Prediction method of anti-parasitic disease synergistic drug combination

Abstract

The invention discloses a method for predicting an anti-parasitic disease synergistic drug combination. The method comprises the following steps: S1, constructing an anti-parasitic disease synergistic drug combination data set; s2, establishing a hypergraph representing a drug cooperation relationship, and constructing an incidence matrix to describe a relationship between nodes and hyperedges in the hypergraph; s3, constructing an information enhanced hypergraph neural network model IHGNN, and learning node features; and S4, predicting a drug synergy score based on node features by using a full-connection neural network so as to achieve the purpose of screening a potential anti-parasitic disease synergy drug combination. According to the method, the defect that an existing prediction algorithm neglects the information of the synergistic relationship between the drug combination and the parasitic disease is overcome, and the potential anti-parasitic disease synergistic drug combination can be accurately recommended.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and particularly relates to a method for predicting a synergistic drug combination for anti-parasitic diseases. Background Art

[0002] Parasitic diseases are a class of infectious diseases caused by parasites and are widely distributed globally. These pathogens include various parasites such as malaria, schistosomiasis, cysticercosis, and echinococcosis, which can cause different degrees of infection and diseases inside the human body. Parasitic diseases pose a major threat to global health and socio-economic well-being, especially in developing countries. One of the key methods for treating parasitic infections currently is drug treatment. There are problems such as drug resistance, uncertain efficacy, side effects, and long treatment time in the monotherapy of parasitic infections. The emergence of resistance reduces the effectiveness of certain drugs, while adverse reactions and toxicity may have a negative impact on the health of patients. Therefore, current research efforts are dedicated to highlighting the potential of combination therapies and developing effective drug combinations to treat parasitic infections.

[0003] Generally speaking, compared with monotherapy, combination drug therapy has several advantages. Due to the synergistic effects of different drug combinations on target biological pathways, the overall effect of treating parasitic infections is enhanced. Synergistic drug combinations can enhance the treatment effect and promote the recovery of patients. When using a combination drug treatment regimen, the dosage of each drug can be reduced, thereby reducing the risk of adverse reactions and drug resistance in patients. This method aims to minimize discomfort during the treatment process.

[0004] However, using biological experiments to infer effective drug combinations for treating parasitic infections is a very obvious labor-intensive and time-consuming process. Deep learning techniques have been widely used to predict synergistic drug combinations for cancer, transforming the task into a binary classification problem. Deep learning techniques can process large-scale and complex biological data, helping to identify complex relationships and interactions between drugs. In addition, deep learning techniques can learn and identify hidden patterns in the data, promote the discovery of non-linear relationships in drug interactions, and improve the accuracy of prediction. Therefore, it is crucial to use deep learning techniques to predict drug combinations related to parasitic diseases. Converting the synergistic drug-drug-parasitic disease triple into a drug synergistic hypergraph can also accurately reflect the synergistic relationship between parasitic diseases and drug combinations. However, the number of hyperedges in the drug synergistic hypergraph is small, belonging to a sparse hypergraph. Therefore, designing effective strategies to fully utilize and exploit the existing information in the drug synergistic hypergraph is the key to predicting effective drug combinations for treating parasitic diseases. Summary of the Invention

[0005] To solve the existing problems, the present invention provides a method for predicting a synergistic drug combination for anti-parasitic diseases, and the specific scheme is as follows:

[0006] A prediction method for an anti-parasitic disease synergistic drug combination, comprising the following steps:

[0007] S1, constructing a dataset of anti-parasitic disease synergistic drug combinations;

[0008] S2, establishing a hypergraph representing drug synergy relationships, and constructing an incidence matrix to describe the relationships between nodes and hyperedges in the hypergraph;

[0009] S3, constructing an information-enhanced hypergraph neural network model IHGNN to learn node features;

[0010] S4, predicting whether a drug combination has a synergistic effect in parasitic diseases through a fully connected neural network FCN.

[0011] Preferably, the data information in the dataset in step S1 is sourced from the PubChem database.

[0012] Preferably, in step S2, a hypergraph G=(V, E, W) representing drug synergy relationships is established, where V = P ∪ D is a node set containing a parasitic disease set P and a drug set D, E ∈ p×d×d (p ∈ P, d ∈ D) is a hyperedge set representing drug-drug-parasitic disease triples with synergistic effects, and W ∈ R |E|×|E| is a diagonal matrix storing the weights assigned to all hyperedges.

[0013] Preferably, the method for constructing an incidence matrix to describe the relationships between nodes and hyperedges in the hypergraph in step S2 is as follows:

[0014] S21, designing a convolutional neural network CNN for target gene data acting on parasitic diseases to extract the feature representation H of parasitic disease nodes P ;

[0015] S22, designing a graph convolutional neural network GCN for SMILES data of drugs to extract the feature representation H of drug nodes D ;

[0016] S23, constructing an incidence matrix Y ∈ R |V|×|E| to describe the relationships between nodes and hyperedges in the hypergraph, and the feature matrix of all nodes is represented by where F is the feature dimension.

[0017] Preferably, in step S3, the hypergraph neural network HGNN is combined with a random hyperedge sampling method and the SE method to construct the IHGNN; at the topological level, designing random hyperedge sampling increases the randomness and diversity of the input data; at the semantic level, the hypergraph neural network learns the attention weights of node attributes by introducing the SE module.

[0018] Preferably, the specific method of step S4 is as follows: the node embeddings output by IHGNN are concatenated into triples as the input of FCN; the fully connected layer in FCN is designed as a spindle-shaped structure, and the Sigmoid function following the output of the last hidden layer is used to calculate the probability of the synergistic effect of two drugs in a specific parasitic disease.

[0019] The present invention also discloses a computer-readable storage medium with a computer program stored thereon. After the computer program runs, it executes the method described in any one of the above.

[0020] The present invention also discloses a computer system, including a processor and a storage medium. There is a computer program stored on the storage medium. The processor reads and runs the computer program from the storage medium to execute the method described in any one of the above.

[0021] The beneficial effects of the present invention are as follows:

[0022] 1. The present invention models the problem of predicting drug combinations for parasitic diseases as a binary classification problem, focusing on drug pairs that can exhibit synergistic effects. This modeling method provides a clear mathematical framework for predicting drug combinations.

[0023] 2. The present invention converts drug combinations and parasitic diseases into a hypergraph form, where drugs and cell lines are represented by nodes, and whether a drug combination has a synergistic therapeutic effect on a parasitic disease is represented by a hyperedge. The initial features of drug nodes are extracted from SMILES information through a graph convolutional neural network, while the initial features of parasitic disease nodes are extracted from target information using a convolutional neural network.

[0024] 3. The present invention proposes a prediction model of information-enhanced hypergraph neural network - IHGNNDDS. This model applies prior knowledge at the topological level and semantic level during the training process to enhance the information in the drug synergistic hypergraph. This information enhancement method helps to solve the problem of insufficient information in the drug synergistic hypergraph, thereby improving the accuracy of the prediction results.

[0025] 4. The present invention uses a random hyperedge sampling method to achieve the information enhancement function at the topological level. This method randomly removes hyperedges of the input hypergraph during the training process, and by encouraging the model to learn various hypergraph structures, it improves the robustness and generalization ability of the model. In addition, at the semantic level, a squeeze and excitation module is explicitly introduced to process node attributes. It assigns attention weights to each attribute of the node, enabling the model to selectively enhance features containing useful information and suppress irrelevant features by learning global information. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0027] Figure 1 Schematic diagram of the method flow of the present invention;

[0028] Figure 2 Method diagram of random hyperedge sampling of the present invention;

[0029] Figure 3 SE module diagram of the present invention;

[0030] Figure 4 Three cross-validation methods for sampling of the present invention. Specific implementation manners

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0032] As Figure 1 , a prediction method for an anti-parasitic disease synergistic drug combination, which is a prediction method for an anti-parasitic disease synergistic drug combination based on an information-enhanced hypergraph neural network, and includes the following steps:

[0033] S1. Construct a dataset of anti-parasitic disease synergistic drug combinations. The data information therein is sourced from the PubChem database. This dataset includes 335 drug combinations for treating 25 kinds of parasitic diseases that have been verified, that is, there are 335 drug-drug-parasitic disease triples. Specifically, through a comprehensive literature search in the PubChem database, about 200 literatures containing information on anti-parasitic disease synergistic drug combinations are screened out, and evidence that 335 drug combinations have a synergistic effect in treating parasitic diseases is extracted. These 335 synergistic drug combinations involve 190 drugs, including 158 approved drugs, 20 research drugs, and 12 experimental drugs.

[0034] S2. Establish a hypergraph representing the drug synergistic relationship and construct an incidence matrix to describe the relationship between nodes and hyperedges in the hypergraph.

[0035] Specifically, using the known information on anti-parasitic drug combinations, a hypergraph G=(V, E, W) of drug synergy relationships is constructed, where V = P ∪ D is the node set containing the set of parasitic diseases P and the set of drugs P, E ∈ p×d×d (p ∈ P, d ∈ D) is the hyperedge set representing the drug-drug-parasitic disease triples with synergistic effects, and W ∈ R |E|×|E| is a diagonal matrix storing the weights assigned to all hyperedges.

[0036] S21, design a convolutional neural network CNN for the target gene data of parasitic diseases to extract the feature representation H of parasitic disease nodes P .

[0037] Specifically, the initial features of the parasitic disease nodes in the hypergraph are composed of target information, which is collected and sorted from the GeneCards database. The target data involved is the set of target genes of all parasitic diseases in the drug synergy dataset, which contains a total of 1,741 target genes. Subsequently, design a convolutional neural network CNN for the target gene data of parasitic diseases to extract the feature representation H of parasitic disease nodes P .

[0038] Based on the CNN learning the gene expression information of cell lines, construct the initial features of cell line nodes. The gene expression data of cell lines is input into the embedding layer, which generates a 256-dimensional vector representation. Next is the 1D convolutional layer, which is applied sequentially to learn different levels of abstract features from the previous layer. The max pooling layer performs downsampling on the output of the convolutional layer. Finally, the fully connected layer uses the result of downsampling to construct the cell line vector representation H P ∈R NP×F , where N P represents the number of cell lines, and F represents the dimension of the features.

[0039] S22, design a graph convolutional neural network GCN for the SMILES data of drugs to extract the feature representation H of drug nodes D .

[0040] Specifically, the initial features of the drug nodes in the hypergraph are composed of the Simplified Molecular Input Line Entry System SMILES of drugs, which is collected and sorted from the DrugBank database. Subsequently, design a graph convolutional neural network GCN for the SMILES data of drugs to extract the feature representation H of drug nodes D .

[0041] Based on the GCN learning the molecular structure information of drugs, construct the initial features of drug nodes. Use the open-source cheminformatics software RDKit to convert the SMILES of drugs into molecular graphs. The graph of a given drug is defined as G d =(X d,A d ), where X d is the attribute matrix of all atoms, and A d is the adjacency matrix representing the chemical bonds between atomic nodes. Apply GCN to the drug molecular graph to construct the drug feature representation where N D represents the number of cell lines, and F represents the dimension of the features.

[0042] S23, construct the association matrix Y ∈ R |V|×|E| to describe the relationship between nodes and hyperedges in the hypergraph. The feature matrix of all nodes is represented by , where F is the feature dimension.

[0043] Specifically, design the association matrix Y ∈ R |V|×|E| to describe the relationship between nodes and hyperedges in the hypergraph. Specifically, if the node v ∈ V is connected by the hyperedge e ∈ E, then Y ve = 1; otherwise, Y ve = 0. The feature matrix of all nodes is represented by , where H P is the target gene feature of parasitic diseases, and H D is the molecular structure feature of the drug, and F is the feature dimension.

[0044] S3, construct the information-enhanced hypergraph neural network model IHGNN to learn node features. IHGNN consists of HGNN, a random sampling method, and an SE module.

[0045] The information-enhanced hypergraph neural network is developed based on HGNN. This network is implemented on a well-constructed hypergraph G = (V, E, W) to learn the node embeddings. HGNN applies the hyperedge convolution operator for hypergraph learning and implements node-edge-node transformation to refine the node features. The specific calculation process is defined as follows:

[0046]

[0047] where represents the node representation of the l-th layer, σ(·) represents the ReLU activation function, Q vv ∈ R |V|×|V| and P ee ∈ R |E|×|E| represent the degree matrix of nodes and the degree matrix of hyperedges, respectively, and is the learnable weight matrix of the l-th layer.

[0048] Due to the limited information contained in the hypergraph representing the drug synergy relationship, the hypergraph neural network HGNN may not be able to accurately reflect the relationships between nodes, thus affecting the accuracy of the prediction results. Therefore, an information-enhanced hypergraph neural network IHGNN is constructed by enhancing HGNN from both topological and semantic perspectives to fully exploit and utilize the information in the hypergraph.

[0049] For example Figure 2 , in terms of enhancing at the topological level, a random hyperedge sampling method is considered to disrupt the original hypergraph to increase the diversity of available information. The design idea of random hyperedge sampling is influenced by the DropEdge method, which is an effective technique to prevent overfitting and oversmoothing when training various deep GCN models. The key of DropEdge is to randomly discard a certain number of edges from the input adjacency matrix in each training epoch. In fact, DropEdge can be regarded as a way of data augmentation. By DropEdge, different variants of the original graph can be randomly generated, which leads to an increase in the randomness and diversity of the input data.

[0050] The random hyperedge sampling method refers to randomly discarding a certain number of hyperedges of the input hypergraph during each training. By performing the random hyperedge sampling technique on the original hypergraph, a sub-hypergraph is obtained, so the sub-hypergraph becomes more random and diverse as the new input data. In addition, during the process of performing random hyperedge sampling, the unsampled hyperedges and their corresponding messages are removed from the original hypergraph, which results in sparser connections between vertices and reduces the message passing in the sub-hypergraph. The random hyperedge sampling technique can mitigate overfitting and oversmoothing by information enhancement and reducing the messages of stacked hyperedge convolutional layers.

[0051] Specifically, at the topological level, inspired by the DropEdge method, designing random hyperedge sampling increases the randomness and diversity of the input data, thus playing a role in information enhancement. To illustrate more specifically, a hypergraph g with N nodes and M hyperedges can be represented by an incidence matrix Y g ∈R N×M . By performing random hyperedge operations on the hypergraph g, a subgraph sg can be obtained, whose incidence matrix is Y sg ∈R N×M′ . It should be noted that M′ represents the number of hyperedges in the subgraph sg, which can be calculated by the following equation: M′ = pM(4). Where the probability p represents the proportion of the original hyperedges retained in sg. Figure 2 Shows the conversion process from Y g to Y sg . The original hypergraph contains eight nodes and four hyperedges, while after performing the random hyperedge sampling operation, with a probability of p = 0.5, the subgraph only contains two hyperedges.

[0052] In terms of semantic-level enhancement, the squeeze and excitation (SE) method is adopted to adjust the features output by the convolutional layer, which enables the neural network to selectively enhance informative features. The SE method is an attention mechanism for neural networks, aiming to improve the network's attention to useful features while suppressing unimportant features. The core idea of this method is to enhance the network's representation ability by explicitly modeling the mutual dependencies between convolutional feature channels.

[0053] In the Squeeze step, the SE architecture uses global average pooling to compress the feature maps of each channel. Specifically, it averages the feature maps of each channel in the spatial dimensions (height and width) to obtain a scalar representing the global information of that channel. This scalar can be regarded as a channel descriptor, which contains the global information of the channel's feature map but does not contain spatial location information. In the Excitation step, the SE architecture uses a self-gating mechanism to learn the weights of each channel. This mechanism first reduces the dimension of the channel descriptor through a fully connected layer (also known as the bottleneck layer) to reduce the computational amount and increase the model's non-linearity. Then, a non-linear transformation is performed through the ReLU activation function, and the number of channels is restored to the original through another fully connected layer. Finally, the Sigmoid activation function is used to compress the weight values to between 0 and 1. The larger the weight value, the more important the features of that channel, and vice versa.

[0054] Specifically, at the semantic level, inspired by the SE method, an SE module is introduced into the hypergraph neural network to learn the attention weights of node attributes. As Figure 3 shown, the SE module follows the convolutional layer and is used to learn attention weights, which are combined with the features extracted by the convolutional layer in a linear combination manner to construct reweighted features. The specific calculation process is as follows:

[0055]

[0056] where and represent the output features of the l-th layer and the node attention weights respectively, Pooling represents the global average pooling function, W1 and W2 are the corresponding weight matrices, and φ and σ represent the Sigmoid and ReLU activation functions respectively.

[0057] To make full use of the information in the drug combination hypergraph, a random hyperedge sampling method and the SE method are selectively introduced to form an information enhancement module, which is combined with HGNN to generate IHGNN. After generating the subgraph sg using random hyperedges, the calculation details of IHGNN are as follows:

[0058]

[0059] where represents the node representation of the l-th layer, and Y sg is the incidence matrix of the sub-hypergraph, and are the degree matrix of the nodes, the weight matrix, and the degree matrix of the hyperedges of the sub-hypergraph, respectively. When introducing the SE module to process the node features output by the convolutional layer, the node representation of the l-th layer in the hypergraph can be changed to:

[0060]

[0061] Through L iterations, the features of the nodes can be updated by aggregating the information of their L-hop neighbors connected by common hyperedges, which means that the complex relationship between parasitic diseases and drug combinations is fully utilized during training. If then z a = Z[a, :], z b = Z[b, :], and z c = Z[c, :] can be used to represent the features of the parasitic disease node a, the drug node b, and the drug node c, respectively.

[0062] S4. Predict whether the drug combination has a synergistic effect in the parasitic disease through a fully connected neural network (FCN).

[0063] After constructing the representations of a parasitic disease and two drugs in the IHGNN, it can be predicted whether the drug combination will have a synergistic effect in this parasitic disease through a fully connected neural network (FCN). Specifically, the node embeddings output by the IHGNN are concatenated into a triple, which serves as the input to the FCN. The fully connected layers in the FCN are designed in a spindle-shaped structure, and the Sigmoid function following the output of the last hidden layer is used to calculate the probability of the synergistic effect of the two drugs in a specific parasitic disease.

[0064] Specifically, after constructing the parasitic disease feature z a and the drug features z b and z c in the IHGNN, the FCN can be used to predict whether drugs b and c have a synergistic effect in this parasitic disease a. Specifically, the node features output by the IHGNN are concatenated into a drug-drug-parasitic disease triple, which serves as the input to the FCN. The fully connected layers in the FCN are designed in a spindle-shaped structure, and the Sigmoid function following the last hidden layer is used to calculate the probability of the synergistic effect of the two drugs in a specific parasitic disease. The process is as follows:

[0065] P abc = ρ(W out ·a l) (9)

[0066] Among them, P represents the collaborative probability, l represents the number of hidden layers, ρ(·) is the Sigmoid function, and W out represents the weight matrix. a in the above equation l represents the feature vector learned from the previous layer and can be calculated by the following equation:

[0067]

[0068] Where represents the weight matrix, δ(·) represents the tanh activation function, and a 0 = concat(z a , z b , z c ) is used to represent the original input vector. Binary cross-entropy is used as the loss function for the collaborative drug combination prediction task.

[0069] IHGNNDDS is compared with the baseline methods HypergraphSynergy, DeepDDS, DeepWalk, node2vec, and TransE under three cross-validation methods (as Figure 4 shown) to reveal the performance of IHGNNDDS. The performance of each model is shown in Table 1.

[0070] Table 1

[0071]

[0072] Specifically, both IHGNNDDS and HypergraphSynergy outperform other comparative models in terms of the scores of AUC (Area Under Curve), AUPR (Area Under the Precision-Recall Curve), and ACC (Accuracy). HypergraphSynergy also uses hypergraphs to represent the relationships between drug combinations and diseases, and uses HGNN to learn the node features in the hypergraph. This indicates that introducing hypergraphs to capture and reflect the synergistic information between drug combinations and parasitic diseases can significantly improve the prediction performance of the model. In the random cross-validation scenario, IHGNNDDS outperforms the hypergraph-based comparative model HypergraphSynergy by 4.25%, 4.32%, and 3.40% in terms of the scores of AUC, AUPR, and ACC, respectively. This can prove that IHGNN is stronger than HGNN in capturing the information existing in the drug synergy hypergraph. Due to the hierarchical cross-validation based on parasitic diseases and drug combinations, the performance of the prediction method on unknown parasitic diseases and drug combinations can be verified. Compared with random CV, the performance of all methods has decreased to varying degrees. Compared with all baseline methods, the decrease of IHGNNDDS in the evaluation metrics is the smallest, but it still continuously obtains the best performance among all methods. Therefore, it can be proved that IHGNNDDS has sufficient ability to infer drug combinations for potentially treatable parasitic diseases.

[0073] The present invention also discloses a computer-readable storage medium and a computer system. Among them, a computer program is stored on a computer-readable storage medium. After the computer program runs, it executes the method as described above. A computer system includes a processor and a storage medium. A computer program is stored on the storage medium. The processor reads and runs the computer program from the storage medium to execute the method as described above.

[0074] Those skilled in the art will further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability of hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Skilled artisans may implement the described functionality in different ways for each specific application, but such implementation decisions should not be construed as causing a departure from the scope of the present invention.

[0075] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functions may be stored on or transmitted via a computer-readable medium as one or more instructions or code. The computer-readable medium includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. The storage media may be any available media that can be accessed by a computer. By way of example and not limitation, such computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a web site, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of the medium. As used herein, disk and disc include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and Blu-ray disc, where disk generally reproduces data magnetically, while disc uses lasers to optically reproduce data. Combinations of the above should also be included within the scope of computer-readable media.

[0076] The foregoing description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0077] Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting synergistic antiparasitic drug combinations, characterized in that: The following steps are involved: S1, Construction of a dataset of synergistic drug combinations against parasitic diseases; S2, establish a hypergraph representing the drug synergy relationship, and construct an association matrix to describe the relationship between nodes and hyperedges in the hypergraph; S3, builds the information enhanced hypergraph neural network model IHGNN to learn the node features in the hypergraph; S4, using a fully connected neural network to predict drug synergy scores based on node features, in order to achieve the purpose of screening potential synergistic anti-parasitic drug combinations.

2. The method according to claim 1, characterized in that: The data information of the data set in step S1 comes from the PubChem database.

3. The method according to claim 1, characterized in that: In step S2, a hypergraph G = (V, E, W) representing the drug synergistic relationship is established, where V = P∪D is a node set containing the parasitic disease set P and the drug set D, E∈p×d×d(p∈P,d∈D) is a hyperedge set representing drug-drug-parasitic disease triples with synergistic effects, and W∈R |E|×|E| is a diagonal matrix that stores the weights assigned to all hyperedges.

4. The method according to claim 3, characterized in that The method of constructing the association matrix in step S2 to describe the relationship between nodes and hyperedges in the hypergraph is as follows: S21, Design a convolutional neural network (CNN) acting on target gene data of parasitic diseases to extract feature representations of parasitic disease nodes H P ; S22, design a graph convolutional neural network (GCN) acting on drug SMILES data to extract the feature representation of drug nodes H D ; S23, construct the incidence matrix Y∈R |V|×|E| To describe the relationship between nodes and hyperedges in the hypergraph, the feature matrix of all nodes is given by It indicates that F is the feature dimension.

5. The method according to claim 1, characterized in that In step S3, the hypergraph neural network HGNN is combined with the random hyperedge sampling method and the Squeeze-and-Excitation method, namely the SE method, to construct the IHGNN. At the topological level, the design of random hyperedge sampling increases the randomness and diversity of the input data. At the semantic level, the hypergraph neural network learns the attention weights of node attributes by introducing the SE module.

6. The method according to claim 1, characterized in that The specific method of step S4 is as follows: the node embeddings output by the IHGNN are concatenated into triplets as the input of the FCN; the fully connected layer in the FCN is designed as a spindle-shaped structure, and the output of the last hidden layer followed by the Sigmoid function is used to calculate the probability of synergistic effects of two drugs in specific parasitic diseases.

7. A computer-readable storage medium, characterized in that: A computer program is stored on the medium, and after the computer program is run, the method according to any one of claims 1 to 6 is executed.

8. A computer system, characterized in that: The method comprises a processor and a storage medium, wherein a computer program is stored in the storage medium, and the processor reads and runs the computer program from the storage medium to execute the method as claimed in any one of claims 1 to 6.

Images