Drug target prediction method and system fusing structural features and system features

Abstract

The invention discloses a drug target prediction method and system fusing structural features and system features, and belongs to the technical field of computer-aided drug design, and the system comprises a data preprocessing module which is used for preprocessing input data information and constructing a dynamic interaction network; the structural biological feature extraction module is used for extracting structural biological features of the medicine and the target spot; the dynamic system biological feature extraction module is used for extracting dynamic system biological features of the drug-target interaction network; and the prediction module is used for acquiring an interaction prediction score of an input drug and a target spot based on the structural biological characteristics and the dynamic system biological characteristics. According to the method, any input drug target pairs can be predicted, the predictable range is expanded, and particularly, the robustness of the method is enhanced on the prediction of new drugs and new targets.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer-aided drug design, and particularly relates to a drug target prediction method and system integrating structural features and systematic features. Background Art

[0002] Drug-Target Interaction (DTI) prediction methods are mainly divided into two categories: structure-based methods and system-based methods. Structure-based methods mainly use one-dimensional sequences as inputs, such as the Simplified Molecular Input Line Entry System (SMILES) or the amino acid sequences of drug fingerprints and targets. These methods usually have a Y-shaped structure. First, the features of drugs and targets are extracted separately, and then connected as the basis for DTI prediction. Many studies also use the attention mechanism to capture the semantic correlation between local and global features during the feature extraction process. Some methods convert drugs or targets into graphs and apply Graph Neural Networks (GNN) technology to extract molecular features, thereby improving the model performance. In addition, other methods use spatial neural networks to extract features from 3D structure data to predict the interactions between related drugs and targets. Although structure-based methods have achieved good performance in DTI tasks, their disadvantages are also obvious. These methods only focus on the intrinsic features of entities and ignore the significant impact of the interactions between entities on DTI prediction. In addition, such methods overly rely on the structural similarity between entities, which greatly limits the potential to improve the model performance.

[0003] System-based methods usually predict DTI relationships within relevant biological networks, combining the relationship information between different entities, thereby providing a new perspective for DTI prediction. Compared with structural information, heterogeneous networks have a stronger ability to describe the complex relationships between entities, thus providing more comprehensive feature information. Some methods regard the biological interaction network as a graph, where entities are nodes and edges represent different relationships between entities, and transform the DTI problem into a link prediction problem on the graph. Some methods integrate various types of similarity relationships to construct more complex heterogeneous networks for reasoning and prediction. Other methods select a third type of entity, rather than drugs or targets, as the "bridging node" and indirectly predict the relationship between drugs and targets by examining the relationships between this node and drugs and targets. Although system-based methods perform well in DTI prediction tasks, the cold start problem of being unable to predict new node relationships limits the scalability of these methods in new drug discovery.

[0004] Heterogeneous interaction networks provide new insights for DTI prediction tasks and demonstrate excellent performance by integrating more diverse attributes. Some methods predict DTI tasks from a more comprehensive feature perspective by introducing other relevant biological entities and constructing more complex and comprehensive interaction networks. Some methods improve DTI prediction performance by defining meta-paths with specific semantic information in heterogeneous interaction networks to capture the complex relationships between different types of nodes and edges. Some knowledge graph-based methods improve the accuracy and interpretability of drug target prediction through high-order relationship reasoning. However, such methods require training and learning on large-scale graph data, which requires a large amount of time and computational resources.

[0005] It can be seen that there are certain deficiencies in the two current methods for predicting drug-target interactions (DTI), mainly focusing on: (1) Structure-based methods only use the structural attribute features of drugs and targets, lacking the application and integration of interaction information, so the prediction effect is poor; (2) System-based methods are limited by the limitations of related computational methods such as graph neural network methods, resulting in their inability to be applied to the prediction of new drugs or targets, restricting the application scope of their methods and reducing their robustness. Therefore, the present invention proposes a drug target prediction method and system that fuses structural features and system features. Summary of the Invention

[0006] To solve the above technical problems, the present invention proposes a drug target prediction method and system that fuses structural features and system features to solve the problems existing in the above prior art.

[0007] To achieve the above object, the present invention provides a drug target prediction system that fuses structural features and system features, including:

[0008] A data preprocessing module for preprocessing input data information and constructing a dynamic interaction network;

[0009] A structural biology feature extraction module for extracting the structural biology features of drugs and targets;

[0010] A dynamic systems biology feature extraction module for extracting the dynamic systems biology features of the drug-target interaction network;

[0011] A prediction module for obtaining the interaction prediction scores of the input drugs and targets based on the structural biology features and the dynamic systems biology features.

[0012] Optionally, the data preprocessing module includes: a data input module, a feature embedding module, and a dynamic interaction network module;

[0013] Among them, the data input module is used to obtain the simplified molecular linear input specification of the drug and the amino acid sequence of the target, and convert the simplified molecular linear input specification of the drug and the amino acid sequence of the target into an initial encoding in the form of a digital vector;

[0014] The feature embedding module is used to process the initial encoding to obtain initial features, where the initial features include: initial drug structural features, initial target structural features, initial drug system features, and initial target system features;

[0015] The dynamic interaction network module is used to construct a dynamic interaction network based on the input drug ID information and target ID information.

[0016] Optionally, the structural biology feature extraction module includes: a drug structural biology feature extraction module and a target structural biology feature extraction module;

[0017] Among them, the drug structural biology feature extraction module is used to obtain drug structural biology features based on the initial drug structural features;

[0018] The target structural biology feature extraction module is used to obtain target structural biology features based on the initial target structural features.

[0019] Optionally, the dynamic system biology feature extraction module includes: an initial dynamic interaction network construction module and a drug-target dynamic interaction network system biology feature extraction module;

[0020] Among them, the initial dynamic interaction network construction module is used to construct a heterogeneous dynamic interaction network based on the dynamic interaction network, the initial drug system features, and the initial target system features;

[0021] The drug-target dynamic interaction network system biology feature extraction module is used to extract features from the drug-target initial dynamic interaction network to obtain system biology features.

[0022] The present invention also discloses a drug-target prediction method that fuses structural features and system features, including the following steps:

[0023] Perform feature embedding on the simplified molecular linear input specification of the drug and the amino acid sequence of the target to obtain initial features, and at the same time construct a heterogeneous dynamic interaction network based on the drug ID information and the target ID information;

[0024] Extract drug structural biology features and target structural biology features from the initial features;

[0025] Use the GraphSAGE graph neural network method to process the initial features and the dynamic interaction network to obtain system biology features;

[0026] Predict the drug-target relationship based on the drug structural biology characteristics, target structural biology characteristics, and systems biology characteristics to obtain the interaction prediction score between the drug and the target.

[0027] Optionally, the process of obtaining the initial features includes:

[0028] Perform character conversion on the simplified molecular linear input specification of the drug and the amino acid sequence of the target to obtain the initial encoding of the drug and the target;

[0029] Perform feature encoding on the initial encoding of the drug and the target respectively to obtain the initial features, where the initial features include: drug initial structural features, target initial structural features, drug initial system features, and target initial system features.

[0030] Optionally, the process of constructing the heterogeneous dynamic interaction network based on the drug ID information and the target ID information includes:

[0031] Create a homogeneous drug-drug interaction network based on the drug ID information;

[0032] Create a homogeneous target-target interaction network based on the target ID information;

[0033] Construct the connecting edges of the homogeneous drug-drug interaction network and the homogeneous target-target interaction network based on the drug-target interaction information;

[0034] Construct a heterogeneous dynamic interaction network based on the connecting edges of the homogeneous drug-drug interaction network and the homogeneous target-target interaction network.

[0035] Optionally, the expression for feature extraction of the initial features is:

[0036]

[0037] In the formula, X is the input feature, k1, k2, and k3 are the learnable parameter matrices of the first, second, and third CNN layers respectively, b1, b2, and b3 are the bias parameters of the first, second, and third CNN layers respectively, * represents the convolution operation, ReLU represents the activation function, and H out represents the output of the module.

[0038] Optionally, the process of obtaining the systems biology characteristics using the GraphSAGE graph neural network method includes: sequentially passing the node features of the drug nodes and the target nodes through message generation, message aggregation, and message update to obtain the systems biology characteristics;

[0039] Among them, when there is no edge between any of the drug nodes and target nodes and feature update cannot be completed, the original node features of the current node are used as the updated features for splicing.

[0040] Optionally, the expressions for generating messages, aggregating messages, and updating messages for drug nodes and target nodes are as follows:

[0041]

[0042]

[0043]

[0044] In the formula, MESSAGE (k) is the message generation function of the k-th layer network, and respectively represent the node features of node v and node u at the (k - 1)-th layer, indicates that the direction of edge e is from node u to node v, represents the message passed from node u to node v generated in the k-th layer network, represents the message aggregation function for relationship r in the k-th layer network, R is the set of relationships r, and u is an element in the neighbor node set N r (v) of node v, represents the transfer information of the messages of all neighbor nodes of node v aggregated in the k-th layer network, UPDATE (k) represents the message update formula of the k-th layer network, represents the node feature of node v at the (k - 1)-th layer, represents the node feature of node v at the k-th layer.

[0045] Compared with the prior art, the present invention has the following advantages and technical effects:

[0046] The drug target prediction system and method of the present invention realize efficient and accurate prediction of drug-target interactions by fusing structural features and system features. First, the system dynamically generates a drug-target interaction network framework through a data preprocessing module, providing a basis for subsequent analysis. The structural biology feature extraction module accurately extracts the structural features of drugs and targets, and combines with the dynamic system biology feature extraction module to analyze the dynamic features of the interaction network, comprehensively capturing the interaction mechanism between drugs and targets. The prediction module then accurately calculates the interaction prediction score based on multi-dimensional features.

[0047] At the method level, the initial structural features are obtained through feature embedding, and the GraphSAGE graph neural network is used to process the dynamic interaction network, further updating the drug and target feature splicing, and finally achieving accurate prediction. This method not only takes into account the static structural information of drugs and targets, but also incorporates the system characteristics of dynamic interaction networks, significantly improving the accuracy and reliability of predictions. Compared with traditional methods, the present invention can more comprehensively reflect the true interaction relationship between drugs and targets, providing strong support for drug development and target discovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0049] Figure 1 A flow chart of a drug target prediction method integrating structural features and system features according to an embodiment of the present invention;

[0050] Figure 2 The heterogeneous drug-target interaction network construction and dynamic system biology characteristics learning of the embodiments of the present invention;

[0051] Figure 3 It is the system biology feature update of the embodiment of the present invention, wherein 1) indicates that neither the drug nor the target node features are updated, 2) indicates that only the drug node feature is updated, 3) indicates that only the target node feature is updated, and 4) indicates that both the drug and target node features are updated;

[0052] Figure 4 The figure is a schematic diagram of the process of predicting and screening asiatic acid-related targets according to an embodiment of the present invention, wherein 1) represents the preliminary test of asiatic acid and the target to be tested, 2) represents the screening of the targets predicted to be "interacting" after the test, 3) performs a virtual molecular docking experiment on the above-mentioned predicted interacting targets, and further screens them according to the binding free energy, and 4) the targets that meet the conditions are regarded as potential interacting targets. DETAILED DESCRIPTION

[0053] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0054] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0055] like Figure 1As shown, in this embodiment, a drug target prediction system that integrates structural features and system features is provided. The DTI prediction system designed in the present invention integrates structural and system features, taking both into account and improving the prediction performance. At the same time, the interaction information is used as auxiliary information rather than model input, enabling the prediction of any input drug-target pair and enhancing the robustness of the method. Specifically, it includes: a data preprocessing module, a structural biology feature extraction module, a dynamic systems biology feature extraction module, and a prediction module. Among them, the data preprocessing module is responsible for preprocessing the input data information and dynamically generating a drug-target interaction network framework. The structural biology feature extraction module and the dynamic systems biology feature extraction module are responsible for extracting the key feature attributes of drugs and targets, providing a basis for subsequent interaction prediction.

[0056] As a specific implementation of this embodiment, the drug target prediction system that integrates structural features and system features includes:

[0057] A data preprocessing module, which includes: a data input module, a feature embedding module, and a dynamic interaction network module. Among them, the data input module is used to obtain the simplified molecular linear input specification of the drug and the amino acid sequence of the target, and convert the simplified molecular linear input specification of the drug and the amino acid sequence of the target into an initial encoding in digital vector form. The feature embedding module is used to process the initial encoding to obtain initial features, where the initial features include: initial drug structural features, initial target structural features, initial drug system features, and initial target system features. The dynamic interaction network module is used to construct a heterogeneous dynamic interaction network based on the input drug ID information and target ID information.

[0058] In this embodiment, the Simplified Molecular Input Line Entry System (SMILES) of the drug and the Amino Acid Sequence of the target are input through the data input module. First, the SMILES of the drug and the Amino Acid Sequence of the target are initially encoded to convert the data in string form into the initial encoding in digital vector form, which facilitates the feature extraction calculation of subsequent modules. Then, the initial encoding needs to pass through the feature embedding module to obtain the initial structural features of the drug and the target, and the initial system features of the drug and the target respectively. At the same time, the dynamic interaction network module searches in the relevant drug-target interaction data according to the ID information of the input drug and target, and dynamically constructs the edge information of the heterogeneous drug-target interaction network. Finally, the initial features are used as the input of the subsequent structural biology feature extraction module, and the initial system features are used as the node features in the drug-target interaction network. Combined with the edge information of the dynamic heterogeneous drug-target interaction network, they jointly form a heterogeneous dynamic interaction network, which is used as the input of the subsequent dynamic systems biology feature extraction module.

[0059] The construction process of the heterogeneous dynamic interaction network is as follows. First, the drug-drug interactions related to the input drug ID are searched, and a homogeneous drug-drug interaction network with a specified neighbor hop number and a specified number of neighbor nodes is created. The same operation is performed on the target to create a homogeneous target-target interaction network. Finally, by querying the drug-target interaction information, the connecting edges between the two homogeneous networks are constructed, and the two homogeneous networks are constructed into a heterogeneous drug-target interaction network, that is, the heterogeneous dynamic interaction network. The dynamic nature of this network is reflected in that for any input drug-target pair, the number of nodes, the number of edges, and the connectivity of its network are different, and only the corresponding network can be determined during the operation of the method.

[0060] The structural biology feature extraction module includes a drug structural biology feature extraction module and a target structural biology feature extraction module. Among them, the drug structural biology feature extraction module is used to obtain the drug structural biology features based on the initial structural features of the drug, and the target structural biology feature extraction module is used to obtain the target structural biology features based on the initial structural features of the target.

[0061] This module is responsible for extracting the structural biology features of drugs and targets. Taking the initial structural features of drugs and targets from the data preprocessing module as input, this module uses the most commonly used sequence feature extraction module in the field to extract the structural biology features of drugs and targets, that is, it adopts 3 consecutive one-dimensional convolutional neural network (CNN) layers and 1 maxpooling layer. CNN can efficiently capture local features and gradually abstract complex patterns of sequence structures through hierarchical learning; the maxpooling layer improves computational efficiency through dimensionality reduction and enhances the robustness of the model to input changes. At the same time, the parameter sharing of CNN reduces the risk of overfitting, automatic feature extraction avoids manual intervention, and its parallel computing ability and adaptability to variable-length sequences further improve the efficiency and scalability of the model.

[0062] Dynamic system biology feature extraction module, which includes: an initial dynamic interaction network construction module and a drug-target dynamic interaction network system biology feature extraction module; among them, the initial dynamic interaction network construction module is used to construct a drug-target initial dynamic interaction network based on the dynamic interaction network, drug initial system features, and target initial system features; the drug-target dynamic interaction network system biology feature extraction module is used to extract features from the drug-target initial dynamic interaction network to obtain system biology features.

[0063] This module mainly conducts feature representation learning based on the initial dynamic heterogeneous drug-target interaction network related to the input drug targets constructed by the above data preprocessing module, uses the GraphSAGE method of graph neural network for learning, and obtains system biology features as the output of this system feature extraction module.

[0064] Prediction module: Using the outputs of both the structural biology feature extraction module and the dynamic system biology feature extraction module as the input of this prediction module, generally adopting a 3-layer fully connected layer as the architecture of the model, to predict the drug-target relationship and obtain the interaction prediction score between the drug and the target.

[0065] Example 2

[0066] The present invention proposes a method for predicting drug-target relationships by fusing structural features and system features. The method for predicting drug-target relationships by fusing structural features and system features is applied to a drug-target prediction system, and specifically includes the following steps:

[0067] Data preprocessing: a) Organize and integrate drug-target interaction data from different data sources as the basis for dynamically constructing a drug-target related dynamic interaction network. The relevant data sources involve multiple databases such as DrugBank, TTD, Uniport, BioGRID, Binding DB, Drugs, PharmgKB, etc. There are 18,897 drug-target interactions, 85,831 target-target interactions, and more than 2 million drug-drug interaction data; b) Initialize the encoding of the SMILES of drugs and the amino acid sequence of targets, convert specific characters into corresponding numbers, and realize the numericalization of sequences; c) Respectively pass the initial encodings of drugs and targets through different Embedding layers for feature encoding to obtain the initial structural features of drugs and targets and the initial system features of drugs and targets. The initialized system features are used as node features in the subsequent dynamic interaction network; d) In the biological interaction data organized and integrated in a), search for relevant interaction information of drugs and targets. First, construct a homogeneous drug interaction network and a homogeneous target interaction network with specified neighbor numbers and neighbor hop numbers respectively, and then construct the above homogeneous networks into a heterogeneous drug-target interaction network through drug-target interaction information to provide connectivity information of the interaction network, that is, edge information; e) The initial structural features of drugs and targets obtained in c) above are used as the input of the structural biology feature extraction module, and the initial system features of drugs and targets obtained are used as the initial node features in the interaction network. Together with the interaction network connectivity information obtained in d), they jointly constitute the input of the dynamic systems biology feature extraction module.

[0068] Structural biology feature extraction: Respectively use a 3-layer one-dimensional convolutional neural network and a 1-layer max pooling layer to extract features from the initial structural features of drugs and targets. The feature extraction processes of drugs and targets are independent of each other and do not affect each other. The calculation formula for this part is as follows:

[0069]

[0070] Among them, X is the input of this module, that is, the initial features of drug SMILES or target sequence. k1, k2, and k3 are the learnable parameter matrices of the first, second, and third CNN layers respectively. b1, b2, and b3 are the bias parameters of the first, second, and third CNN layers respectively. * represents the convolution operation, and ReLU represents the activation function. H out represents the output of this module, that is, the structural biology features of drugs or targets.

[0071] Dynamic systems biology feature extraction: As Figure 2 and Figure 3As shown in the figure, a) The GraphSAGE method of the graph neural network is used to extract the dynamic system biology characteristics of the drug-target interaction network. Specifically, in the process of d) in the data preprocessing stage, by controlling two parameters, namely the number of isomorphic interaction network layers of the target node and the number of neighbor nodes of each layer of nodes, the scale of the generated dynamic interaction network is controlled. Therefore, theoretically, for different input drug-target pairs, the model will generate interaction networks with different scales, shapes, and densities. Subsequently, GraphSAGE generates the embedded representation of the target node by sampling neighbor nodes and aggregating neighbor information. This embedded representation contains not only its own features but also relevant features of the network structure. Especially when performing message aggregation, first, separate aggregation is performed for different relationships r, and then the features under all different relationships are uniformly aggregated to form the final transmitted message and passed downwards. Therefore, it contains richer and more comprehensive network structure feature information. The corresponding message generation, message aggregation, and message update formulas are as follows:

[0072]

[0073]

[0074]

[0075] The first formula above is the message generation formula: MESSAGE (k) represents the message generation function of the k-th layer network, where and represent the node features of node v and node u at the (k - 1)-th layer respectively, indicates that the direction of edge e is from node u to node v, that is, node v is the target node; represents the message passed from node u to node v generated in the k-th layer network. The second formula is the message aggregation formula: represents the message aggregation function for relationship r in the k-th layer network, and R is the set of relationship r; is the message generated by the above message generation formula, where u is an element in the neighbor node set N r (v) of node v, that is, node u is a neighbor node of node v, represents the transmission information that aggregates the messages of all neighbor nodes of node v in the k-th layer network. The third formula is the message update formula, UPDATE (k) represents the message update formula of the k-th layer network, is the message from all neighbor nodes, represents the node feature of node v at the (k - 1)-th layer, that is, the updated feature of the target node is composed of the node feature of this node in the previous layer network and the aggregated messages of all neighbor nodes in this layer network.

[0076] b) If the node features of both the drug node and the target node can be updated through the above methods, the system biology feature of the interaction network is the concatenation of the updated node features of the drug node and the target node, that is, the system biology feature, namely

[0077]

[0078] where H network represents the dynamic system biology feature of the current input drug-target pair, h drugnode represents the updated node feature of the drug node in the current input drug-target pair, and h targetnode represents the updated node feature of the target node in the current input drug-target pair. If any of the nodes cannot be updated through the GraphSAGE method due to the absence of edges, the original node feature of that node is used as the updated feature for the above concatenation.

[0079] Prediction: a) Concatenate the above structural biology features of the drug and the target and the system biology feature of the drug-target interaction network as the input of this module. The calculation formula is as follows:

[0080]

[0081] where H represents the final feature output of this method, H drug represents the structural biology feature of the current input drug, H target represents the structural biology feature of the current input target, H network represents the dynamic system biology feature of the current input drug-target pair, and ⊕ represents the concatenation operation of feature vectors. Subsequently, through three fully connected layers of calculation and learning, the interaction prediction score of the input drug and target is finally obtained.

[0082] b) Use the cross-entropy loss function to train the parameters of the model. The formula is as follows:

[0083]

[0084] where y represents the true label of the training data, represents the predicted label of the model.

[0085] The present invention provides rich and effective attribute features for the prediction of drug-target interactions by fusing structural biology features and system biology features, and improves the accuracy of drug-target relationship prediction to a certain extent.

[0086] The heterogeneous drug target interaction network related to the system biology characteristics of the present invention is dynamically constructed, and the scale of the network can be actively controlled. Compared with other prediction methods using biological networks, the adopted interaction network has a smaller scale and a higher correlation with the input drug-target pairs, reducing the dependence on computing resources to a certain extent and simultaneously reducing the time complexity.

[0087] Benefiting from the fact that the input of the method only requires the structural information of the drug and the target, the present invention can predict for any input drug-target pair, expanding the predictable range, especially in the prediction of new drugs and new targets, and enhancing the robustness of the method at the same time.

[0088] The above are only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A drug target prediction system that integrates structural features and system features, characterized in that, Including: A data preprocessing module for preprocessing input data information and constructing a dynamic interaction network; A structural biology feature extraction module for extracting the structural biology features of drugs and targets; A dynamic systems biology feature extraction module for extracting the dynamic systems biology features of the drug-target interaction network; A prediction module for obtaining the interaction prediction score of the input drug and target based on the structural biology features and the dynamic systems biology features.

2. The drug target prediction system integrating structural features and system features according to claim 1, characterized in that The data preprocessing module includes: a data input module, a feature embedding module, and a dynamic interaction network module; Among them, the data input module is used to obtain the simplified molecular linear input specification of the drug and the amino acid sequence of the target, and convert the simplified molecular linear input specification of the drug and the amino acid sequence of the target into an initial encoding in digital vector form; The feature embedding module is used to process the initial encoding to obtain initial features, where the initial features include: initial drug structural features, initial target structural features, initial drug system features, and initial target system features; The dynamic interaction network module is used to construct a dynamic interaction network based on the input drug ID information and target ID information.

3. The drug target prediction system integrating structural features and system features according to claim 2, characterized in that, The structural biology feature extraction module includes: a drug structural biology feature extraction module and a target structural biology feature extraction module; Among them, the drug structural biology feature extraction module is used to obtain drug structural biology features based on the initial drug structural features; The target structural biology feature extraction module is used to obtain target structural biology features based on the initial target structural features.

4. The drug target prediction system integrating structural features and system features according to claim 3, characterized in that, The dynamic systems biology feature extraction module includes: an initial dynamic interaction network construction module and a drug-target dynamic interaction network systems biology feature extraction module; Among them, the initial dynamic interaction network construction module is used to construct a heterogeneous dynamic interaction network based on the dynamic interaction network, the initial drug system features, and the initial target system features; The drug-target dynamic interaction network systems biology feature extraction module is used to extract feature extraction from the drug-target initial dynamic interaction network to obtain systems biology features.

5. A drug target prediction method that integrates structural features and system features, characterized in that, Including the following steps: Performing feature embedding on the simplified molecular linear input specification of the drug and the amino acid sequence of the target to obtain initial features, and at the same time constructing a heterogeneous dynamic interaction network based on the drug ID information and the target ID information; Performing feature extraction on the initial features to obtain drug structural biology features and target structural biology features; Using the GraphSAGE graph neural network method to process the initial features and the dynamic interaction network to obtain systems biology features; Performing drug-target relationship prediction based on the drug structural biology features, target structural biology features, and systems biology features to obtain the interaction prediction score of the drug and the target.

6. The method for predicting drug targets by integrating structural features and system features according to claim 5, wherein The process of obtaining the initial features includes: Performing character conversion on the simplified molecular linear input specification of the drug and the amino acid sequence of the target to obtain the initial encoding of the drug and the target; The initial encodings of the drug and the target are respectively subjected to feature encoding to obtain initial features, where the initial features include: drug initial structural features, target initial structural features, drug initial systematic features, and target initial systematic features.

7. The drug target prediction method integrating structural features and system features according to claim 6, wherein The process of constructing a heterogeneous dynamic interaction network based on drug ID information and target ID information includes: Creating a homogeneous drug-drug interaction network based on the drug ID information; Creating a homogeneous target-target interaction network based on the target ID information; Constructing the connecting edges of the homogeneous drug-drug interaction network and the homogeneous target-target interaction network based on drug-target interaction information; Constructing a heterogeneous dynamic interaction network based on the connecting edges of the homogeneous drug-drug interaction network and the homogeneous target-target interaction network.

8. The method for predicting drug targets by integrating structural features and system features according to claim 7, wherein The expression for feature extraction of the initial features is: where X is the input feature, k1, k2, and k3 are the learnable parameter matrices of the 1st, 2nd, and 3rd CNN layers respectively, b1, b2, and b3 are the bias parameters of the 1st, 2nd, and 3rd CNN layers respectively, * represents the convolution operation, ReLU represents the activation function, and H out represents the output of the module.

9. The method for predicting a drug target by integrating structural features and system features according to claim 8, characterized in that, The process of obtaining systematic biology features using the GraphSAGE graph neural network method includes: successively passing the node features of drug nodes and target nodes through message generation, message aggregation, and message update to obtain systematic biology features; Among them, when there is no edge between any of the drug nodes and target nodes and feature update cannot be completed, the original node features of the current node are used as the updated features to be concatenated.

10. The method for predicting a drug target by integrating structural features and system features according to claim 9, wherein The expression for successively passing the node features of drug nodes and target nodes through message generation, message aggregation, and message update is: where MESSAGE (k) is the message generation function of the k-th layer network, and represent the node features of nodes v and u at the (k - 1)-th layer respectively, indicates that the direction of edge e is from node u to node v, represents the message passed from node u to node v generated by the k-th layer network, represents the message aggregation function for relationship r in the k-th layer network, R is the set of relationships r, and u is an element in the neighbor node set N r (v) of node v, represents the transfer information aggregating the messages of all neighbor nodes of node v in the k-th layer network, UPDATE (k) represents the message update formula of the k-th layer network, represents the node feature of node v at the (k - 1)-th layer, represents the node feature of node v at the k-th layer.

Images