Drug-target interaction prediction method and device based on representation alignment space

Abstract

A method and apparatus for predicting drug-target interactions based on a representation alignment space. The method initializes the protein representation dimension, drug representation dimension, common embedding representation dimension, the number of multilayer perceptron layers, and batch dimension parameters. Multiple single-source network data are obtained from multiple pharmaceutical databases and compared to construct a biomedical heterogeneous network. Based on a contrastive learning strategy, protein target and drug representations are integrated into a common embedding representation space, and a model is trained using DTI associations. The protein target and its corresponding drug representations are input into the trained model for DTI prediction, obtaining the most relevant drugs. By constructing a biomedical heterogeneous network, the present invention facilitates the prediction of DTI associations associated with unknown drugs. Furthermore, by establishing an embedding representation alignment framework, the method utilizes high-quality representation methods to characterize biomedical networks and biological entities, while also addressing the issue of sparse training data using contrastive loss.

Description

Technical Field

[0001] The present invention relates to the technical field of new drug discovery prediction, and in particular to a method and device for predicting drug-target interactions based on characterization alignment space. Background Art

[0002] Drug discovery is a complex, multi-stage process aimed at identifying and developing new compounds that can effectively treat specific diseases. Predicting drug-target interactions (DTIs) is a key challenge in drug discovery, helping scientists determine which drug molecules are likely to effectively bind to and regulate specific biological targets. However, traditional drug discovery methods, such as protein structure analysis, are not only time-consuming but also costly to identify DTIs. Furthermore, experimental procedures may be limited by factors such as sample preparation, further complicating drug discovery.

[0003] To address this issue, researchers have developed machine learning and artificial intelligence-driven models to improve prediction efficiency and accuracy. These algorithms can analyze heterogeneous biomedical networks composed of biological entities such as drugs, target structures, and drug targets to predict possible DTI patterns and associations within the network.

[0004] Existing methods are unable to predict DTI associations for novel drugs outside of biopharmaceutical networks because they rely on extensive prior data and known associations. In the absence of this data, models struggle to accurately predict, a phenomenon known as the cold-start problem. Addressing the cold-start problem is crucial for new drug discovery and improving the accuracy of new drug-target association predictions.

[0005] At the same time, network-based methods are increasingly favored in DTIs prediction due to their ability to effectively utilize biomedical heterogeneous networks (BioHNs). In particular, graph neural networks (GNNs) aggregate neighborhood information through a message passing mechanism and show significant advantages in DTI prediction.

[0006] However, methods relying on knowledge graphs or BioHNs typically treat DTI prediction as a binary classification or regression task, focusing primarily on memorizing the binding annotation ratios in the training data. This approach neglects the learning of molecular interaction features, limiting generalization to new data. In other words, GNN-based methods struggle to effectively predict DTIs when graph associations of unknown entities are lacking, making such methods unsuitable for drug discovery.

[0007] Therefore, how to invent a method that can effectively predict DTI has become an urgent problem to be solved. Summary of the Invention

[0008] To this end, the present invention provides a method and apparatus for predicting drug-target interactions based on representation alignment space. This method constructs large-scale heterogeneous biomedical networks, providing multi-layered association knowledge that facilitates the prediction of DTI associations associated with unknown drugs. Furthermore, based on the principles of contrastive learning, an embedded representation alignment framework is constructed. This approach not only utilizes high-quality representation methods to characterize biomedical networks and biological entities, but also leverages contrastive loss to address the issue of sparse training data.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting drug-target interactions based on characterizing alignment space, comprising:

[0010] Initialize the protein representation dimension, drug representation dimension, common embedding representation dimension, multi-layer perceptron layer number, and batch dimension parameters;

[0011] Acquire a plurality of single-source network data through a plurality of pharmaceutical databases; perform comparison processing on the plurality of single-source network data to construct a biomedical heterogeneous network;

[0012] Integrating the protein target and drug representations obtained by the network representation strategy into a common embedding representation space according to a contrastive learning strategy; and setting a model through DTI association training using the contrastive learning strategy.

[0013] The protein target and the drug representations corresponding to the protein target are input into the trained setting model to perform DTI prediction, and obtain the drugs most relevant to the protein target.

[0014] As a preferred solution of the drug-target interaction prediction method based on characterizing the alignment space, in the process of obtaining the plurality of single-source network data through a plurality of pharmaceutical databases, the plurality of pharmaceutical databases include the drug database DrugBank, the therapeutic target database TTD, the pharmacogenomics database PharmGKB, the comprehensive drug database DrugCentral and the drug repositioning database repoDB;

[0015] The single-source network data includes drug-target interactions, drug-disease interactions, and drug-drug interactions.

[0016] As a preferred solution of the drug-target interaction prediction method based on characterizing the alignment space, in the process of comparing and processing a number of the single-source network data to construct the biomedical heterogeneous network, the single-source network data are used to remove non-compliant data and redundant data according to set rules to construct the biomedical heterogeneous network; the drugs, diseases and targets in the biomedical heterogeneous network are used as nodes, the interactions between entities are used as edges, and the nodes in the biomedical heterogeneous network are numbered.

[0017] As a preferred solution of the drug-target interaction prediction method based on representation alignment space, in the process of integrating the protein target and drug representation obtained by the network representation strategy into the common embedding representation space according to the comparative learning strategy, the drug and protein target are respectively mapped to independent representation spaces through the network representation strategy or the biological entity representation strategy; a multi-layer perceptron is set as a projection module to map the drug representation and protein representation of the set dimension to the common embedding representation space of the same dimension.

[0018] As a preferred solution of the drug-target interaction prediction method based on the representation alignment space, in the process of training the set model through DTI association using the contrastive learning strategy, the parameters of the projection module are optimized by calculating the bidirectional cross entropy loss function. The formula for calculating the bidirectional cross entropy loss function is:

[0019]

[0020] Where n is the number of labels; and Represents sim and sim respectively T The probability of the label corresponding to the i-th row in .

[0021] The present invention also provides a drug-target interaction prediction device based on the representation alignment space, based on the above drug-target interaction prediction method based on the representation alignment space, comprising:

[0022] Parameter initialization module, used to initialize the protein representation dimension, drug representation dimension, common embedding representation dimension, multi-layer perceptron layer number and batch dimension parameters;

[0023] A biomedical heterogeneous network construction module is used to obtain a plurality of single-source network data through a plurality of pharmaceutical databases; compare and process the plurality of said single-source network data to construct a biomedical heterogeneous network;

[0024] a common embedding representation space representation integration module for integrating the protein target and drug representations obtained by the network representation strategy into the common embedding representation space according to a contrastive learning strategy; and setting a model through DTI association training using the contrastive learning strategy;

[0025] The DTI prediction module is used to input the protein target and the drug representations corresponding to the protein target into the trained setting model to perform DTI prediction and obtain the drugs most relevant to the protein target.

[0026] As a preferred embodiment of the drug-target interaction prediction device based on the representation alignment space, in the biomedical heterogeneous network construction module, in the process of obtaining the plurality of single-source network data through a plurality of pharmaceutical databases, the plurality of pharmaceutical databases include the drug database DrugBank, the therapeutic target database TTD, the pharmacogenomics database PharmGKB, the comprehensive drug database DrugCentral and the drug repositioning database repoDB;

[0027] The single-source network data includes drug-target interactions, drug-disease interactions, and drug-drug interactions.

[0028] As a preferred solution of the drug-target interaction prediction device based on the representation alignment space, in the biomedical heterogeneous network construction module, in the process of comparing and processing a plurality of the single-source network data to construct the biomedical heterogeneous network, the plurality of the single-source network data are removed from the non-compliant data and redundant data according to the set rules to construct the biomedical heterogeneous network; the drugs, diseases and targets in the biomedical heterogeneous network are used as nodes, the interactions between entities are used as edges, and the nodes in the biomedical heterogeneous network are numbered.

[0029] As a preferred solution of the drug-target interaction prediction device based on representation alignment space, in the common embedding representation space representation integration module, in the process of integrating the protein target and drug representation obtained by the network representation strategy into the common embedding representation space according to the comparative learning strategy, the drug and protein target are respectively mapped to independent representation spaces through the network representation strategy or the biological entity representation strategy; a multi-layer perceptron is set as a projection module to map the drug representation and protein representation of the set dimension to the common embedding representation space of the same dimension.

[0030] As a preferred solution of the drug-target interaction prediction device based on the representation alignment space, in the common embedding representation space representation integration module, in the process of training the set model through DTI association using the contrastive learning strategy, the parameters of the projection module are optimized by calculating the bidirectional cross entropy loss function. The formula for calculating the bidirectional cross entropy loss function is:

[0031]

[0032] Where n is the number of labels; and Represents sim and sim respectively T The probability of the label corresponding to the i-th row in .

[0033] The present invention has the following advantages: initializing the protein representation dimension, drug representation dimension, public embedding representation dimension, number of multilayer perceptron layers, and batch dimension parameters; obtaining a plurality of single-source network data through a plurality of pharmaceutical databases; performing comparison processing on the plurality of said single-source network data to construct a biomedical heterogeneous network; integrating the protein target and drug representation obtained by the network representation strategy into a public embedding representation space according to a contrastive learning strategy; utilizing the contrastive learning strategy to train a model through DTI association; inputting the protein target and a plurality of drug representations corresponding to the protein target into the trained set model for DTI prediction to obtain a plurality of drugs most relevant to the protein target. The present invention proposes a contrastive learning-based embedding representation alignment framework, constructing a public embedding representation space, extensively learning DTI association patterns by learning sparse DTI labels in the network, and further extending it to predicting the relevant DTI of new drugs. A multilayer perceptron collaborative projection mapping module is adopted to generate a cross-dimensional public embedding representation space, thereby effectively unifying multi-source, multimodal, and heterogeneous network representations and entity representations. The present invention also proposes a weak label loss function label calculation method that can reduce overfitting learning of other drug-target pairs, thereby improving generalization and being effectively applicable to completely new drugs. The present invention constructs a large-scale biomedical heterogeneous network for multi-source, heterogeneous biomedical data, which includes multiple biological entities and multiple associations between entities, provides multi-layer association knowledge, contains more hidden biological meanings, and is conducive to predicting DTI associations related to unknown drugs; based on the idea of ​​contrastive learning, an embedded representation alignment framework is constructed, which can not only use high-quality representation methods to characterize biomedical networks and biological entities, but also use contrastive loss to solve the problem of sparse training data; the designed weak label calculation method does not explicitly specify the degree of association between other drug-target pairs. While helping the model learn the drug-target pairs that are actually associated, it reduces overfitting of other drug-target pairs, thereby retaining the learning ability of one-to-many and many-to-many DTI. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can, without inventive effort, derive other implementation drawings based on the provided drawings.

[0035] The structures, proportions, sizes, etc. illustrated in this specification are intended solely to complement the contents disclosed herein and to facilitate understanding and reading by persons skilled in the art. They are not intended to limit the conditions under which the present invention may be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportions, or adjustments in sizes, without affecting the efficacy and objectives of the present invention, shall remain within the scope of the technical contents disclosed herein.

[0036] Figure 1 This is a schematic flow chart of the method for predicting drug-target interactions based on characterization alignment space provided in Example 1 of the present invention;

[0037] Figure 2 Schematic diagram of a public embedding representation alignment space constructed based on BioHNs in the drug-target interaction prediction method based on representation alignment space provided in Example 1 of the present invention;

[0038] Figure 3 Schematic diagram of the association calculation and ranking prediction module in the drug-target interaction prediction method based on the representation alignment space provided in Example 1 of the present invention;

[0039] Figure 4 Schematic diagram of the architecture of the drug-target interaction prediction device based on characterization alignment space provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0040] The following describes the implementation of the present invention using specific embodiments. Those skilled in the art will readily understand the other advantages and benefits of the present invention from the disclosure herein. Obviously, the embodiments described are only a portion of the present invention, not all of it. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0041] Example 1

[0042] See also Figure 1 , Example 1 of the present invention provides a drug-target interaction prediction method based on characterization alignment space, comprising the following steps:

[0043] S1. Initialize the protein representation dimension, drug representation dimension, common embedding representation dimension, multilayer perceptron layer number and batch dimension parameters;

[0044] S2. Acquire a plurality of single-source network data through a plurality of pharmaceutical databases; perform comparison processing on the plurality of single-source network data to construct a biomedical heterogeneous network;

[0045] S3. Integrating the protein target and drug representations obtained by the network representation strategy into a common embedding representation space according to a contrastive learning strategy; and setting a model through DTI association training using the contrastive learning strategy;

[0046] S4. Inputting the protein target and the representations of several drugs corresponding to the protein target into the trained setting model to perform DTI prediction, and obtaining several drugs that are most relevant to the protein target.

[0047] In this embodiment, in step S1, the protein representation dimension, drug representation dimension, common embedding representation dimension, number of multilayer perceptron layers, and batch dimension parameters are initialized.

[0048] Specifically, the protein representation dimension dim p , drug characterization dimension dim d , the public embedding representation dimension dim, the number of multi-layer perceptron layers l, the batch dimension n and other parameters are initialized.

[0049] In this embodiment, in step S2, in the process of obtaining the plurality of single-source network data through a plurality of pharmaceutical databases, the plurality of pharmaceutical databases include a drug database DrugBank, a therapeutic target database TTD, a pharmacogenomics database PharmGKB, a comprehensive drug database DrugCentral, and a drug repositioning database repoDB;

[0050] The single-source network data includes drug-target interactions, drug-disease interactions, and drug-drug interactions.

[0051] Specifically, multiple single-source network data such as drug-target interaction, drug-disease interaction, and drug-drug interaction are obtained through the drug database DrugBank, the therapeutic target database TTD, the pharmacogenomics database PharmGKB, the comprehensive drug database DrugCentral, and the drug repositioning database repoDB.

[0052] In this embodiment, in step S2, it also includes removing non-compliant data and redundant data from the single-source network data according to set rules in the process of comparing and processing the plurality of single-source network data to construct the biomedical heterogeneous network, thereby constructing the biomedical heterogeneous network; taking drugs, diseases and targets in the biomedical heterogeneous network as nodes, and interactions between entities as edges, and numbering the nodes in the biomedical heterogeneous network.

[0053] Specifically, all databases are compared, and according to the two rules that the target has a unique UniProt number and the target is marked as "reviewed", the data that does not meet the requirements and redundant data are removed to form a biomedical heterogeneous network. Drugs, diseases, and targets in the biomedical heterogeneous network are regarded as nodes, and the interactions between entities are regarded as edges. The nodes in the biomedical heterogeneous network are numbered: d represents drug nodes; p represents protein target nodes; di represents disease nodes; where num d 、num p 、num di Represent the total number of drug, target, and disease nodes, respectively.

[0054] In this embodiment, in step S3, as Figure 2 As shown, in the process of integrating the protein target and drug representation obtained by the network representation strategy into the common embedding representation space according to the contrastive learning strategy, the drug and protein target are respectively mapped to independent representation spaces through the network representation strategy or the biological entity representation strategy; a multi-layer perceptron is set as a projection module to map the drug representation and protein representation of the set dimension to the common embedding representation space of the same dimension.

[0055] Specifically, based on mature network representation methods or biological entity representation methods, drugs and protein targets are mapped to independent representation spaces respectively; a multi-layer perceptron (MLP) is deployed as a projection module to transform the dim d The drug representation and dimension are dim p The protein representations are mapped to a common embedding representation space with the same dimension dim.

[0056] In this embodiment, step S3 further includes using the contrastive learning strategy to train the set model through DTI association.

[0057] Specifically, the cos similarity is calculated as the distance matrix sim between the drug and the target in the representation space and its transposed sim T Represents the bidirectional distance matrix. Based on contrastive learning, the drug-target pairs that actually exist in the biomedical network are positive samples, and the rest are negative samples;

[0058] The labels are constructed in a weakly associated manner. Each sample is considered an independent category, and its label directly corresponds to the index in the batch, that is, label = [0, 1, 2, ..., n-1]. The bidirectional cross entropy loss function is calculated to optimize the projection module parameters. The formula for calculating the bidirectional cross entropy loss function is:

[0059]

[0060] Where n is the number of labels; and Represents sim and sim respectively T The probability of the label corresponding to the i-th row in .

[0061] In this embodiment, in step S4, the protein target and the representations of several drugs corresponding to the protein target are input into the trained setting model to perform DTI prediction, and obtain several drugs that are most relevant to the protein target.

[0062] Specifically, such as Figure 3 As shown, the protein target and N drug representations are input into the trained setting model to predict the top k drugs related to the protein target; they are mapped to the common embedding representation space through the projection module, and the cos similarity between the protein target representation and the N drug representations is calculated to measure the degree of association; the drugs are sorted from large to small according to the degree of association, and the top k drugs are the k drugs predicted by the setting model to be associated with the protein target.

[0063] In summary, the present invention initializes the protein representation dimension, drug representation dimension, public embedding representation dimension, multi-layer perceptron layer number and batch dimension parameters; obtains a number of single-source network data through a number of pharmaceutical databases; compares and processes a number of said single-source network data to construct a biomedical heterogeneous network; integrates the protein targets and drug representations obtained by the network representation strategy into a public embedding representation space according to a contrastive learning strategy; utilizes the contrastive learning strategy to train a model through DTI association; inputs the protein targets and a number of drug representations corresponding to the protein targets into the trained set model for DTI prediction to obtain a number of drugs that are most relevant to the protein targets. The present invention proposes an embedding representation alignment framework based on contrastive learning, constructs a public embedding representation space, and extensively learns DTI association patterns by learning sparse DTI labels in the network, and further expands it to predict the relevant DTI of new drugs. A multi-layer perceptron collaborative projection mapping module is adopted to generate a cross-dimensional public embedding representation space, thereby effectively unifying multi-source, multi-modal, and heterogeneous network representations and entity representations. The present invention also proposes a weak label loss function label calculation method that can reduce overfitting learning of other drug-target pairs, thereby improving generalization and being effectively applicable to completely new drugs. The present invention constructs a large-scale biomedical heterogeneous network for multi-source, heterogeneous biomedical data, which includes multiple biological entities and multiple associations between entities, provides multi-layer association knowledge, contains more hidden biological meanings, and is conducive to predicting DTI associations related to unknown drugs; based on the idea of ​​contrastive learning, an embedded representation alignment framework is constructed, which can not only use high-quality representation methods to characterize biomedical networks and biological entities, but also use contrastive loss to solve the problem of sparse training data; the designed weak label calculation method does not explicitly specify the degree of association between other drug-target pairs. While helping the model learn the drug-target pairs that are actually associated, it reduces overfitting of other drug-target pairs, thereby retaining the learning ability of one-to-many and many-to-many DTI.

[0064] It should be noted that the method of the embodiments of the present disclosure can be performed by a single device, such as a computer or server. The method of the embodiments of the present disclosure can also be applied in a distributed scenario, where multiple devices cooperate to perform the method. In such a distributed scenario, one of the multiple devices may only perform one or more steps of the method of the embodiments of the present disclosure, and the multiple devices will interact with each other to complete the method.

[0065] It should be noted that the above description is limited to some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in an order different from that described in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0066] Example 2

[0067] See also Figure 4 , Embodiment 2 of the present invention also provides a drug-target interaction prediction device based on characterization alignment space, comprising:

[0068] Parameter initialization module 001, used to initialize the protein representation dimension, drug representation dimension, common embedding representation dimension, multi-layer perceptron layer number and batch dimension parameters;

[0069] The biomedical heterogeneous network construction module 002 is used to obtain a plurality of single-source network data through a plurality of pharmaceutical databases; compare and process the plurality of single-source network data to construct a biomedical heterogeneous network;

[0070] The common embedding representation space representation integration module 003 is used to integrate the protein target and drug representations obtained by the network representation strategy into the common embedding representation space according to the contrastive learning strategy; using the contrastive learning strategy, the model is set through DTI association training;

[0071] The DTI prediction module 004 is used to input the protein target and the drug representations corresponding to the protein target into the trained setting model to perform DTI prediction and obtain the drugs most relevant to the protein target.

[0072] In this embodiment, in the biomedical heterogeneous network construction module 002, in the process of obtaining the plurality of single-source network data through a plurality of pharmaceutical databases, the plurality of pharmaceutical databases include the drug database DrugBank, the therapeutic target database TTD, the pharmacogenomics database PharmGKB, the comprehensive drug database DrugCentral, and the drug repositioning database repoDB;

[0073] The single-source network data includes drug-target interactions, drug-disease interactions, and drug-drug interactions.

[0074] In the process of comparing and processing the plurality of single-source network data to construct the biomedical heterogeneous network, the plurality of single-source network data are used to remove non-compliant data and redundant data according to set rules to construct the biomedical heterogeneous network; the drugs, diseases and targets in the biomedical heterogeneous network are used as nodes, the interactions between entities are used as edges, and the nodes in the biomedical heterogeneous network are numbered.

[0075] In this embodiment, in the common embedding representation space representation integration module 003, in the process of integrating the protein target and drug representation obtained by the network representation strategy into the common embedding representation space according to the comparative learning strategy, the drug and protein target are respectively mapped to independent representation spaces through the network representation strategy or the biological entity representation strategy; a multi-layer perceptron is set as a projection module to map the drug representation and protein representation of the set dimension to the common embedding representation space of the same dimension.

[0076] In the process of training the set model by DTI association using the contrastive learning strategy, the parameters of the projection module are optimized by calculating the bidirectional cross entropy loss function. The formula for calculating the bidirectional cross entropy loss function is:

[0077]

[0078] Where n is the number of labels; and Represents sim and sim respectively T The probability of the label corresponding to the i-th row in .

[0079] It should be noted that the information interaction, execution process, etc. between the modules of the above-mentioned system are based on the same concept as the method embodiment in Example 1 of the present application, and the technical effects they bring are the same as those of the method embodiment of the present application. For specific contents, please refer to the description in the method embodiment shown above in the present application, and no further details will be given here.

[0080] Example 3

[0081] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, in which a program code for a drug-target interaction prediction method based on characterization alignment space is stored. The program code includes instructions for executing the drug-target interaction prediction method based on characterization alignment space of embodiment 1 or any possible implementation thereof.

[0082] Computer-readable storage media can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0083] Example 4

[0084] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0085] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the drug-target interaction prediction method based on the characterization alignment space of Example 1 or any possible implementation thereof.

[0086] Specifically, the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor, which is implemented by reading software code stored in a memory. The memory can be integrated into the processor or located outside the processor and exist independently.

[0087] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode.

[0088] Obviously, those skilled in the art will appreciate that the various modules or steps of the present invention described above can be implemented using a general-purpose computing system. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Alternatively, they can be implemented using program code executable by a computing system, and thus, they can be stored in a storage system and executed by the computing system. In some cases, the steps shown or described herein can be performed in a different order than that shown, or they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0089] Although the present invention has been described in detail above using general descriptions and specific embodiments, it will be apparent to those skilled in the art that modifications and improvements may be made thereto. Therefore, such modifications and improvements, without departing from the spirit of the present invention, are intended to be within the scope of protection claimed herein.

Claims

1. A drug-target interaction prediction method based on characterization alignment space, characterized in that: include: Initialize the protein representation dimension, drug representation dimension, common embedding representation dimension, multi-layer perceptron layer number, and batch dimension parameters; Acquire a plurality of single-source network data through a plurality of pharmaceutical databases; perform comparison processing on the plurality of single-source network data to construct a biomedical heterogeneous network; According to the contrastive learning strategy, the protein target and drug representations obtained by the network representation strategy are integrated into the common embedding representation space; Using the contrastive learning strategy, the model was set through DTI correlation training; Inputting the protein target and the corresponding drug representations of the protein target into the trained setting model to perform DTI prediction, and obtaining the drugs most relevant to the protein target; In the process of integrating the protein target and drug representations obtained by the network representation strategy into the common embedding representation space according to the contrastive learning strategy, the drug and protein target are respectively mapped to independent representation spaces by the network representation strategy or the biological entity representation strategy; Setting a multi-layer perceptron as a projection module to map the drug representation and protein representation of set dimensions to the common embedding representation space of the same dimension; In the process of training the set model by DTI association using the contrastive learning strategy, the parameters of the projection module are optimized by calculating the bidirectional cross entropy loss function. The formula for calculating the bidirectional cross entropy loss function is: Where n is the number of labels; and Represents sim and sim respectively T The probability of the label corresponding to the i-th row in .

2. The drug-target interaction prediction method based on representation alignment space according to claim 1, characterized in that: In the process of obtaining the plurality of single-source network data through a plurality of pharmaceutical databases, the plurality of pharmaceutical databases include a drug database DrugBank, a therapeutic target database TTD, a pharmacogenomics database PharmGKB, a comprehensive drug database DrugCentral, and a drug repositioning database repoDB; The single-source network data includes drug-target interactions, drug-disease interactions, and drug-drug interactions.

3. The drug-target interaction prediction method based on representation alignment space according to claim 2, characterized in that: In the process of comparing and processing the plurality of single-source network data to construct the biomedical heterogeneous network, the plurality of single-source network data are used to remove non-compliant data and redundant data according to set rules to construct the biomedical heterogeneous network; the drugs, diseases and targets in the biomedical heterogeneous network are used as nodes, the interactions between entities are used as edges, and the nodes in the biomedical heterogeneous network are numbered.

4. A drug-target interaction prediction device based on a representation alignment space, using any of the drug-target interaction prediction methods based on a representation alignment space according to claims 1-3, characterized in that: include: Parameter initialization module, used to initialize the protein representation dimension, drug representation dimension, common embedding representation dimension, multi-layer perceptron layer number and batch dimension parameters; A biomedical heterogeneous network construction module is used to obtain a plurality of single-source network data through a plurality of pharmaceutical databases; compare and process the plurality of said single-source network data to construct a biomedical heterogeneous network; A common embedding representation space representation integration module is used to integrate the protein target and drug representations obtained by the network representation strategy into the common embedding representation space based on the contrastive learning strategy; Using the contrastive learning strategy, the model was set through DTI correlation training; A DTI prediction module is used to input the protein target and the drug representations corresponding to the protein target into the trained setting model to perform DTI prediction and obtain the drugs most relevant to the protein target; In the common embedding representation space representation integration module, in the process of integrating the protein target and drug representations obtained by the network representation strategy into the common embedding representation space according to the contrastive learning strategy, the drug and protein target are respectively mapped to independent representation spaces by the network representation strategy or the biological entity representation strategy; Setting a multi-layer perceptron as a projection module to map the drug representation and protein representation of set dimensions to the common embedding representation space of the same dimension; In the common embedding representation space representation integration module, in the process of training the set model through DTI association using the contrastive learning strategy, the parameters of the projection module are optimized by calculating the bidirectional cross entropy loss function. The formula for calculating the bidirectional cross entropy loss function is: Where n is the number of labels; and Represents sim and sim respectively T The probability of the label corresponding to the i-th row in .

5. The drug-target interaction prediction device based on representation alignment space according to claim 4, characterized in that: In the biomedical heterogeneous network construction module, in the process of obtaining the plurality of single-source network data through a plurality of pharmaceutical databases, the plurality of pharmaceutical databases include a drug database DrugBank, a therapeutic target database TTD, a pharmacogenomics database PharmGKB, a comprehensive drug database DrugCentral, and a drug repositioning database repoDB; The single-source network data includes drug-target interactions, drug-disease interactions, and drug-drug interactions.

6. The drug-target interaction prediction device based on representation alignment space according to claim 5, characterized in that: In the biomedical heterogeneous network construction module, in the process of comparing and processing the plurality of single-source network data to construct the biomedical heterogeneous network, the plurality of single-source network data are removed from the plurality of single-source network data according to set rules to construct the biomedical heterogeneous network; the drugs, diseases and targets in the biomedical heterogeneous network are used as nodes, the interactions between entities are used as edges, and the nodes in the biomedical heterogeneous network are numbered.

Images