Drug interaction prediction method based on interaction enhancement graph self-attention mechanism

Abstract

The invention discloses a drug interaction prediction method based on an interaction enhancement graph self-attention mechanism, and belongs to the technical field of drug interaction prediction, and the method comprises the steps: carrying out the modeling of a drug under a local view and a global view, so as to obtain a molecular graph and a motif graph; designing an interaction-enhanced graph Transform module, respectively updating node embedding in two views of the molecular graph through an interaction-enhanced attention mechanism, and fusing through a multi-channel self-adaptive fusion module; enabling the extracted features to pass through a feedforward neural network, a residual connection layer and a normalization layer to obtain a final molecular representation of a molecular graph and a motif graph; designing a dynamic interaction scaling prediction module, and calculating a DDI score; and constructing a drug interaction prediction model, designing a loss function, and optimizing model parameters. According to the invention, the accuracy and stability of DDI prediction can be significantly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of drug interaction prediction, and particularly relates to a drug interaction prediction method based on an interactive enhanced graph self-attention mechanism. Background Art

[0002] With the continuous development of modern medicine, for patients suffering from multiple complications or intractable diseases, combined drug use has become a quite common treatment mode in clinical practice. However, the combined intake of multiple drugs may cause them to interact with each other in the patient's body, thus triggering the phenomenon of drug-drug interaction (DDI). DDI may lead to weakened drug efficacy, increased toxicity or other adverse reactions, and even threaten life in severe cases. Therefore, studying and predicting drug interactions not only has important significance for improving the safety of drug treatment, but also provides a scientific basis for personalized medicine and precision medicine.

[0003] Traditional DDI identification mainly relies on limited sources such as professional databases, clinical trial reports and existing literature. Researchers and clinicians avoid common interaction risks by retrieving existing cases or using empirical knowledge. However, due to the large variety of drugs, complex medication scenarios, and the continuous emergence of new drugs or new medication combinations, it is difficult to effectively predict potential unknown interactions only relying on traditional methods, which cannot meet the current clinical needs. In recent years, with the rapid development of computational methods, algorithm-based DDI prediction has gradually become a powerful tool. Compared with traditional experimental methods, computational methods can process large amounts of data in a shorter time and do not require costly and time-consuming experimental operations, thus providing a more efficient and accurate solution for DDI prediction and prevention.

[0004] At present, computational DDI identification methods can be roughly divided into indirect methods (based on drug association) and direct methods (based on drug structure). Indirect methods mainly rely on building associations between drugs, often representing drugs and their interactions through heterogeneous networks or knowledge graphs. Drugs and their related biomarkers, targets, pathways and other entities are regarded as nodes in the network, and the relationships between them are represented by edges. By learning the potential association information between drugs, these methods can effectively capture the interactions between drugs and thus be used for DDI prediction. However, without stable data and financial support, indirect DDI prediction methods cannot obtain effective data quality assurance. Direct methods focus on encoding the molecular structure information of drugs, and believe that the occurrence of drug interactions is mainly driven by the interaction between substructures or functional groups within the molecule. Therefore, through in-depth analysis of drug structures, their potential DDI characteristics can be revealed. Graph neural networks (GNNs), as a deep learning method for processing graph structures, have achieved remarkable results in many fields in recent years, especially showing strong modeling capabilities in drug molecular graphs. Nyamabo AK et al. used the graph attention network (GAT) to extract substructure information at different levels in drugs, and then used the co-attention mechanism to measure the importance of interactions between substructures for DDI prediction. Yang Z et al. proposed D-MPNN, which assigns different weights to substructures of different radii during message passing, and obtains size-adaptive drug substructures for subsequent DDI prediction. Transformer was originally designed to process sequence data, and has achieved great success in natural language processing (NLP) tasks. Its self-attention mechanism can capture long-distance dependencies between elements in a sequence, making it perform well in tasks such as translation and text generation. By adaptively adjusting the self-attention mechanism, the graph transformer can effectively process complex relationships between nodes and structural information of the graph, and has gradually become an important tool in graph data analysis. Compared with traditional GNNs, graph transformers show greater flexibility and efficiency in processing large-scale graph data and long-distance dependencies. Niu D et al. proposed a substructure refined mechanism based on an improved graph transformer to update the substructure of drugs and extract the most accurate substructure that determines the drug properties. In addition, DAS-DDI proposed a DS encoder based on the graph transformer module to achieve synchronous update of nodes and keys to extract the substructure of drugs, demonstrating the feasibility of graph transformers in graph structured data.

[0005] However, with the continuous advancement of the DDI prediction task, existing methods based on GNN and graph Transformers have several significant limitations when exploring the interactions between drug pairs. Existing drug encoders limit their perspective to a single drug and only perform feature propagation between nodes within the graph structure of a single drug through self-attention mechanisms, which restricts the exploration of interactions between drug pairs. In addition, although many methods extract drug information by designing multiple drug views, most of these methods simply concatenate different views together and fail to effectively utilize the potential synergistic effects between different views, resulting in a significant reduction in the effect of multi-view information fusion. This not only leads to insufficient information utilization but also inevitably results in a certain degree of information redundancy. Moreover, in the pairwise interaction prediction task of drug pairs, existing methods generally assist prediction by calculating the interaction relationships between substructures or functional groups between drug pairs. Although this can capture certain local structural information, its limitation lies in the insufficient dynamic modeling of interactions, ignoring that the interaction intensity between different substructures or functional groups in a specific drug pair is dynamically changing. Summary of the Invention

[0006] In view of the above problems existing in the prior art, the present invention proposes a drug interaction prediction method based on an interaction-enhanced graph self-attention mechanism, which solves the deficiencies of the prior art.

[0007] The drug interaction prediction method based on an interaction-enhanced graph self-attention mechanism includes the following steps: S1. Model the drug as a molecular graph and a motif graph respectively under the local view and the global view; S2. Design a graph Transformer module for enhancing interactions, update the node embeddings respectively through the interaction-enhanced attention mechanism in the two views of the molecular graph, and fuse them through a multi-channel adaptive fusion module; S3. Obtain the final molecular representations of the molecular graph and the motif graph by passing the extracted features through a feed-forward neural network, a residual connection layer, and a normalization layer; S4. Design a dynamic interaction scaling prediction module to calculate the DDI score; S5. Construct a drug interaction prediction model with the graph Transformer module, the multi-channel adaptive fusion module, the feed-forward neural network, the residual connection layer, the normalization layer, and the dynamic interaction scaling prediction module, design a loss function, and optimize the model parameters.

[0008] Further, in S1, in the local view, each atom in the drug molecule is defined as a node, and the chemical bond is used as the edge connecting the nodes. A one-hot is used to construct a feature vector to describe the atomic symbol and chemical properties, and the chemical bond also uses a one-hot to construct a feature vector to characterize the bond type and conjugation information; use the local molecular graph Characterize the drug in the local view is the node representation of the drug in the local view is the representation of the chemical bonds of the drug in the local view is the adjacency information between nodes of the drug in the local view; In the global view, a molecular graph is constructed with atoms as nodes and chemical bonds as edges. A relative distance matrix is introduced to describe the global positional relationship between atoms. The element values in the relative distance matrix are determined by calculating the shortest path between nodes. This value reflects the relative distance between atoms in the molecular structure. Use the global molecular graph Characterize the drug in the global view is the node representation of the drug in the global view is the representation of the chemical bonds of the drug in the global view is the adjacency information between nodes of the drug in the global view represents the relative distance matrix between each node in the drug; Through a chemistry-inspired molecular fragmentation function, the complex molecular structure is decomposed into multiple chemically meaningful motifs. A vocabulary containing unique motifs is constructed using the ChEMBL29 database. Then, the local molecular graph and the global molecular graph are respectively converted into a local motif graph and a global motif graph with motifs as nodes and shared atom connections as edges.

[0009] Furthermore, in S2, the drug pair is divided into a head drug and a tail drug. For the head drug in the molecular graph, in the local view, the initial embeddings of the head drug nodes are respectively mapped to the query space and the value space, and the initial embeddings of the tail drug nodes are mapped to the key space. Then, the interaction strength between the drug pair nodes is calculated and the features of the nodes are updated. The expression is: ;(1) ;(2) ;(3) ;(4) Among them, , and are both learnable weight matrices, is the node embedding matrix of the head drug in the local view at the th layer. When = 0, it is the initial embedding of the head drug, is the embedding vector of the +1th node of the head drug in the local view at the th layer; is the softmax activation function, is the dimension of, is the number of neighbors of the th node in the head drug, is the number of attention heads in interactive enhanced attention; For the tail drug at the layer, it is the node embedding matrix under the local view, , , are respectively the query matrix of the th node in the local view of the head drug, the key matrix of the th node in the local view of the tail drug, and the value matrix of the th node in the local view of the head drug; In the global view, the initial embeddings of the head drug nodes are respectively mapped to the query space and the value space, the initial embeddings of the tail drug nodes are mapped to the key space, and at the same time, the relative distance matrix between the nodes in the head drug is constructed ; Given the node set of the head drug, for any two nodes and , calculate their relative distance in the head drug molecular graph, which is defined as the shortest path between the th node and the th node, and calculate the interactive attention between the drug pair nodes and update the node features through the following formula: ; (5) ; (6) ; (7) ; (8) where, , and are learnable weight matrices, is the number of nodes in the head drug, is the dimension of, is a hyperparameter, is the node embedding matrix of the head drug under the global view at the l-th layer, is the node embedding matrix of the tail drug under the global view at the th layer, is the embedding vector of the +1-th node of the head and tail drugs under the global view at the th layer, , , are respectively for the head drug at the The query matrix of the th node in the global view, the key matrix of the th node in the global view, and the value matrix of the

[0010] th node in the global view. ; (9) ; (10) where is a learnable multi-channel operator for storing the attention map of the interaction under the two views, is the normalization operation on the embedding, and are both learnable weight matrices, represents the matrix multiplication operation, represents the element-wise multiplication operation, C is the number of channels in is the bias vector, is the attention map between the th node and the th node in the c-th channel, is the fused feature of the th node of the head drug after passing through the multi-channel adaptive fusion module.

[0011] Furthermore, in the step S3, the extracted features are passed through a feed-forward neural network, a residual connection layer, and a normalization layer to obtain the final molecular representation of the head drug, and the expression is: ; (11) ; (12) ; (13) where is a learnable weight matrix, is the GELU activation function, is the layer normalization operation, is the th node embedding of the head drug output by the th layer, is the +1 th node embedding of the head drug output by the th layer; The +1 th node embedding of the tail drug output by the th layer is , the tail drug molecule is denoted as , is the number of tail drug nodes; Embed the obtained molecular graph into and respectively update the initial embeddings of the motif graphs of the head drug and the initial embeddings of the motif graphs of the tail drug , the expression is: ; (14) ; (15) where is the feature fusion function, is the initial embedding of the motif graph after the head drug is updated, is the initial embedding of the motif graph after the tail drug is updated; and are used as the input of the interaction-enhanced graph Transformer module again to obtain the final hidden embeddings of the motif graphs of the head drug and the tail drug and .

[0012] Furthermore, in S4, first, use a co-attention-based interaction layer to capture the interaction intensity between drugs, and generate a drug representation with context information by weighted fusion of the features of the drug pair , the expression is: ; (16) where and respectively represent the first elements corresponding to the final hidden embeddings of the head drug and the tail drug; Next, calculate the interaction information between drug pairs, the expression is: ; (17) ; (18) where and respectively represent the remaining parts of the final hidden embeddings of the head drug and the tail drug except the first element, is the interaction information of the head drug, is the interaction information of the tail drug, is the trainable weight matrix; In order to dynamically adjust the interaction intensity between drug pairs, the Gumbel-Softmax distribution is used, and the expression is: ; (19) ;(20) ;(21) Among them, is the noise sampled from the uniform distribution , is the probability distribution matrix sampled from the head drug Gumbel-Softmax distribution of the th element, indicating the relative preference for the th functional group classification category in the head drug, is the temperature parameter, is the number of functional groups; Finally, the features of the drug pair are aggregated and input into a multi-layer perceptron for DDI score calculation. The expression is: ;(22) ;(23) ;(24) Among them, is the probability distribution matrix sampled from the tail drug from the Gumbel-Softmax distribution, is the prediction activation function, represents the vector concatenation operation, is the trainable weight matrix, is the DDI prediction probability.

[0013] Furthermore, in the S5, by collecting various types of drug pairs in DrugBank, the drug interaction prediction model is tested using multi-classification tasks and binary classification tasks. The multi-classification task is to identify which DDI reaction the drug pair will have, and the binary classification task is to identify whether there is a DDI reaction between the drug pair; For the multi-classification task, the following cross-entropy loss function is used to optimize the model parameters: ;(25) Among them, Z is the number of samples, is the number of classes, is the true label of the th sample on the th class, is the prediction probability of the model for the th sample in the th class; For the binary classification task, binary cross-entropy is used as the loss function: ;(26) Among them, is the true label of the th sample, represents the class prediction probability of the th sample. When , it means that the model predicts the th sample as class 1, otherwise 0.

[0014] Beneficial technical effects brought by the present invention: The present invention combines an interactive enhanced graph Transformer module, a multi-channel adaptive fusion module, and an annealing algorithm-driven dynamic interaction scaling prediction module, successfully breaking through the limitations of the traditional single-drug perspective, while fully integrating multi-perspective drug features, and achieving accurate modeling of the interaction relationship between drug pairs; during the prediction process, the Gumbel-Softmax distribution and the annealing concept are introduced to dynamically adjust the interaction strength, thereby significantly improving the accuracy and stability of DDI prediction.

[0015] (1) By designing an interactive enhanced attention mechanism in the graph Transformer module, the limitation of the traditional single-drug perspective is broken through, and the interaction between drug pairs can be explored more deeply. This mechanism models the interaction of drug pairs in the graph structure, thereby effectively improving the ability to capture the drug interaction relationship.

[0016] (2) Aiming at the problem of fusing multi-view information of drugs, a module capable of interactive learning from different perspectives is proposed, which effectively integrates and utilizes the features and potential associations of drugs in each view.

[0017] (3) To overcome the problem of insufficient interaction modeling in existing methods, the annealing idea and the Gumbel-Softmax distribution are introduced to explore and scale the interaction force between drug pairs, and the interaction strength is automatically adjusted during the training process. Description of the Drawings

[0018] Figure 1 is the local view of the drug constructed in the present invention.

[0019] Figure 2 is the global view of the drug constructed in the present invention.

[0020] Figure 3 is the workflow diagram of the interactive enhanced graph Transformer module in the present invention.

[0021] Figure 4 is the flowchart of the multi-channel adaptive fusion module in the present invention.

[0022] Figure 5This is the flowchart of the dynamic interaction scaling prediction module in the present invention.

[0023] Figure 6 This is the prediction result diagram of the drug interaction of two groups of drug pairs in the present invention. Detailed implementation manners

[0024] The following further describes the detailed implementation manners of the present invention in combination with specific embodiments: A drug interaction prediction method based on an interactive enhanced graph self-attention mechanism includes the following steps: S1. Model the drugs as molecular graphs and motif graphs respectively under the local view and the global view; For the modeling of drugs, the present invention considers two representation methods of drugs. One is the drug based on the molecular graph (collectively referred to as the molecular graph), and the other is the drug based on the motif (that is, the functional groups that repeatedly appear in the drug) (collectively referred to as the motif graph). And based on these two methods, the local view and the global view of the drug are respectively constructed, as Figure 1 and Figure 2 shown.

[0025] At the microscopic level, that is, the local view, specific functional groups serve as the carriers of local information, and their chemical properties and spatial structures directly determine the affinity and specific binding ability between drug molecules and targets. Taking the molecular graph as an example, in the local view, each atom in the drug molecule is defined as a node, and the chemical bond is used as the edge connecting the nodes. The one-hot is used to construct the feature vector to describe the atomic symbol and chemical properties, and the chemical bond also uses the one-hot to construct the feature vector to characterize the bond type and conjugation information; the local molecular graph represents the drug in the local view, is the node representation of the drug in the local view, is the representation of the chemical bond of the drug in the local view, is the adjacency information between nodes of the drug in the local view; At the macroscopic level, that is, the global view, the three-dimensional conformation, size scale of the whole molecule, and the interaction between long-distance regions cannot be ignored, and play a decisive role in regulating the biological activity, stability of the molecule, and the synergistic or antagonistic effects between drug molecules. In the global view, a molecular graph is also constructed with atoms as nodes and chemical bonds as edges, but a relative distance matrix is introduced to describe the global position relationship between atoms. For a given node set, the element values in the relative distance matrix are determined by calculating the shortest path between nodes, and this value reflects the relative position distance between atoms in the molecular structure. The global molecular graph represents the drug in the global view, is the node representation of the drug in the global view, It is the representation of chemical bonds of drugs in the global view, It is the adjacency information between nodes of drugs in the global view, It represents the relative distance matrix between each node in the drug; Secondly, the molecular graph provides high-resolution information for understanding the chemical environment of molecules. However, the complexity of drug molecules is not only reflected in the connections at the atomic level, but more importantly in the combination of substructures with specific chemical meanings. Therefore, in addition to the molecular graph, the motif graph is also introduced into the molecular representation process. Through a chemically inspired molecular fragmentation function, the complex molecular structure is decomposed into multiple motifs with chemical meanings. A vocabulary containing unique motifs is constructed using the ChEMBL29 database, and then the local molecular graph and the global molecular graph are respectively converted into a local motif graph and a global motif graph with motifs as nodes and shared atomic connections as edges, which can effectively capture key structural information such as functional groups in the molecule, thereby representing the molecular structure at a higher semantic level.

[0026] S2. Design a graph Transformer (self-attention mechanism) module with enhanced interaction, and update the node embeddings respectively through the interaction-enhanced attention mechanism in two views of the molecular graph, and fuse them through a multi-channel adaptive fusion module; S3. Since the problem of gradient vanishing or gradient explosion may occur in deep networks, and at the same time, in order to eliminate the feature distribution differences between different layers, avoid biases during the model training process, and improve the stability of the training process, the extracted features are passed through a feed-forward neural network, a residual connection layer, and a normalization layer to obtain the final molecular representations of the molecular graph and the motif graph; The above graph Transformer module, multi-channel adaptive fusion module, feed-forward neural network, residual connection layer, and normalization layer have L layers. The following takes the layer as an example for calculation; Traditional drug interaction prediction methods based on graph model modeling are limited to a single drug space. In order to accurately capture the interaction between drug pairs and enhance the interaction modeling ability, the present invention designs a graph Transformer module with enhanced interaction, as Figure 3 shown; Taking the molecular graph as an example, the drug pair is divided into a head drug and a tail drug. The node embeddings are updated respectively through the interaction-enhanced attention mechanism in two views, and fused through a multi-channel adaptive fusion module. Taking the head drug as an example, first in the local view, the initial embeddings of the head drug nodes are respectively mapped to the query space and the value space, and the initial embeddings of the tail drug nodes are mapped to the key space, and then the interaction intensity between the drug pair nodes is calculated and the features of the nodes are updated. The expression is: ;(1) ;(2) ; (3) ; (4) Among them, , and are all learnable weight matrices, is the node embedding matrix of the head drug in the local view at the -th layer. When = 0, it is the initial embedding of the head drug, is the embedding vector of the +1-th node of the head drug in the local view at the -th layer; is the softmax activation function, is 's dimension, is the number of neighbors of the -th node in the head drug. Neighbors are adjacent nodes, is the number of attention heads in the interaction-enhanced attention; is the node embedding matrix of the tail drug in the local view at the -th layer, , , are respectively the query matrix of the -th node of the head drug in the local view, the key matrix of the -th node of the tail drug in the local view, and the value matrix of the -th node of the head drug in the local view; In the global view, the initial embedding of the head drug nodes is mapped to the query space and the value space respectively, and the initial embedding of the tail drug nodes is mapped to the key space. At the same time, the relative distance matrix between the nodes of the head drug is constructed; Given the node set of the head drug, for any two nodes and , calculate their relative distance in the head drug molecular graph, which is defined as the shortest path between the -th node and the -th node. For example, if the shortest path from the node to passes through , and the edge distances connecting and , and are both 1, then . Calculate the interaction attention between the drug pair nodes and update the node features through the following formula: ; (5) ; (6) ; (7) ; (8) Among them, , and are learnable weight matrices, is the number of nodes of the head drug, is 's dimension, is a hyperparameter, is the node embedding matrix of the head drug under the global view in the th layer, is the node embedding matrix of the tail drug under the global view in the th layer, is the embedding vector of the + 1th node of the head drug under the global view in the th position, , , are respectively the query matrix of the th node of the head drug in the global view, the key matrix of the th node of the tail drug in the global view, and the value matrix of the th node of the head drug in the global view.

[0027] Taking the tail drug as an example, obtain the embedding vector of the + 1th node of the tail drug under the local view in the th position , and the embedding vector of the + 1th node of the tail drug under the global view in the th position .

[0028] In S2, a multi-channel adaptive fusion module is designed to fuse the local view and the global view corresponding to the molecular graph respectively. As Figure 4 shown, the expression is: ; (9) ; (10) Among them, is a learnable multi-channel operator for storing the attention mapping of the interaction under the two views, is the normalization operation on the embedding, and are both learnable weight matrices, represents the matrix multiplication operation, Represents the element-wise product operation, where C is the number of channels in the bias vector, is the th node and the th attention map between the th

[0029] The fused feature of the th

[0030] node of the head drug after passing through the multi-channel adaptive fusion module. The fused feature is obtained by fusing the local view and the global view corresponding to the molecular graph for the tail drug.

[0031] In S3, the extracted passes through a feed-forward neural network, a residual connection layer, and a normalization layer to obtain the final molecular representation , and the expression is: ; (11) ; (12) ; (13) where is a learnable weight matrix, is the GELU activation function, is the layer normalization operation, is the th node embedding of the head drug output by the th +1 layer, is the The extracted passes through a feed-forward neural network, a residual connection, and a normalization to obtain the final molecular representation , , is the The th node embedding of the tail drug output by the +1 layer, where is the number of tail drug nodes.

[0032] Embed the obtained molecular graph and respectively update the initial embedding of the motif graph of the head drug and the initial embedding of the motif graph of the tail drug , and the expression is: ; (14) ; (15) where, is the feature fusion function, is the initial embedding of the motif graph after the head drug is updated, is the initial embedding of the motif graph after the tail drug is updated; and are used as the input of the interactive enhanced graph Transformer module again, and repeat the above process, that is, formulas (1)-(12), to obtain the final hidden embeddings and of the motif graphs of the head drug and the tail drug.

[0033] S4. Design a dynamic interaction scaling prediction module to calculate the DDI score; Aiming at the deficiencies of the existing DDI prediction methods in interaction modeling, the present invention designs a dynamic interaction scaling prediction module, which explores and scales the interaction force between drug pairs by introducing the concept of annealing and the Gumbel-Softmax distribution, and automatically adjusts the interaction strength between drug pairs during the training process, as Figure 5 shown. Specifically, the interaction scaling factor is dynamically adjusted by the annealing algorithm, so that the interaction is stronger in the initial stage of training and gradually weakens in the later stage, avoiding overfitting.

[0034] First, use a co-attention-based interaction layer to capture the interaction strength between drugs, and generate a drug representation with context information by weighted fusion of the features of the drug pair , and the expression is: ; (16) where, and respectively represent the first elements corresponding to the final hidden embeddings of the head drug and the tail drug; Next, calculate the interaction information between drug pairs, and the expression is: ; (17) ;(18) Among them, and respectively represent the remaining parts of the final hidden embeddings of the head drug and the tail drug except the first element, is the interaction information of the head drug, is the interaction information of the tail drug, is the trainable weight matrix; Taking the head drug as an example, in order to dynamically adjust the interaction strength between drug pairs, the Gumbel-Softmax distribution is used, and the expression is: ;(19) ;(20) ;(21) Among them, is the noise sampled from the uniform distribution to introduce randomness, so that the shape of the probability distribution is different for each sampling. The noise will perturb the obtained interaction strength, making the sample selection process exploratory rather than simply choosing the maximum value (greedy strategy). This makes Gumbel-Softmax a smooth approximation to the discrete sampling mechanism, is the probability distribution matrix sampled by the head drug from the Gumbel-Softmax distribution of the th element, represents the relative preference of the th functional group classification category in the head drug, that is , combines the sampling noise with to generate a new score, is the temperature parameter, which is used to control the smoothness of this process. When the temperature is high, the distribution becomes smoother; when the temperature is low, the distribution becomes sharper, approaching a hard decision (i.e., approaching the selection of the maximum value). At the beginning of model training, will hold a high temperature, making the probability distribution tend to be uniform and the sampling result more random. As the model training process progresses, the temperature is gradually reduced, which means annealing, making the probability distribution tend to a single extreme value, so that a certain degree of randomness and exploration can be maintained during the sampling process, is the exponential function, which converts the new score into a positive value to form the original form of the probability distribution, is the number of functional groups; Taking the tail drug as an example, the probability distribution matrix sampled by the tail drug from the Gumbel-Softmax distribution is .

[0035] Finally, the features of the drug pairs are aggregated and input into a multi-layer perceptron for DDI score calculation. The expression is: ;(22) ;(23) ;(24) where is the prediction activation function. Since two classification tasks are designed, when facing a multi-classification task, is set to the Softmax activation function, and when facing a binary classification task, is set to the Sigmoid activation function. represents the vector concatenation operation, is the trainable weight matrix, is the DDI prediction probability.

[0036] S5. Construct a drug interaction prediction model with the L-layer graph Transformer module, multi-channel adaptive fusion module, feed-forward neural network, residual connection layer, normalization layer, and dynamic interaction scaling prediction module, design a loss function, and optimize the model parameters.

[0037] DrugBank is a classic drug database containing rich and reliable drug information. By collecting various types of drug pairs in DrugBank, the drug interaction prediction model is tested using multi-classification tasks and binary classification tasks. The multi-classification task is to identify which DDI reaction the drug pair will have, and the binary classification task is to identify whether there is a DDI reaction between the drug pair; For the multi-classification task, the following cross-entropy loss function is used to optimize the model parameters: ;(25) where Z is the number of samples, is the number of classes, is the true label of the th sample on the th class, is the prediction probability of the model for the th sample in the th class; For the binary classification task, binary cross-entropy is used as the loss function: ;(26) where is the true label of the th sample, Indicates the class prediction probability of the th sample. When it indicates that the model predicts the th sample as class 1, otherwise 0.

[0038] The comparison results of the model (AMIE-DDI) proposed by the present invention with other advanced models (MR-GNN, SA-DDI, DSN-DDI, PEB-DDI, MeTDDI) in two datasets. ZhongDDI is a multi-class dataset, and ZhangDDI is a binary-class dataset. ACC, AUROC, and AUPR are used as the evaluation metrics of the model, as shown in Table 1. The model proposed by the present invention has achieved performance improvement compared with the current advanced models in both classification scenarios.

[0039] Table 1 Comparison of evaluation metrics between the model of the present invention and other advanced models ; As Figure 6 shown, among the drug pairs with label 0 composed of DB01558 and DB00277, the model proposed by the present invention marked the pyridine ring (C1CN=CccN1) and alkyl group (Cc) in DB01558, and these two motifs are closely related to the binding site of the metabolic enzyme. For DB00277, the model marked the pyrimidine ring (c1cncnc1) and methylamine (Cn). The pyrimidine ring usually competitively occupies the active site of the metabolic enzyme through coordination, inhibiting the metabolic activity of other drugs.

[0040] In the drug pair with label 1 composed of DB06266 and DB00454, the aminomethyl group (Cn) marked by the model proposed by the present invention is the key functional group of DB06266 and is often used as the recognition and binding site of metabolic enzymes (such as the CYP450 family). The methylamine group (CN), as an electron-donating group, is prone to form a bond with the metabolic enzyme. DB00454 may bind to the active site of the metabolic enzyme (such as CYP450) through its methylamine group. In medicinal chemistry, compounds containing methylamine groups often undergo phase I reactions such as N-oxidation during metabolism, indicating that this group is prone to interact with metabolic enzymes.

[0041] The above shows that the model proposed by the present invention can identify the key functional groups in the drug interaction process, improving the model interpretability in the drug interaction process.

[0042] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions, or substitutions made by those skilled in the art within the scope of the essence of the present invention should also fall within the protection scope of the present invention.

Claims

1. A drug interaction prediction method based on interactive enhanced graph self-attention mechanism, characterized in that: The following steps are involved: Drug interaction prediction method based on interaction-enhanced graph self-attention mechanism S1, modeling drugs as molecular graphs and motif graphs in local and global views, respectively; S2, design an interaction-enhanced graph Transformer module, which updates node embeddings separately in two views of the molecular graph through the interaction-enhanced attention mechanism and fuses them through a multi-channel adaptive fusion module; S3, the extracted features are passed through a feedforward neural network, a residual connection layer and a normalization layer to obtain the molecular representation of the final molecular graph and motif graph; S4, design a dynamic interaction scaling prediction module to calculate the DDI score; S5. The graph Transformer module, multi-channel adaptive fusion module, feedforward neural network, residual connection layer and normalization layer and dynamic interaction scaling prediction module are combined to construct a drug interaction prediction model, design the loss function and optimize the model parameters.

2. The drug interaction prediction method based on interactive enhanced graph self-attention mechanism according to claim 1, characterized in that: In S1, in the local view, each atom in the drug molecule is defined as a node, and the chemical bond is used as the edge connecting the nodes. The feature vector is constructed using one-hot to describe the atomic symbol and chemical properties. The chemical bond also uses one-hot to construct the feature vector to represent the bond type and conjugation information; using the local molecular graph Characterize the drug in a local view, is the node representation of the drug in the local view, is a representation of the chemical bonds of the drug in a local view, It is the adjacency information between nodes of the drug in the local view; In the global view, a molecular graph is constructed with atoms as nodes and chemical bonds as edges. A relative distance matrix is ​​introduced to describe the global position relationship between atoms. The element value in the relative distance matrix is ​​determined by calculating the shortest path between nodes. This value reflects the relative position of atoms in the molecular structure. Characterize drugs in a global view, is the node representation of the drug in the global view, is the representation of the chemical bonds of the drug in the global view, is the adjacency information between nodes of the drug in the global view, Represents the relative distance matrix between nodes in the drug; Through a chemically inspired molecular fragmentation function, the complex molecular structure is decomposed into multiple chemically meaningful motifs. A vocabulary containing unique motifs is constructed using the ChEMBL29 database, and then the local molecular graph and the global molecular graph are converted into local motif graph and global motif graph respectively with motifs as nodes and shared atomic connections as edges.

3. The drug interaction prediction method based on interactive enhanced graph self-attention mechanism according to claim 2, characterized in that: In S2, drug pairs are divided into head drugs and tail drugs. For the head drug in the molecular graph, in the local view, the initial embedding of the head drug node is mapped to the query space and the value space respectively, and the initial embedding of the tail drug node is mapped to the key space. Then, the interaction strength between the drug pair nodes is calculated and the node features are updated. The expression is: ；(1) ；(2) ；(3) ； (4) in, , and are all learnable weight matrices, It is the first drug in The node embedding matrix in the local view of the layer is = 0 is the initial insertion of the drug, It is the first drug in +1 layer in the local view The embedding vector of each node; is the softmax activation function, yes The dimension of It is the first drug The number of neighbors of a node, is the number of attention heads in interactive enhanced attention; For the last drug The node embedding matrix under the local view in the layer, , , The first The query matrix of the node in the local view, the tail drug The key matrix of the node in the local view and the first drug in the head The value matrix of nodes in the local view; In the global view, the initial embedding of the head drug node is mapped to the query space and the value space respectively, and the initial embedding of the tail drug node is mapped to the key space. At the same time, the relative distance matrix between the nodes in the head drug is constructed. ; A set of nodes for a given head drug , for any two nodes and , calculate their relative distances in the head drug molecule graph , defined as Nodes and The shortest path between nodes is used to calculate the interactive attention between the drug pairs and update the features of the nodes using the following formula: ；(5) ；(6) ；(7) ；(8) in, , and is a learnable weight matrix, is the number of nodes of the head drug, yes The dimension of is a hyperparameter, is the node embedding matrix of the head drug in the global view in layer l, For the last drug The node embedding matrix in the global view of the layer, The first and last drugs are +1 layer in the global view The embedding vector of each node, , , The first The query matrix of the node in the global view, the tail drug The key matrix of the node in the global view and the head drug The value matrix of each node in the global view.

4. The drug interaction prediction method based on interactive enhanced graph self-attention mechanism according to claim 3, characterized in that: In S2, a multi-channel adaptive fusion module is designed to fuse the local view and the global view corresponding to the molecular graph, and the expression is: ；(9) ；(10) in, is a learnable multi-channel operator that stores the attention map of the interaction between two views. is the normalization operation on the embedding, and are all learnable weight matrices, represents the matrix multiplication operation, represents the element-wise multiplication operation, C is The number of channels in is the bias vector, is the number of channels in the cth Nodes and The attention map between nodes, After the drug passes through the multi-channel adaptive fusion module, The fusion features of the nodes.

5. The drug interaction prediction method based on interactive enhanced graph self-attention mechanism according to claim 4, characterized in that: In S3, the extracted features are passed through a feedforward neural network, a residual connection layer, and a normalization layer to obtain the final head drug molecule representation. , the expression is: ；(11) ；(12) ；(13) in, is a learnable weight matrix, is the GELU activation function, is the layer normalization operation, For the The first drug of the layer output Node embedding, For the +1 layer output of the first drug Node embedding; No. +1 layer of output of the last drug The nodes are embedded as , the tail drug molecule is represented by , is the number of tail drug nodes; Embed the obtained molecular graph into and Update the initial embedding of the motif graph of the head drug separately Initial embedding of the motif graph of the and tail drugs , the expression is: ；(14) ；(15) in, is the feature fusion function, is the initial embedding of the motif graph after the head drug update, Initial embedding of the motif graph after updating for the tail drug; and Re-use it as the input of the interactive enhancement graph Transformer module to obtain the motif graph of the head drug and the tail drug and finally the hidden embedding and .

6. The drug interaction prediction method based on interactive enhanced graph self-attention mechanism according to claim 5, characterized in that: In S4, first, a co-attention-based interaction layer is used to capture the interaction strength between drugs, and drug representations with contextual information are generated by weighted fusion of drug pair features. , the expression is: ；(16) in, and Represent the first element of the final hidden embedding corresponding to the head drug and the tail drug respectively; Next, the interaction information between drug pairs is calculated, and the expression is: ；(17) ；(18) in, and denote the remainder of the final hidden embeddings of the head drug and the tail drug, except the first element, respectively. For the interaction information of the head drug, is the interaction information of the tail drug, is a trainable weight matrix; In order to dynamically adjust the interaction strength between drug pairs, the Gumbel-Softmax distribution is used, which is expressed as: ；(19) ；(20) ；(21) in, From a uniform distribution The noise sampled in is the probability distribution matrix sampled from the Gumbel-Softmax distribution of the initial drug No. elements, Indicates the first drug The relative preference of the functional group classification categories, is the temperature parameter, is the number of functional groups; Finally, the features of the drug pairs are aggregated and input into a multi-layer perceptron for DDI score calculation, expressed as: ；(22) ；(23) ；(24) in, is the probability distribution matrix of the tail drug sampled from the Gumbel-Softmax distribution, is the prediction activation function, represents the vector concatenation operation, is the trainable weight matrix, Predict the probability of DDI.

7. The drug interaction prediction method based on interactive enhanced graph self-attention mechanism according to claim 6, characterized in that: In S5, by collecting various types of drug pairs in DrugBank, a multi-classification task and a binary classification task are used to test the drug interaction prediction model. The multi-classification task is to identify which type of DDI reaction will occur in the drug pairs, and the binary classification task is to identify whether there is a DDI reaction between the drug pairs. For multi-classification tasks, the following cross entropy loss function is used To optimize the model parameters: ；(25) Where Z is the sample size, is the number of categories, It is The sample in The true labels on the categories, The model is The sample in The predicted probability of being in each category; For binary classification tasks, use binary cross entropy As a loss function: ；(26) in, It is The true labels of samples, Indicates The class prediction probability of samples is When the model predicts samples are of class 1, otherwise they are 0.

Images