A drug-drug interaction prediction method and device based on distribution matching

By adopting the distribution matching method in drug-drug interaction prediction, combining molecular structure and topological relationship characteristics, the problems of low prediction accuracy and computational efficiency in the existing technology are solved, and efficient drug interaction prediction is achieved, which significantly improves the Model performance and robustness.

CN118711701BActive Publication Date: 2024-11-19ZHEJIANG LAB
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411215460.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-02
Publication Date
2024-11-19
Estimated Expiration
2044-09-02

AI Technical Summary

Technical Problem

The existing drug-drug interaction prediction methods have high experimental costs, long time, limited coverage and single information perspective problems, and the GNN model has challenges in computing efficiency and multi-view information fusion.

Method used

Using a distribution matching method, the molecular structural characteristics and topological relationship characteristics of the drug are extracted, and distribution matching is performed in the feature space and output space to maximize their mutual information to predict drug-drug interactions, and design a central matching distribution based on central matching The loss function of the equilibrium embedding information from different perspectives.

Benefits of technology

It significantly improves prediction accuracy and computing efficiency, reduces calculation overhead, maintains high prediction accuracy in the case of scarcity of data, adapts to complex changes in drug data, and provides reliable prediction results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118711701B_ABST
    Figure CN118711701B_ABST
Patent Text Reader

Abstract

The present invention discloses a drug-drug interaction prediction method and device based on distribution matching, including: extracting the molecular structure characteristics of the drug; based on the molecular structure characteristics, further extracting the topological relationship characteristics between the drugs; in the feature space and the output space, respectively, the molecular structure characteristics and the topological relationship characteristics are distributed and matched to maximize their mutual information to predict DDI. The present invention not only considers the molecular structure information of the drug, but also integrates the topological relationship information between the drugs. This integration of multi-perspective information comprehensively captures the complex interactions between drugs, thereby significantly improving the accuracy of the prediction, and adopts a distribution matching strategy to maximize the mutual information between the molecular structure characteristics and the topological relationship characteristics, ensuring the integrity and accuracy of the information, and further improving the prediction performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of molecular interaction prediction, and in particular to a method and device for predicting drug-drug interaction based on distribution matching. Background Art

[0002] Drug-Drug Interactions (DDIs) refer to the complex interactions that may occur between drugs when two or more drugs coexist in the patient's body. This interaction may lead to changes in drug efficacy or induce adverse reactions, and in severe cases may even threaten the patient's life safety. The risk of DDIs is particularly important in clinical treatment, especially in elderly patients, patients with multiple co-morbidities, or patients who require long-term multi-drug treatment. The incidence of DDIs is higher and the potential hazards are more prominent. Therefore, accurate prediction and identification of drug interactions are of vital importance in drug development, clinical treatment, and drug safety management.

[0003] With the development of modern medicine and the need for complex disease management, patients often need to take multiple drugs at the same time (i.e., polydrug therapy), which increases the probability of DDIs. At the same time, the diversity of drug combinations and the complexity of pharmacological mechanisms also make DDIs prediction more challenging. Failure to identify potential DDIs in a timely manner may lead to reduced drug treatment efficacy or induce unforeseen drug toxicity reactions, thereby posing a threat to the patient's health. Therefore, in drug development and clinical practice, predicting and preventing DDIs has become a critical task.

[0004] Traditional methods for detecting drug interactions mainly rely on in vitro experiments, animal experiments, and clinical trials. These methods usually have high accuracy and reliability by directly observing the interactions of drugs under specific experimental conditions. However, these methods also have significant limitations:

[0005] 1. High experimental costs: Experimental testing of drug interactions often requires the use of expensive reagents, instruments and equipment, as well as the input of professionals, resulting in high overall experimental costs. This poses a significant economic burden on small and medium-sized pharmaceutical companies and research institutions.

[0006] 2. Time-consuming: The experimental process is usually very cumbersome, and it may take several months or even years from preliminary screening to final clinical verification. For some new drugs that are urgently needed to be launched, the long cycle of traditional methods will undoubtedly slow down their market launch.

[0007] 3. Limited coverage: Due to the limitation of experimental resources and time, traditional methods cannot cover all possible drug combinations. Therefore, some potential DDIs are not discovered in time, which increases the potential risk of drug use for patients.

[0008] 4. Limitations of experimental conditions: There are differences between laboratory conditions and the actual environment in the human body. Some drug interactions may be difficult to simulate or observe in a laboratory environment, resulting in limited extrapolation of predicted results.

[0009] In recent years, Graph Neural Networks (GNNs) have been widely used in the field of cheminformatics, especially in the prediction of drug interactions, due to their powerful ability to process complex structured data. The core advantage of GNNs is that they can represent drug molecules as graph structures and extract local structural features of molecules through the relationship between network nodes (such as atoms) and edges (such as chemical bonds). These features are used to predict possible interactions between drugs.

[0010] Although the application of GNNs in DDI prediction shows great potential, existing methods still have certain limitations:

[0011] 1. Single information perspective: Most current GNN models mainly use the molecular structure information of drugs for prediction, but often ignore the topological relationship between drugs (such as the interaction between drugs in biological networks) and the fusion of other multi-perspective information. This single information perspective processing method limits the accuracy of model prediction to a certain extent.

[0012] 2. Model complexity and computational efficiency: As the scale of drug data increases, the complexity and computational overhead of the GNN model also increase. How to improve computational efficiency while maintaining high accuracy is an urgent problem to be solved.

[0013] 3. Challenges in fusion of multi-view information: GNNs still face challenges in aligning and integrating information from different viewpoints. In particular, how to achieve effective information fusion between multimodal data and integrate multi-view information while maintaining efficient computing is an important research direction in GNNs applications. Summary of the invention

[0014] The purpose of the present invention is to provide a method and device for predicting drug-drug interaction based on distribution matching in view of the deficiencies of the prior art.

[0015] The object of the present invention is achieved through the following technical solution: a drug-drug interaction prediction method based on distribution matching, comprising:

[0016] Extract the molecular structure characteristics of drugs;

[0017] Based on the molecular structure features, further extracting topological relationship features between drugs;

[0018] In the feature space and the output space, the molecular structure features and the topological relationship features are distributed and matched respectively to maximize their mutual information to predict DDI.

[0019] Preferably, the molecular structure characteristics of the extracted drug include:

[0020] The molecular structure characteristics of drugs are extracted through graph neural networks or graph convolutional networks.

[0021] Preferably, further extracting the topological relationship features between drugs based on the molecular structure features includes:

[0022] Through graph neural network or graph convolutional network, the topological relationship features between drugs are extracted.

[0023] Preferably, it also includes designing a loss function based on central matching distribution to balance the embedded information of the molecular structure characteristics of the drugs and the topological relationship characteristics between the drugs, as follows:

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032] in, is the total loss, To monitor losses, is the feature space matching loss, is the output space matching loss, , is a hyperparameter, , is the model parameter set, Indicates the molecular structure characteristics of the drug. Represents the topological relationship characteristics between drugs, represents the expectation, n represents the number of drug molecules, , express , The k-order central moment corresponding to the characteristic coordinates, , represent the model distribution obtained from the molecular structure characteristics and the topological relationship characteristics, respectively, is the number of labels, To predict the label, The real label, which is whether there is a drug-drug interaction, yes The real label, is a connection with label information, where , It is composed of various drug molecules. The topological relationship network diagram is composed of It is a collection of all connection information in the topology relationship.

[0033] The present invention also provides a drug-drug interaction prediction device based on distribution matching, comprising:

[0034] The first extraction module is used to extract the molecular structure characteristics of the drug;

[0035] A second extraction module is used to further extract topological relationship features between drugs based on the molecular structure features;

[0036] The prediction module is used to perform distribution matching on the molecular structure features and the topological relationship features in the feature space and the output space respectively, and maximize their mutual information to predict DDI.

[0037] Preferably, the molecular structure characteristics of the extracted drug include:

[0038] The molecular structure characteristics of drugs are extracted through graph neural networks or graph convolutional networks.

[0039] Preferably, further extracting the topological relationship features between drugs based on the molecular structure features includes:

[0040] Through graph neural network or graph convolutional network, the topological relationship features between drugs are extracted.

[0041] Preferably, it also includes designing a loss function based on central matching distribution to balance the embedded information of the molecular structure characteristics of the drugs and the topological relationship characteristics between the drugs, as follows:

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] in, is the total loss, To monitor losses, is the feature space matching loss, is the output space matching loss, , is a hyperparameter, , is the model parameter set, Indicates the molecular structure characteristics of the drug. Represents the topological relationship characteristics between drugs, represents the expectation, n represents the number of drug molecules, , express , The k-order central moment corresponding to the characteristic coordinates, , represent the model distribution obtained from the molecular structure characteristics and the topological relationship characteristics, respectively, is the number of labels, To predict the label, The real label, which is whether there is a drug-drug interaction, yes The real label, is a connection with label information, where , It is composed of various drug molecules. The topological relationship network diagram is composed of It is a collection of all connection information in the topology relationship.

[0051] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the above-mentioned drug-drug interaction prediction method based on distribution matching.

[0052] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the above-mentioned drug-drug interaction prediction method based on distribution matching is implemented.

[0053] Compared with the prior art, the beneficial effects of the embodiments of the present invention are:

[0054] 1. The embodiment of the present invention not only considers the molecular structure information of the drug, but also integrates the topological relationship information between drugs. This integration of multi-perspective information fully captures the complex interactions between drugs, thereby significantly improving the accuracy of prediction.

[0055] 2. Adopting distribution matching strategy: By performing distribution matching in feature space and output space, the DM-DDI model effectively improves the alignment effect of information from different perspectives. This strategy maximizes the mutual information between molecular structure features and topological relationship features, ensures the integrity and accuracy of the information, and further improves the prediction performance.

[0056] 3. Design the CMD loss function. Compared with the traditional contrastive learning method, the CMD loss function does not need to construct positive and negative sample pairs, which reduces the computational overhead. Under the premise of ensuring the prediction accuracy, the CMD loss function effectively improves the computational efficiency of the model. Experimental data shows that the DM-DDI model using the CMD loss function increases the computational speed by about 50% during the training process, greatly shortens the training time, and reduces the consumption of computing resources.

[0057] 4. The distribution matching strategy effectively avoids information loss, allowing the model to maintain stable prediction performance when processing drug interaction data with different data characteristics. This robustness enables the DM-DDI model to adapt to the complex changes and uncertainties in drug data and provide reliable prediction results.

[0058] 5. When data is scarce, the DM-DDI model can still maintain a high prediction accuracy, which shows that the model has strong adaptability and stability when dealing with insufficient data. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0060] Figure 1 A schematic flow chart of a method for predicting drug-drug interaction based on distribution matching provided by an embodiment of the present invention;

[0061] Figure 2 A schematic diagram of a topological relationship between drugs provided in an embodiment of the present invention;

[0062] Figure 3 A schematic diagram of a DM-DDI framework provided in an embodiment of the present invention;

[0063] Figure 4 A schematic diagram of a drug-drug interaction prediction device module based on distribution matching provided by an embodiment of the present invention;

[0064] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The present invention is described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the features of the following embodiments and implementations can be combined with each other.

[0066] A drug-drug interaction prediction method based on distribution matching of the present invention, such as Figure 1 As shown, the following steps are included:

[0067] (1) Extract the molecular structure characteristics of drugs;

[0068] Specifically, the molecular structure characteristics of the drug are extracted through graph neural networks or graph convolutional networks.

[0069] In one embodiment, the drug molecule is represented as a graph structure, where the nodes of the graph represent atoms in the molecule and the edges represent chemical bonds between atoms. The molecular graph is modeled and features are extracted through a graph neural network (GNN) to capture the local structural features within the molecule. GNN can effectively handle complex relationships in the molecular graph and improve the ability to learn the structural features of drug molecules. Of course, a graph convolutional network (GCN) can also be used, which will not be repeated here.

[0070] (2) Based on the molecular structure characteristics, further extracting the topological relationship characteristics between drugs;

[0071] Specifically, the topological relationship features between drugs are extracted through graph neural networks or graph convolutional networks.

[0072] In one embodiment, the interactions between drugs are represented as a topological graph and modeled using GCN. The construction of the topological graph takes into account the interactive relationship between drugs in the biological network and extracts the topological relationship features between drugs. GCN can capture the global and local relationships between drugs and enhance the understanding of the interaction patterns between drugs.

[0073] (3) In the feature space and the output space, the molecular structure features and the topological relationship features are distributed and matched respectively to maximize their mutual information to predict DDI.

[0074] Specifically, by maximizing the mutual information between the embedded representations of the molecular graph and the topological graph, the alignment effect between the two is ensured and information fusion is enhanced; multi-perspective information is effectively integrated to improve the comprehensive prediction ability of the model.

[0075] In one embodiment, a loss function based on central matching distribution (CMD) (hereinafter referred to as CMD loss function) is designed to balance the embedding information of the molecular structure characteristics of the drugs and the topological relationship characteristics between the drugs, as follows:

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084] in, is the total loss, To monitor losses, is the feature space matching loss, is the output space matching loss, , is a hyperparameter, , is the model parameter set, Indicates the molecular structure characteristics of the drug. Represents the topological relationship characteristics between drugs, represents the expectation, n represents the number of drug molecules, , express , The k-order central moment corresponding to the characteristic coordinates, , represent the model distribution obtained from the molecular structure characteristics and the topological relationship characteristics, respectively, is the number of labels, To predict the label, The real label, which is whether there is a drug-drug interaction, yes The real label, is a connection with label information, where , It is composed of various drug molecules. The topological relationship network diagram is composed of It is a collection of all connection information in the topology relationship.

[0085] Compared with traditional unsupervised contrastive learning methods, the CMD loss function can improve computational efficiency and model prediction without constructing positive and negative sample pairs. This method more efficiently captures the semantic relationship between different perspectives and optimizes the model training process.

[0086] Based on this, the present invention designs a DM-DDI model, see Figure 3 , including the following three modules:

[0087] Molecular graph module, used to extract the molecular structure features of drugs;

[0088] Topological map module, used to extract the topological relationship features between drugs;

[0089] The distribution matching module is used to perform distribution matching on the molecular structure features and the topological relationship features in the feature space and the output space respectively, and maximize their mutual information to predict DDI.

[0090] Experimental data and results:

[0091] 1. To ensure the comprehensiveness of the experiment and the reliability of the results, we selected the following three representative DDI datasets:

[0092] Dataset A: Contains large-scale drug-drug interaction data, covering the interaction information of multiple drugs. This dataset has high complexity and diversity, and is suitable for evaluating the performance of the model when processing complex data.

[0093] Dataset B: mainly includes medium-scale drug interaction data, providing a relatively balanced ratio of positive and negative samples. The characteristic of this dataset is that it covers a wide range of interaction information, which is helpful for testing the generalization ability of the model.

[0094] Dataset C: A smaller-scale drug interaction dataset, but with higher standards in terms of data annotation accuracy and quality. This dataset is used to evaluate the performance of the model in data-scarce situations.

[0095] 2. Evaluation Metrics

[0096] In order to comprehensively evaluate the performance of the DM-DDI model, we used the following key classification metrics:

[0097] AUC (Area Under the Curve): Measures the overall predictive performance of the model. The higher the AUC value, the better the model is at distinguishing between positive and negative samples.

[0098] F1-score: It is the harmonic mean of precision and recall. It is a commonly used indicator for evaluating model performance, especially for cases with unbalanced samples.

[0099] Precision: It indicates the proportion of samples predicted by the model to be positive that are actually positive, and measures the accuracy of the model's prediction results.

[0100] Recall: It indicates the proportion of actual positive samples that the model can correctly identify, and measures the model's coverage of positive samples.

[0101] 3. Experimental Results

[0102] The DM-DDI model was subjected to multiple experiments and compared with the existing state-of-the-art methods. The experimental results are as follows:

[0103] Dataset A:

[0104] The results show that the performance of our method DM-DDI is compared with that of other methods on the \emph{ZhangDDI} dataset, mainly evaluated from the following three indicators: AUROC, AUPRC and F1 score. The following are the main results and summary of the comparison:

[0105] 1.AUROC (Area Under the Receiver Operating Characteristic Curve):

[0106] DM-DDI reached 98.42, second only to MIRACLE's 98.95, with a very small gap.

[0107] The AUROCs of other methods range from 67.81 (NN) to 95.20 (Ens), and DM-DDI significantly outperforms all methods except MIRACLE.

[0108] 2.AUPRC (Area Under the Precision-Recall Curve):

[0109] DM-DDI’s performance on AUPRC is 96.31, which is also second only to MIRACLE’s 98.17.

[0110] The AUPRC of other methods are mostly between 52.61 (NN) and 93.07 (NFP-GCN), again indicating that DM-DDI and MIRACLE are significantly better than other methods.

[0111] 3.F1 score:

[0112] The F1 score of DM-DDI is 92.72, which is also slightly lower than MIRACLE's 93.20.

[0113] The F1 scores of other methods range from 49.84 (NN) to 85.41 (Ens), and the performance of DM-DDI and MIRACLE is also significantly higher than that of other methods.

[0114] Note: When comparing DM-DDI with other methods (especially MIRACLE), in addition to performance indicators (such as AUROC, AUPRC and F1 score), the reduction in computation is also a significant advantage. Specifically, DM-DDI achieves comparable or even better performance than the most advanced methods by optimizing the algorithm or model architecture, but the amount of computation required is greatly reduced. Compared with MIRACLE, the amount of computation has been reduced by 50%.

[0115] Dataset B:

[0116] In the experimental results, the DM-DDI algorithm showed excellent performance in many indicators, especially in AUPRC and F1 scores, which exceeded the MIRACLE algorithm. Specifically, DM-DDI achieved 97.94 in AUPRC, which is a significant improvement compared to MIRACLE's 95.57; in F1 score, DM-DDI also led, achieving a high score of 97.82, while MIRACLE was 92.26. These data show that the DM-DDI algorithm not only surpassed MIRACLE in overall performance, but also had significant advantages in the comprehensive performance of precision and recall.

[0117] Dataset C:

[0118] In the comparative experiment on Dataset C, the DM-DDI algorithm continued to demonstrate strong performance advantages. Although MIRACLE is slightly higher in AUROC, reaching 95.51, slightly ahead of DM-DDI's 93.17, the DM-DDI algorithm is significantly better than MIRACLE in AUPRC and F1 scores. Specifically, DM-DDI achieved 96.48 in AUPRC, far exceeding MIRACLE's 92.34, and in F1 score, DM-DDI is significantly ahead of MIRACLE's 83.60 with a score of 95.82. This result shows that DM-DDI has shown excellent results while maintaining a high overall performance, especially in the balance between precision and recall, further verifying its superiority in deep learning models.

[0119] Computational efficiency: In addition, we also evaluated the impact of the CMD loss function on the computational efficiency of the model. Experimental results show that compared with the traditional contrastive learning method, the DM-DDI model has a computational speed-up of about 50% during training. This improvement not only shortens the model training time, but also reduces the consumption of computing resources.

[0120] 4. Conclusion

[0121] The experimental results fully verified the effectiveness of the DM-DDI model in drug-drug interaction prediction. Through the evaluation of three datasets of different scales and difficulties, we observed that the DM-DDI model performed well in terms of prediction accuracy, computational efficiency, and model robustness. In particular, the introduction of the distribution matching strategy and the CMD loss function significantly improved the overall performance and computational efficiency of the model. These results show that the DM-DDI model has great potential in practical applications and can effectively support drug development and clinical decision-making.

[0122] The present invention also provides a drug-drug interaction prediction device based on distribution matching, such as Figure 3 As shown, including:

[0123] The first extraction module is used to extract the molecular structure characteristics of the drug;

[0124] A second extraction module is used to further extract topological relationship features between drugs based on the molecular structure features;

[0125] The prediction module is used to perform distribution matching on the molecular structure features and the topological relationship features in the feature space and the output space respectively, and maximize their mutual information to predict DDI.

[0126] It should be noted that the device embodiment shown in this embodiment matches the content of the above method embodiment, and the content of the above method embodiment can be referred to, which will not be repeated here.

[0127] Figure 5 This is a schematic diagram of an electronic device structure provided by an embodiment of the present invention. Figure 5 The electronic device provided in this embodiment includes: a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, a drug-drug interaction prediction method based on distribution matching of the present invention is implemented.

[0128] It should be noted that, in addition to Figure 5 In addition to the memory and processor shown, the electronic device may also include other hardware according to its actual functions, which will not be described in detail.

[0129] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a drug-drug interaction prediction method based on distribution matching in the above embodiment is implemented.

[0130] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capability and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0131] The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A drug-drug interaction prediction method based on distribution matching, characterized in that: include: Extract the molecular structure characteristics of drugs; Based on the molecular structure features, further extracting topological relationship features between drugs; In the feature space and the output space, the molecular structure features and the topological relationship features are respectively distributed and matched to maximize their mutual information to predict DDI; It also includes the design of a loss function based on central matching distribution to balance the embedding information of the molecular structure characteristics of the drug and the topological relationship characteristics between the drugs, as follows: L=L s +α1l fm +α2l om Among them, L is the total loss, L s To monitor the loss, l fm is the feature space matching loss, l om is the output space matching loss, α1 and α2 are hyperparameters, φ, is the model parameter set, Indicates the molecular structure characteristics of the drug. represents the topological relationship characteristics between drugs, E represents the expectation, n represents the number of drug molecules, Indicates g φ , The k-order central moment corresponding to the feature coordinates, p and r represent the model distribution obtained from the molecular structure feature and the topological relationship feature, respectively, c is the number of labels, Y is the predicted label, y is the actual label, and the label is whether there is a drug-drug interaction, y i Yes i The real label, is a connection with label information, where It is composed of various drug molecules. The topological relationship network diagram is composed of It is a collection of all connection information in the topology relationship.

2. A drug-drug interaction prediction method based on distribution matching according to claim 1, characterized in that: The molecular structure characteristics of the extracted drug include: The molecular structure characteristics of drugs are extracted through graph neural networks or graph convolutional networks.

3. A drug-drug interaction prediction method based on distribution matching according to claim 1, characterized in that: Based on the molecular structure features, further extracting the topological relationship features between drugs includes: Through graph neural network or graph convolutional network, the topological relationship features between drugs are extracted.

4. A drug-drug interaction prediction device based on distribution matching, characterized in that: include: The first extraction module is used to extract the molecular structure characteristics of the drug; The second extraction module is used to extract the molecular structure features and further extract the topological relationship features between drugs; A prediction module, used for performing distribution matching on the molecular structure features and the topological relationship features in the feature space and the output space respectively, maximizing their mutual information, so as to predict DDI; It also includes the design of a loss function based on central matching distribution to balance the embedding information of the molecular structure characteristics of the drug and the topological relationship characteristics between the drugs, as follows: <h2 style=";text-align:left;direction:ltr">L=L<h2 style=";text-align:left;direction:ltr"> s <h2 style=";text-align:left;direction:ltr"> +a1l<h2 style=";text-align:left;direction:ltr"> fm <h2 style=";text-align:left;direction:ltr"> +a2l<h2 style=";text-align:left;direction:ltr"> om Among them, L is the total loss, L s To monitor the loss, l fm is the feature space matching loss, l om is the output space matching loss, α1 and α2 are hyperparameters, φ, is the model parameter set, Indicates the molecular structure characteristics of the drug. represents the topological relationship characteristics between drugs, E represents the expectation, n represents the number of drug molecules, Indicates g φ , The k-order central moment corresponding to the feature coordinates, p and r represent the model distribution obtained from the molecular structure feature and the topological relationship feature, respectively, c is the number of labels, Y is the predicted label, y is the actual label, and the label is whether there is a drug-drug interaction, y i Yes i The real label, is a connection with label information, where It is composed of various drug molecules. The topological relationship network diagram is composed of It is a collection of all connection information in the topology relationship.

5. The drug-drug interaction prediction device based on distribution matching according to claim 4, characterized in that: The molecular structure characteristics of the extracted drug include: The molecular structure characteristics of drugs are extracted through graph neural networks or graph convolutional networks.

6. The drug-drug interaction prediction device based on distribution matching according to claim 4, characterized in that: Based on the molecular structure features, further extracting the topological relationship features between drugs includes: Through graph neural network or graph convolutional network, the topological relationship features between drugs are extracted.

7. An electronic device comprising a memory and a processor, characterized in that: The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement a drug-drug interaction prediction method based on distribution matching as described in any one of claims 1-3.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for predicting drug-drug interaction based on distribution matching as described in any one of claims 1 to 3 is implemented.