

[0026] Step S31: Find the entity set related to drugs / cell lines in the knowledge graph, and define drugs / cell lines as parameters m;

[0027] Step S32: Recursively represent the adjacent entity set of m as:

[0028]

[0029] Where l is the propagation depth of KG, (h, t, r) represents a set of triples, G is the knowledge graph, and L is the total propagation depth of the knowledge graph;

[0030] Step S33: Construct entity adjacency matrix A according to entity sets e And the relationship adjacency matrix A r , where (A e ,A r ∈R v*n ), v refers to the number of entities, and n refers to the number of fixed neighbor entities;

[0031] In the relation adjacency matrix A r The relationship between each entity and its neighbor entities includes the four types of corresponding connections obtained in step S1;

[0032] Step S34: Based on the grouping results of the knowledge graph and drug combinations, the propagation mechanism of the knowledge graph is used to collect high-order association information with drugs / cell lines, and the vector representation of the enhanced entity is improved.

[0033] In step S4, the specific method of extracting high-order structures and semantic relationships through the graph neural network and learning the representation of drugs and their neighborhoods is:

[0034] Step S41: learn the embedded representation of the node, introduce the graph neural network, and perform neighborhood sampling on each node;

[0035] Step S42: learning neighborhood information and setting parameter H, which is defined as the perception domain in CNN;

[0036] The receptive field depth of a given drug pair is set to 2, which means that the second-order connected nodes are considered;

[0037] Step S43: Obtain the final embedding representation of the entity by aggregating neighborhood information, and define the node directly connected to the drug node as N neigh (e), using a fixed neighborhood range S(e).

[0038] The specific method of constructing the attention-based pooling module in step S5 to learn the interactive information between drug pairs and perform aggregation is:

[0039] Step S51: input the aggregated entity into the constructed attention-based pooling module;

[0040] Step S52: The attention-based pooling module assigns a score to each drug substructure and performs weighted summation of the embeddings of all nodes to obtain a graph-level representation;

[0041] Step S53: After sampling is completed, the embedded representation of the entity itself and the embedded representation of the neighborhood information are aggregated through three aggregation methods to finally obtain the embedded representation of the current entity:

[0042] The final entity representation is aggregated through three aggregation methods and the output is a single vector. The sum aggregation method is defined as an overlay operation, the concat aggregation method is a concatenation operation, and the neighbor aggregation method only considers the neighborhood information and ignores its own information. The corresponding aggregation expression is:

[0043] Sum aggregation:

[0044] Concat aggregation:

[0045] Neighbor aggregation:

[0046] Among them, σ is the Sigmoid activation function; W and b are trainable weights and biases; e is an entity and S(e) is the neighborhood range.

[0047] The specific method for predicting the synergy score in step S6 is:

[0048] Specifically, a bidirectional long short-term memory network (Bi-LSTM) is used to predict the collaboration score:

[0049] Step S61: preparing the input sequence, dividing the input sequence into time steps of equal length, each time step being a feature vector;

[0050] Step S62: construct a Bi-LSTM model, including a forward LSTM and a backward LSTM, wherein the forward LSTM is responsible for processing from the beginning of the sequence backward, and the backward LSTM is responsible for processing from the end of the sequence forward, and each LSTM unit includes an input gate, a forget gate, and an output gate component;

[0051] Step S63: forward propagation, inputting the input sequence into the forward LSTM time step by time step to obtain the forward hidden state sequence of each time step;

[0052] Step S64: Backward propagation, inputting the input sequence into the backward LSTM time step by time step to obtain the backward hidden state sequence of each time step;

[0053] Step S65: merging hidden states, merging the forward and backward hidden state sequences according to the time step, the merged hidden state sequence contains the comprehensive information at each time step in the sequence, and considers the information from the beginning and the end of the sequence;

[0054] Step S66: Output layer processing, inputting the merged hidden state sequence into two fully connected layers.

[0055] The specific method of the model continuously learning the output solution in step S7 is:

[0056] Specifically, we use MSELoss to calculate the error and use the Adam optimizer for training:

[0057] Step S71: Calculate the difference between the predicted value and the true value, the difference of each sample is a vector or a matrix;

[0058] Step S72: Calculate the square difference, square each difference value to obtain the corresponding square error;

[0059] Step S73: Calculate the mean, sum the square errors of all samples, and then divide by the number of samples to obtain the average square error, which is the value of the mean square error loss function;

[0060] Step S74: Back propagation, taking the mean square error as the loss, and calculating the gradient of the loss function with respect to the model parameters through the back propagation algorithm;

[0061] Step S75: Use the Adam optimizer to update the model parameters according to the gradient of the loss function so that the model approaches the optimal solution:

[0062] According to the gradient of the loss function, the learning rate lr, and the momentum parameter β1, the exponential decay rate β2 of the gradient second moment is calculated. The first-order moment estimate is updated first, and then the second-order moment estimate is updated. Then, the first-order and second-order moment estimate deviations are corrected, the parameters are updated, and the iteration is repeated.

[0063] The beneficial effects of the present invention compared with the prior art are as follows: the present invention proposes an anticancer drug synergy prediction method based on knowledge graph embedding, graph neural network training and bidirectional long short-term memory network prediction. The knowledge graph used is a heterogeneous graph, in which nodes represent entities and edges represent relationships between entities. The rich semantic connections between entities can help the system discover potential relationships between entities. The knowledge graph is used to learn entities such as drugs, proteins and their topological neighbors to obtain a more stable embedding representation; a graph neural network is used to receive the topological information of each entity in the knowledge graph, and features are learned through an attention-based pooling module to better understand the module information; a Bi-LSTM model is used to capture the interaction between drug combinations and cells. Two reverse LSTM layers are intricately arranged together, and its bidirectional structure allows the model to consider past and future information at the same time. After the regression analysis is completed, the output is projected into a fully connected network to obtain a synergy score, thereby improving the prediction accuracy and efficiency of anticancer drug combinations and optimizing the screening and evaluation process of drug combinations in clinical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The present invention will be further described below in conjunction with the accompanying drawings:

[0065] Figure 1 is a flowchart of the steps of the anticancer drug synergy prediction method of the present invention;

[0066] Figure 2 This is a model structure diagram of a neural network used in the prediction method of the present invention. DETAILED DESCRIPTION

[0067] like Figure 1 and Figure 2 As shown, the present invention proposes a synergistic prediction method for anticancer drug combination based on knowledge graph embedding, graph neural network training (GNN), and bidirectional long short-term memory network (Bi-LSTM). The prediction method can effectively improve the prediction accuracy and efficiency of anticancer drug combination, and optimize the screening and evaluation process of drug combination in clinical application; the method specifically constructs a knowledge graph, uses graph neural network training, learns interactive information based on the attention pooling module, and performs Bi-LSTM network regression analysis, and finally outputs the synergistic score through a fully connected network, thereby realizing intelligent and efficient prediction of the effect of anticancer drug combination use; the method specifically includes the following steps:

[0068] Step S1: Collect and obtain four types of entities and their corresponding relationships;

[0069] Among them: four types of entities include drugs, proteins, cell lines, and tissues;

[0070] The corresponding relationships refer to: protein and drug-protein interactions, cell line-protein interactions, protein-protein interactions, and cell line-tissue interactions.

[0071] The data used in this invention are from public databases, specifically:

[0072] Drug-protein interactions are collected from six commonly used databases: DrugBank, BindingDB, ChEMBL, etc., which contain 15,015 drug-protein associations.

[0073] Cell line-protein interactions were obtained from the reaction dataset of Cancer Cell Line Encyclopedia (CCLE), which includes 18,022 protein-coding genes, 1,035 cancer cell lines and 749,551 cell line-protein associations.

[0074] Protein-protein interactions were derived from the human protein-protein interactome.

[0075] Cell line-tissue interactions were obtained from DrugCombDB and Oncology-Screen.

[0076] Step S2: pre-process the data and group them according to the collaboration score;

[0077] First, we screened out groups with potential predictive associations between drug combinations and cell lines. Each sample consisted of two anticancer drugs, a cancer cell, and a related synergy score. The synergy score in DrugCombDB was the ZIP value, and the synergy score in Oncology-Screen was the Loewe additive value.

[0078] All drug combinations are grouped based on their synergy scores: all drug combinations are sorted in descending order of synergy scores;

[0079] Drug combinations corresponding to the first quarter of the synergy scores were classified as synergistic, and drug combinations corresponding to the last quarter of the synergy scores were classified as antagonistic, so that low-confidence combination drug samples could be excluded.

[0080] Synergy was considered as a positive sample, and antagonism was considered as a negative sample. The preprocessed data was saved and divided into a training set and a test set in an 8:2 ratio to train and evaluate the effectiveness of the treatment recommendation model.

[0081] Then build the knowledge graph:

[0082] The knowledge graph displays information about drug-protein, cell line-protein, cell line-tissue, and protein-protein associations. When a piece of information is passed to the target protein and cell line through the knowledge graph, their embeddings will increase the neighbor information of other relationships;

[0083] The knowledge graph can be represented as multiple entity-relationship-entity triplets T=(h,r,T); specifically, the knowledge graph of drug i-drug j-cell line is represented as G=(ε,R), where ε refers to the above four types of entities (drug, protein, cell line, tissue), and R refers to the four types of relationships (drug-protein, cell line-protein and cell line-tissue and protein-protein associations), and h,T∈ε in the triplet represent the head entity and the tail entity respectively, and r∈R represents the relationship between two entities.

[0084] Step S3: Extract neighbor entity sets and improve the initial entity embedding representation:

[0085] Adjacent entities in the knowledge graph are always highly correlated, and this module is used to obtain multi-hop neighbor nodes of drugs / cell lines.

[0086] Step S31: First, find the entity set related to the drug / cell line in the knowledge graph. In the embodiment of the present invention, the drug / cell line is abbreviated as m; the adjacent entities of m can be recursively represented as:

[0087]

[0088] Where l is the propagation depth of KG, (h, t, r) represents a set of triples, where G is the knowledge graph, and L is the total propagation depth of the knowledge graph;

[0089] Step S33: Then construct entity adjacency matrix A according to entity sets e And the relationship adjacency matrix A r , where (A e ,A r ∈R v*n ), where v refers to the number of entities and n refers to the fixed number of neighbor entities. The relationship adjacency matrix includes the relationship between each entity and its neighbor entities, including four types of relationships: drug-protein, protein-protein, cell line-protein, and cell line-tissue;

[0090] Step S34: Finally, based on the grouping results of the knowledge graph and drug combinations, the propagation mechanism of the knowledge graph is used to collect high-order association information with drugs / cell lines, thereby improving and enhancing the vector representation of the entity.

[0091] Step S4: Build a network model, including a graph neural network (GNN) and an attention based pooling module, to encode drug features and neighborhood structures between drug pairs, extract high-order structures and semantic relationships through the graph neural network, and learn the representation of drugs and their neighborhoods:

[0092] Step S41: Directly learn the embedding representation of the node without considering the rich neighborhood information of each entity. Introduce the graph neural network to perform neighborhood sampling on each node.

[0093] Step S42: In the process of learning neighborhood information, in order to learn more neighborhood information, a parameter H is set considering the neighborhood range of two hops of each drug node, which can be understood as the perception field in CNN. In the present invention, the receptive field depth of a given drug pair is set to 2, indicating that the second-order connected node situation is considered. Of course, H can take a larger value.

[0094] Step S43: Obtain the final embedding representation of the entity by aggregating neighborhood information, and define the node directly connected to the drug node as N neigh (e) Since the neighborhood distribution of each drug node is different, for the convenience of calculation, the GraphSAGE calculation method is used, using a fixed neighborhood range S(e).

[0095] Step S5: Construct an attention-based pooling module to learn the interaction information between drug pairs and perform aggregation:

[0096] Step S51: input the aggregated entity into the constructed attention-based pooling module;

[0097] Step S52: The above-mentioned aggregated entities are input into the pooling module as input. The use of attention-based pooling helps the model consider which substructures in the chemical substances are more important for predicting synergistic drugs. By using the attention-based pooling module, not only the interaction information between drug pairs can be obtained, but also the important chemical substructures of the drugs can be identified; the attention-based pooling module is used to assign a score to each drug substructure and perform weighted summation of the embeddings of all nodes to obtain a graph-level representation, expressed as:

[0098]

[0099] In the above formula, and They are the graph embedding matrices of drug x and drug, taken from the last layer of GNN, and obtained by the operation of tanh() function. x ,A y , represents the score assigned to each substructure of the drug. x and ay is the attention score of the drug pair, which is obtained by summing up each score and then activating it with softmax(). Where M represents the node count in the graph. The vectors of all nodes are weighted and summed according to the attention score to obtain the final graph-level representation, which is expressed as:

[0100] g x =multiply(a x ,H x );

[0101] g y =multiply(a y ,H y );

[0102] Therefore, any entity node can be enhanced and represented by the operation of step S52, and the expression is:

[0103] e=attention-basedpooling(E);

[0104] Among them, attention-based pooling() represents the above pooling operation.

[0105] Step S53: In the aggregation process, the embedded representation of the entity itself and the embedded representation of the neighborhood information are aggregated through three aggregation methods to finally obtain the embedded representation of the current entity.

[0106] The final entity representation is aggregated through three aggregation methods and output as a single vector. Among them, the sum aggregation method is an overlay operation, concat is a concatenation operation, and neighbor only considers the neighborhood information and ignores its own information. The expression is:

[0107] Sum aggregation:

[0108] Concat aggregation:

[0109] Neighbor aggregation:

[0110] Among them, σ is the Sigmoid activation function; W and b are trainable weights and biases; e is an entity and S(e) is the neighborhood range.

[0111] Step S6: Use a bidirectional long short-term memory network (Bi-LSTM) to make predictions. The specific steps include:

[0112] Step S61: preparing the input sequence, dividing the input sequence into time steps of equal length, where each time step is usually a feature vector;

[0113] Step S62: construct a Bi-LSTM model, where the forward LSTM is responsible for processing from the beginning of the sequence backward, and the backward LSTM is responsible for processing from the end of the sequence forward. Each LSTM unit includes components such as an input gate, a forget gate, and an output gate;

[0114] Step S63: Forward propagation, input the input sequence into the forward LSTM time step by time step, and obtain the forward hidden state sequence of each time step. The forward hidden state sequence represents the state information at each time step in the sequence, and can capture the temporal dependency;

[0115] Step S64: Backward propagation, inputting the input sequence into the backward LSTM time step by time step to obtain a backward hidden state sequence of each time step, and the backward hidden state sequence reflects the state information at each time step in the sequence;

[0116] Step S65: merge hidden states, merge the forward and backward hidden state sequences by time step, usually by simply concatenating by position. The merged hidden state sequence contains comprehensive information at each time step in the sequence, while taking into account information from both the beginning and the end of the sequence;

[0117] Step S66: Output layer processing, inputting the merged hidden state sequence into two fully connected layers;

[0118] This setting is beneficial for the task because the effects of the interaction between drugs and cells are not affected by the order in which they are taken, and future contexts of the sequence are as important as past contexts.

[0119] Step S7: Use MSELoss to calculate the error and use the Adam optimizer for training. The specific steps include:

[0120] Step S71: Calculate the difference between the predicted value and the true value. For each sample, this difference can be a vector or a matrix.

[0121] Step S72: Calculate the square difference, square each difference value, and obtain the corresponding square error. The purpose is to eliminate positive and negative differences and only focus on the size of the difference between the predicted value and the true value.

[0122] Step S73: Calculate the mean, sum the square errors of all samples, and then divide by the number of samples to get the average square error, which is the value of the mean square error loss function.

[0123] Step S74: Back propagation, taking the mean square error as the loss, and calculating the gradient of the loss function to the model parameters through the back propagation algorithm, so as to update the parameters of the model and minimize the loss function.

[0124] Step S75: Use the Adam optimizer to update the model parameters according to the gradient of the loss function so that the model approaches the optimal solution;

[0125] The model is trained using the Adam optimizer with a maximum of 100,000 iterations. Based on the gradient of the loss function, the learning rate (lr), the momentum parameter (β1), the exponential decay rate (β2) of the gradient second moment, and the parameter stability parameter (ε) are used to calculate the gradient second moment. First, the first-order moment estimate is updated, then the second-order moment estimate is updated, and then the first-order and second-order moment estimate deviations are corrected, the parameters are updated, and the iterations are repeated. The Adam optimizer adaptively adjusts the learning rate based on the historical gradient of each parameter. Different parameters have different learning rates, so that learning is more effective.

[0126] The model is continuously optimized and adjusted through multiple rounds of training and learning, and related weight parameters. The model effectively improves the prediction accuracy and efficiency of the anti-cancer drug combination, thereby optimizing the screening and evaluation process of drug combinations in clinical applications. Ultimately, it can provide patients with diagnosis and treatment recommendations tailored to their individual conditions, including measures such as changing drugs or adjusting dosages for different drug combinations, to assist doctors in providing more personalized and efficient diagnosis and treatment.

[0127] 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. Anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM, characterized by: The prediction steps include the following: Step S1: respectively obtaining from the database: corresponding connections between drug and protein interaction, corresponding connections between cell line and protein interaction, corresponding connections between protein and protein interaction, and corresponding connections between cell line and tissue interaction; Step S2: construct a knowledge graph based on the above correspondence, and group all drug combinations based on the synergy scores of different drug combinations; Step S3: Based on the drug combination grouping results and the constructed knowledge graph, extract the neighbor entity set that interacts with the drug or cell line, build the attention layer of the knowledge graph, and iteratively update the initial entity embedding representation of each entity in the neighbor entity set; Step S4: Taking the entity neighborhood structure generated by iterative updating as input, extracting high-order structures and semantic relationships through graph neural networks, and learning the representation of drugs and their neighborhoods; Step S5: construct an attention-based pooling module to learn the interactive information between drug pairs, and aggregate the embedded representation into a single vector by strengthening the representation of drug pairs; Step S6: Use Bi-LSTM network to perform regression analysis on drug and cell embedding, and after analysis, connect them to a multi-layer fully connected network to predict the synergy score; Step S7: Use the MSELoss function to calculate the error, use the Adam optimizer to perform multiple rounds of training on the model, and output the final anticancer drug combination solution.

2. The anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM according to claim 1, characterized in that: The specific method of constructing the knowledge graph and grouping according to the collaborative scores in step S2 is: Step S21: Screening out groups of drug combinations and cell lines with potential predictive associations; Step S22: Each sample consists of two anticancer drugs, a cancer cell, and a related synergy score. The synergy score in DrugCombDB is defined as a ZIP value, and the synergy score in Oncology-Screen is defined as a Loewe additive value. Group them according to the score relationship and arrange all drug combinations in descending order of synergy scores: The drug combinations corresponding to the synergistic scores in the first quarter are defined as synergistic, and the drug combinations corresponding to the synergistic scores in the second quarter are defined as antagonistic; Step S23: The knowledge graph displays the information of the association between drugs and proteins, cell lines and proteins, cell lines and tissues, and proteins and proteins. When a piece of information is transmitted to the target protein and cell line through the knowledge graph, the embedding of the target protein and cell line will increase the neighbor information of other relationships. Step S24: Represent the knowledge graph in the form of a triple T = (h, t, r) of multiple entities and relationships. The specific method is: Define the knowledge graph of drug i-drug j-cell line, expressed as: G = (ε, R); Where ε is drug, protein, cell line, and tissue, and R is the four corresponding relationships obtained in step S1; Define h, t∈ε in the triple to represent the head entity and the tail entity respectively; Define r∈R in a triple to represent the relationship between two entities.

3. The anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM according to claim 2, characterized in that: The specific method for iteratively updating the initial entity embedding representation of each entity in the neighbor entity set in step S3 is: Step S31: Find the entity set related to drugs / cell lines in the knowledge graph, and define drugs / cell lines as parameters m; Step S32: Recursively represent the adjacent entity set of m as: Where l is the propagation depth of KG, (h, t, r) represents a set of triples, G is the knowledge graph, and L is the total propagation depth of the knowledge graph; Step S33: Construct entity adjacency matrix A according to entity sets e And the relationship adjacency matrix A r , where (A e ,A r ∈R v*n ), v refers to the number of entities, and n refers to the number of fixed neighbor entities; In the relation adjacency matrix A r The relationship between each entity and its neighbor entities includes the four types of corresponding connections obtained in step S1; Step S34: Based on the grouping results of the knowledge graph and drug combinations, the propagation mechanism of the knowledge graph is used to collect high-order association information with drugs / cell lines, and the vector representation of the enhanced entity is improved.

4. The anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM according to claim 3, characterized in that: In step S4, the specific method of extracting high-order structures and semantic relationships through the graph neural network and learning the representation of drugs and their neighborhoods is: Step S41: learn the embedded representation of the node, introduce the graph neural network, and perform neighborhood sampling on each node; Step S42: learning neighborhood information and setting parameter H, which is defined as the perception domain in CNN; The receptive field depth of a given drug pair is set to 2, which means that the second-order connected nodes are considered; Step S43: Obtain the final embedding representation of the entity by aggregating neighborhood information, and define the node directly connected to the drug node as N neigh (e), using a fixed neighborhood range S(e).

5. The anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM according to claim 4, characterized in that: The specific method of constructing the attention-based pooling module in step S5 to learn the interactive information between drug pairs and perform aggregation is: Step S51: input the aggregated entity into the constructed attention-based pooling module; Step S52: The attention-based pooling module assigns a score to each drug substructure and performs weighted summation of the embeddings of all nodes to obtain a graph-level representation; Step S53: After sampling is completed, the embedded representation of the entity itself and the embedded representation of the neighborhood information are aggregated through three aggregation methods to finally obtain the embedded representation of the current entity: The final entity representation is aggregated through three aggregation methods and the output is a single vector. The sum aggregation method is defined as an overlay operation, the concat aggregation method is a concatenation operation, and the neighbor aggregation method only considers the neighborhood information and ignores its own information. The corresponding aggregation expression is: Sum aggregation: Concat aggregation: Neighbor aggregation: Among them, σ is the Sigmoid activation function; W and b are trainable weights and biases; e is an entity and S(e) is the neighborhood range.

6. The anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM according to claim 5, characterized in that: The specific method for predicting the synergy score in step S6 is: Specifically, a bidirectional long short-term memory network (Bi-LSTM) is used to predict the collaboration score: Step S61: preparing the input sequence, dividing the input sequence into time steps of equal length, each time step being a feature vector; Step S62: construct a Bi-LSTM model, including a forward LSTM and a backward LSTM, wherein the forward LSTM is responsible for processing from the beginning of the sequence backward, and the backward LSTM is responsible for processing from the end of the sequence forward, and each LSTM unit includes an input gate, a forget gate, and an output gate component; Step S63: forward propagation, inputting the input sequence into the forward LSTM time step by time step to obtain the forward hidden state sequence of each time step; Step S64: Backward propagation, inputting the input sequence into the backward LSTM time step by time step to obtain the backward hidden state sequence of each time step; Step S65: merging hidden states, merging the forward and backward hidden state sequences according to the time step, the merged hidden state sequence contains the comprehensive information at each time step in the sequence, and considers the information from the beginning and the end of the sequence; Step S66: Output layer processing, inputting the merged hidden state sequence into two fully connected layers.

7. The anticancer drug synergy prediction method based on knowledge graph embedding, GNN and Bi-LSTM according to claim 6, characterized in that: The specific method of the model continuously learning the output solution in step S7 is: Specifically, we use MSELoss to calculate the error and use the Adam optimizer for training: Step S71: Calculate the difference between the predicted value and the true value, the difference of each sample is a vector or a matrix; Step S72: Calculate the square difference, square each difference value to obtain the corresponding square error; Step S73: Calculate the mean, sum the square errors of all samples, and then divide by the number of samples to obtain the average square error, which is the value of the mean square error loss function; Step S74: Back propagation, taking the mean square error as the loss, and calculating the gradient of the loss function with respect to the model parameters through the back propagation algorithm; Step S75: Use the Adam optimizer to update the model parameters according to the gradient of the loss function so that the model approaches the optimal solution: According to the gradient of the loss function, the learning rate lr, and the momentum parameter β1, the exponential decay rate β2 of the gradient second moment is calculated. The first-order moment estimate is updated first, and then the second-order moment estimate is updated. Then, the first-order and second-order moment estimate deviations are corrected, the parameters are updated, and the iteration is repeated.

Images