Drug target prediction method and system based on drug meta-learning
Through the small-sample drug target prediction method of drug meta-learning, a drug target prediction meta-learning framework was established using drug meta-learner and protein feature extraction technology, which solved the problem of low drug target prediction accuracy in small-sample scenarios and achieved efficient drug target prediction.
Patent Information
- Application Number
- CN202411041107.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2025-09-19
- Estimated Expiration
- 2044-07-31
AI Technical Summary
The prediction performance of existing drug target prediction methods decreases significantly in small sample scenarios, making it difficult to effectively use a small amount of drug-target relationship data for accurate prediction.
A small-sample drug target prediction method based on drug meta-learning is adopted. Through drug meta-learning and cross-task division of drug target prediction meta-tasks, a drug-target relationship prediction meta-learning framework is established. A drug meta-learner containing multiple graph convolutional layers is used to realize high-order graph feature learning of drugs, and combined with protein feature extraction and fully connected layers to output the relationship prediction score between drugs and targets.
It improves the accuracy of drug target prediction in small sample scenarios and enhances the ability to predict drug-target relationships under low-resource conditions.
Smart Images
Figure CN119181417B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of drug target prediction based on deep learning, and in particular to a small-sample drug target prediction method and system based on drug meta-learning. Background Art
[0002] Studying the association between drug molecules and potential target proteins is crucial for drug development. With the advancement of artificial intelligence (AI) and the accumulation of biomedical experimental data, the prediction of drug-target relationships using computational methods such as deep learning has made rapid progress. Existing methods achieve excellent drug-target prediction performance primarily due to the large amount of drug-target relationships used as model training data. However, these methods are limited in their ability to learn from drugs with a small number of targets. Specifically, when there is less data on drug-target relationships, the model's predictive performance significantly decreases, severely hindering real-world drug target discovery.
[0003] Computational drug target prediction refers to the use of computational technologies such as artificial intelligence and data mining to train and learn existing drug-target relationship data to construct a drug-target relationship prediction model. For example, Zhou Xuezhong et al. proposed a method and system for predicting traditional Chinese medicine targets based on matrix completion. This method uses a graph convolutional network to learn low-dimensional feature vectors of drugs and targets, and uses the matrix completion method to predict the relationship between the learned drug features and target features to obtain target prediction results. Li Yulin invented a method for predicting drug-target binding energy based on the quantum GAT algorithm. This method implements a quantum computing method of the graph attention mechanism, which can effectively avoid the problem of molecular spatial structure loss, capture the characteristics of interatomic bonds, and improve the model prediction performance. Liu Yongguo et al. invented a drug-target relationship prediction method based on collaborative matrix decomposition. This method not only considers the similarity of drug properties and target properties, but also combines the similarity of drug topological structures and target topological structures, thereby improving the accuracy of drug-target relationship prediction.
[0004] In real-world scenarios, target data for newly developed drugs is also very scarce, and high-quality drug-target relationship data is still scarce and difficult to obtain. This seriously restricts the predictive performance of existing drug target prediction methods. Therefore, there is an urgent need for a drug target prediction model that can still achieve good results in small sample scenarios. First, wet experiment-based drug-target relationship discovery technology requires verification of each drug target, but the potential target search space of drugs is generally very large (up to tens of thousands), which will consume a lot of manpower and material resources. Combining with artificial intelligence computing methods for drug target prediction is an effective way to solve this problem. However, most of these computing methods based on machine learning or deep learning are based on inductive reasoning, which often requires a large amount of drug-target relationship data to train the model in order to obtain better predictive performance. However, large amounts of high-quality drug-target relationship data are usually difficult to obtain, which limits the performance improvement of such methods and seriously restricts their further application. Summary of the Invention
[0005] The purpose of the present invention is to provide a small-sample drug target prediction method and system based on drug meta-learning. By utilizing the cross-task learning technology of meta-learning, through drug target meta-task division and drug meta-learning, a drug target prediction meta-learning framework is established, which improves the drug target prediction accuracy in small-sample scenarios to solve at least one technical problem existing in the above-mentioned background technology.
[0006] In order to achieve the above object, the present invention adopts the following technical solutions:
[0007] In a first aspect, the present invention provides a drug target prediction method based on drug meta-learning, comprising:
[0008] Obtain the chemical molecular formula information of the drug to be predicted and the amino acid sequence information of the target protein;
[0009] A pre-trained drug target prediction model is used to process the obtained drug molecular formula to be predicted and the amino acid sequence of the target protein to obtain a structure indicating whether there is a relationship between the drug to be predicted and the target; wherein, the drug target prediction model includes a drug meta-learning module, a protein feature extraction module and a prediction scoring module; the drug meta-learning module is used to implement high-order graph feature learning of the drug using a drug meta-learner containing multiple graph convolutional layers to obtain a meta-representation of the drug; the protein feature extraction module is used to extract protein features based on the amino acid sequence of the target protein; the prediction scoring module is used to splice the drug meta-representation and protein features, and output the drug-target relationship prediction score value through a fully connected layer.
[0010] Furthermore, the drug meta-learning module is used to use a drug meta-learner containing multiple graph convolutional layers to realize high-order graph feature learning of drugs and obtain the meta-representation of drugs, including: in each graph convolution layer, the learned node representation matrix, by performing an average pooling operation on the embedding of all node representation matrices, and through a fully connected layer, to obtain the meta-representation of the drug in the current task.
[0011] Furthermore, in each graph volume base layer, the learned node representation matrix includes: first converting the drug molecules into SMILES format, and using the RDKit library to convert the SMILES format drug molecules into molecular graphs; based on the atomic type features and chiral label features, the initial feature representation of the atoms is calculated; based on the adjacency matrix of the molecular graph and the initial feature representation, the node representation matrix is learned.
[0012] Furthermore, for the amino acid sequence of the protein, Conjoint Triad coding is used to convert it into a fingerprint representation, and then a protein feature extractor consisting of a two-layer fully connected neural network is used to achieve effective extraction of protein features.
[0013] Furthermore, the representation matrix X obtained by the lth convolutional layer l for:
[0014]
[0015] Where D represents the Laplace matrix calculated from the drug adjacency matrix A, W l-1 represents the parameter matrix of the l-1th convolutional layer, X l-1 represents the representation matrix of the l-1th convolutional layer, and σ represents the sigmoid activation function.
[0016] Furthermore, the meta-representation C of the drug in the current task is obtained Tc , as shown below:
[0017]
[0018] Where W and b are parameter matrices, N represents the number of atoms, X i L Represents the representation vector of the i-th atom in the L-th convolutional layer.
[0019] Furthermore, the fully connected layer outputs the predicted score of the relationship between drugs and proteins, including:
[0020]
[0021] h l =σ(W l h l-1 +b l )
[0022] s(d,p)=W L h L-1 +b L
[0023] Among them, MLP represents a multi-layer perceptron network. represents vector concatenation, σ is the sigmoid activation function; e p The fingerprint representation of the amino acid sequence of protein p, h 0 represents the input features of the fully connected layer, h l represents the output vector of the lth fully connected layer, and s(d,p) represents the predicted score values of drug d and protein p.
[0024] Furthermore, the cross entropy loss function L for the drug-target relationship binary classification task is:
[0025]
[0026] y i Represents the sample label, if drug d and protein p i If there is a relationship between i =1, if there is no relationship, y i =0; N represents the total number of samples in the support set under this task, s(d,p i ) represents drug d and protein p i The scoring value between .
[0027] In a second aspect, the present invention provides a drug target prediction system based on drug meta-learning, comprising:
[0028] An acquisition module is used to obtain the chemical molecular formula information of the drug to be predicted and the amino acid sequence information of the target protein;
[0029] The prediction module is used to use a pre-trained drug target prediction model to process the obtained drug molecular formula to be predicted and the amino acid sequence of the target protein to obtain a structure of whether the relationship between the drug to be predicted and the target exists; wherein, the drug target prediction model includes a drug meta-learning module, a protein feature extraction module and a prediction scoring module; the drug meta-learning module is used to use a drug meta-learner containing multiple graph convolution layers to realize high-order graph feature learning of the drug to obtain the meta-representation of the drug; the protein feature extraction module is used to extract protein features based on the amino acid sequence of the target protein; the prediction scoring module is used to splice the drug meta-representation and protein features, and output the predicted score value of the relationship between the drug and the target through the fully connected layer.
[0030] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the drug target prediction method based on drug meta-learning as described in the first aspect is implemented.
[0031] In a fourth aspect, the present invention provides a computer device comprising a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the drug target prediction method based on drug meta-learning as described in the first aspect.
[0032] In a fifth aspect, the present invention provides an electronic device comprising: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the drug target prediction method based on drug meta-learning as described in the first aspect.
[0033] The beneficial effects of the present invention are as follows: a drug target prediction algorithm based on drug meta-learning utilizes the cross-task learning technology of meta-learning, establishes a drug target prediction meta-learning framework through drug target meta-task division and drug meta-learning, and improves the drug target prediction accuracy in small sample scenarios.
[0034] Additional advantages of the present invention will be more clearly given in the following description or learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0036] Figure 1 This is a flow chart of the drug target prediction method based on drug meta-learning according to an embodiment of the present invention. DETAILED DESCRIPTION
[0037] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention.
[0038] Those skilled in the art will understand that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.
[0039] It should also be understood that terms, such as those defined in commonly used dictionaries, should be understood to have a meaning consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless as defined herein.
[0040] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.
[0041] In the description of this specification, reference to the terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Those skilled in the art may combine and integrate different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless otherwise contradictory.
[0042] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.
[0043] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.
[0044] In this invention, the low-resource scenario of drug-target relationship is modeled as a multi-task problem based on different drugs, and a drug target prediction method based on drug meta-learning is proposed. Through drug-target meta-task division and drug meta-learning in small sample scenarios, a drug target prediction meta-learning framework is established. It is expected that the drug-target relationship prediction capability in low-resource scenarios can be obtained by learning only a small amount of drug-target relationships.
[0045] Example 1
[0046] In this embodiment 1, a drug target prediction system based on drug meta-learning is first provided, which includes: an acquisition module for acquiring the chemical formula information of the drug to be predicted and the amino acid sequence information of the target protein. A prediction module is used to process the acquired drug formula to be predicted and the amino acid sequence of the target protein using a pre-trained drug target prediction model to obtain a structure indicating whether there is a relationship between the drug to be predicted and the target; wherein the drug target prediction model includes a drug meta-learning module, a protein feature extraction module, and a prediction scoring module; the drug meta-learning module is used to implement high-order graph feature learning of the drug using a drug meta-learner containing multiple graph convolutional layers to obtain a meta-representation of the drug; the protein feature extraction module is used to extract protein features based on the amino acid sequence of the target protein; the prediction scoring module is used to splice the drug meta-representation and protein features, and output the drug-target relationship prediction score value through a fully connected layer.
[0047] In this embodiment, the above-mentioned system is used to practice a drug target prediction method based on drug meta-learning, which includes: using an acquisition module to obtain the chemical formula information of the drug to be predicted and the amino acid sequence information of the target protein. Using a prediction module to process the obtained drug formula to be predicted and the amino acid sequence of the target protein using a pre-trained drug target prediction model to obtain a structure indicating whether there is a relationship between the drug to be predicted and the target; wherein the drug target prediction model includes a drug meta-learning module, a protein feature extraction module, and a prediction scoring module; the drug meta-learning module is used to implement high-order graph feature learning of the drug using a drug meta-learner containing multiple graph convolutional layers to obtain a meta-representation of the drug; the protein feature extraction module is used to extract protein features based on the amino acid sequence of the target protein; the prediction scoring module is used to splice the drug meta-representation and protein features, and output the drug-target relationship prediction score value through a fully connected layer.
[0048] Among them, the drug meta-learning module is used to use a drug meta-learner containing multiple graph convolutional layers to realize high-order graph feature learning of drugs and obtain the meta-representation of drugs, including: in each graph convolution layer, the learned node representation matrix, by averaging the embedding of all node representation matrices, and through a fully connected layer, to obtain the meta-representation of the drug in the current task.
[0049] In each graph volume base layer, the learned node representation matrix includes: first converting the drug molecules into SMILES format, and using the RDKit library to convert the SMILES format drug molecules into molecular graphs; based on the atomic type features and chiral label features, the initial feature representation of the atoms is calculated; based on the adjacency matrix of the molecular graph and the initial feature representation, the node representation matrix is learned.
[0050] For the amino acid sequence of the protein, Conjoint Triad coding is used to convert it into a fingerprint representation, and then a protein feature extractor consisting of a two-layer fully connected neural network is used to effectively extract protein features.
[0051] The node representation matrix X obtained by the lth convolutional layer l for:
[0052]
[0053] Where D represents the Laplace matrix calculated from the drug adjacency matrix A, W l-1 represents the parameter matrix of the l-1th convolutional layer, X l-1 represents the representation matrix of the l-1th convolutional layer, and σ represents the sigmoid activation function.
[0054] The meta-representation C of the drug in the current task is obtained Tc , as shown below:
[0055]
[0056] Among them, W and b are parameter matrices, N represents the number of atoms, X i L Represents the representation vector of the i-th atom in the L-th convolutional layer.
[0057] Furthermore, the fully connected layer outputs the predicted score of the relationship between drugs and proteins, including:
[0058]
[0059] h l =σ(W l h l-1 +b l )
[0060] s(d,p)=W L h L-1 +b L
[0061] Among them, MLP represents a multi-layer perceptron network. represents vector concatenation, σ is the sigmoid activation function; e p The fingerprint representation of the amino acid sequence of protein p, h 0 represents the input features of the fully connected layer, h l represents the output vector of the lth fully connected layer, and s(d,p) represents the predicted score values of drug d and protein p.
[0062] Furthermore, the cross entropy loss function L for the drug-target relationship binary classification task is:
[0063]
[0064] y i Represents the sample label, if drug d and protein p i If there is a relationship between i =1, if there is no relationship, y i =0; N represents the total number of samples in the support set under this task, s(d,p i ) represents drug d and protein p i The scoring value between .
[0065] Example 2
[0066] In this Example 2, a small-sample drug target prediction method based on drug meta-learning is provided. This method establishes a drug target prediction meta-learning framework by partitioning drug target meta-tasks and conducting drug meta-learning in small-sample scenarios, thereby improving the accuracy of drug target prediction in these scenarios. First, a drug target prediction model is trained. The model is then used to predict the drug's chemical formula and target protein's amino acid sequence information to obtain the final drug-target relationship prediction results.
[0067] First, in this embodiment, it is proposed to divide the meta-task according to different drugs. Specifically, the drug-target relationship data set is divided into multiple independent prediction tasks according to different drugs, that is, m drugs correspond to m tasks. Then, in each task, the drug target is divided into two parts according to the support set and the query set, where there is only one drug-target relationship sample in the support set for the model to learn the drug molecule representation of the current task, and there are multiple drug-target relationships that need to be predicted in the query set and are used for the final relationship prediction. In addition, these tasks are divided into training task sets and test task sets according to the difference between training and testing, where the training task set is used for the model to learn the drug-target relationship, and the test task set is used to evaluate the model learning effect.
[0068] like Figure 1As shown in the figure, when training on the support set, the input is the drug chemical formula and protein amino acid sequence of the task in the support set. First, the drug meta-representation is obtained using the drug meta-learning module. For proteins, fingerprint coding (Conjoint Triad coding) is first used to convert them into fingerprint representations. Then, a protein feature extractor containing two layers of fully connected neural networks is constructed to achieve effective extraction of protein features. After that, the drug meta-representation and protein feature representation are spliced and passed through a prediction and scoring module containing two layers of fully connected layers. In the prediction stage on the query set, the fast gradient of the compound meta-representation obtained by training on the support set is used to achieve rapid learning of the drug meta-representation on the query set. Finally, the updated compound meta-representation and candidate protein are subjected to the same calculation process as the support stage. Finally, the predicted scores of the drug and candidate protein can be obtained, and the drug-target relationship can be predicted.
[0069] In this embodiment 2, the training method of the drug target prediction model includes the following steps:
[0070] The drug target prediction meta-learning framework includes a training part on the support set and a prediction part on the query set.
[0071] When training on the support set, the input is the drug-target relationship of the task in the support set as well as the drug chemical formula and the amino acid sequence of the protein.
[0072] First, the drug meta-representation C is obtained using the drug meta-learning module. Tc In the drug meta-learning part, a drug meta-learner containing multiple graph convolutional layers is designed to realize the high-order graph feature learning of drugs and obtain the meta-representation of drugs. Specifically, each drug molecule is first converted into SMILES format, and then the SMILES format is converted into a molecular graph using the RDKit library (each atom corresponds to a node, and the relationship between atoms corresponds to the edge in the graph). The adjacency matrix of the molecular graph is denoted as A. Then, by obtaining the atom type feature (denoted as Xa) and the chiral label feature (denoted as Xc), the initial feature representation of the atom can be calculated, that is, X 0 =Xa+Xc. The drug meta-learner contains multiple graph convolutional layers. The node representation matrix learned by the lth graph convolutional layer is as follows.
[0073]
[0074] D represents the Laplace matrix calculated from the drug adjacency matrix A, W l-1 represents the parameter matrix of the l-1th convolutional layer, X l-1represents the representation matrix of the l-1th convolutional layer, and σ is the sigmoid activation function. After multiple graph convolutional layers, we can finally obtain the meta-representation C of the drug in the current task by performing an average pooling operation on all node embeddings and passing through a fully connected layer. Tc , as shown below.
[0075]
[0076] W and b are parameter matrices, N represents the number of atoms, X i L Represents the representation vector of the i-th atom in the L-th convolutional layer.
[0077] For the amino acid sequence of the protein, first use Conjoint Triad coding to convert it into a fingerprint representation. p We then constructed a protein feature extractor consisting of a two-layer fully connected neural network to effectively extract protein features. We then concatenated the drug meta-representation and protein feature representation and passed them through a prediction and scoring module consisting of two fully connected layers to achieve scoring that supports drug-target relationships, as shown below.
[0078]
[0079] h l =σ(W l h l-1 +b l )
[0080] s(d,p)=W L h L-1 +b L
[0081] W and b are both parameter matrices, MLP represents a multi-layer perceptron network, represents vector concatenation, σ is the sigmoid activation function; e p The fingerprint representation of the amino acid sequence of protein p, h 0 represents the input features of the fully connected layer, h l represents the output vector of the lth fully connected layer, and s(d,p) represents the predicted score values of drug d and protein p.
[0082] In order to implement model training, a cross entropy loss function for the drug-target relationship binary classification task is defined as follows.
[0083]
[0084] y i Represents the sample label, if drug d and protein p i If there is a relationship between i=1, if there is no relationship, y i =0; N represents the total number of samples in the support set under this task, s(d,p i ) represents drug d and protein p i The scoring value between .
[0085] Finally, the target loss function is used to represent the drug element C Tc By performing derivative calculations, we can obtain the rapid gradient G of the drug element. Tc ,as follows.
[0086]
[0087] In the prediction stage on the query set, the fast gradient G of the compound element representation trained on the support set can be used Tc , to achieve rapid learning of drug meta-representation on the query set, as shown in the following formula.
[0088]
[0089] Where β is the rate, which represents the step size of the drug meta-fast gradient.
[0090] Finally, the updated compound element is represented as After the candidate protein undergoes the same calculation process as in the support phase, the predicted scores of the drug and the candidate protein can be obtained, thus realizing the prediction of the drug-target relationship.
[0091] Example 3
[0092] This embodiment 3 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the small sample drug target prediction method based on drug meta-learning as described above is implemented. The method includes:
[0093] Obtain the chemical molecular formula information of the drug to be predicted and the amino acid sequence information of the target protein;
[0094] A pre-trained drug target prediction model is used to process the obtained drug molecular formula to be predicted and the amino acid sequence of the target protein to obtain a structure indicating whether there is a relationship between the drug to be predicted and the target; wherein, the drug target prediction model includes a drug meta-learning module, a protein feature extraction module and a prediction scoring module; the drug meta-learning module is used to implement high-order graph feature learning of the drug using a drug meta-learner containing multiple graph convolutional layers to obtain a meta-representation of the drug; the protein feature extraction module is used to extract protein features based on the amino acid sequence of the target protein; the prediction scoring module is used to splice the drug meta-representation and protein features, and output the drug-target relationship prediction score value through a fully connected layer.
[0095] Example 4
[0096] This embodiment 4 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the small sample drug target prediction method based on drug meta-learning as described above, the method comprising:
[0097] Obtain the chemical molecular formula information of the drug to be predicted and the amino acid sequence information of the target protein;
[0098] A pre-trained drug target prediction model is used to process the obtained drug molecular formula to be predicted and the amino acid sequence of the target protein to obtain a structure indicating whether there is a relationship between the drug to be predicted and the target; wherein, the drug target prediction model includes a drug meta-learning module, a protein feature extraction module and a prediction scoring module; the drug meta-learning module is used to implement high-order graph feature learning of the drug using a drug meta-learner containing multiple graph convolutional layers to obtain a meta-representation of the drug; the protein feature extraction module is used to extract protein features based on the amino acid sequence of the target protein; the prediction scoring module is used to splice the drug meta-representation and protein features, and output the drug-target relationship prediction score value through a fully connected layer.
[0099] Example 5
[0100] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the small sample drug target prediction method based on drug meta-learning as described above, the method including:
[0101] Obtain the chemical molecular formula information of the drug to be predicted and the amino acid sequence information of the target protein;
[0102] A pre-trained drug target prediction model is used to process the obtained drug molecular formula to be predicted and the amino acid sequence of the target protein to obtain a structure indicating whether there is a relationship between the drug to be predicted and the target; wherein, the drug target prediction model includes a drug meta-learning module, a protein feature extraction module and a prediction scoring module; the drug meta-learning module is used to implement high-order graph feature learning of the drug using a drug meta-learner containing multiple graph convolutional layers to obtain a meta-representation of the drug; the protein feature extraction module is used to extract protein features based on the amino acid sequence of the target protein; the prediction scoring module is used to splice the drug meta-representation and protein features, and output the drug-target relationship prediction score value through a fully connected layer.
[0103] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0104] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0105] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0106] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide the functions for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.
[0107] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solutions disclosed in the present invention without the need for creative work should be included in the scope of protection of the present invention.
Claims
1. A drug target prediction method based on drug meta-learning, characterized in that: include: Obtain the chemical molecular formula information of the drug to be predicted and the amino acid sequence information of the target protein; The obtained drug molecular formula to be predicted and the amino acid sequence of the target protein are processed using a pre-trained drug target prediction model to obtain a structure indicating whether a relationship exists between the drug to be predicted and the target. The drug target prediction model includes a drug meta-learning module, a protein feature extraction module, and a prediction scoring module. The drug meta-learning module is used to learn high-order graph features of the drug using a drug meta-learner comprising multiple graph convolutional layers to obtain a meta-representation of the drug. The protein feature extraction module is used to extract protein features based on the amino acid sequence of the target protein. The prediction scoring module is used to splice the meta-representation of the drug and the protein features, and output the predicted score value of the relationship between the drug and the target through the fully connected layer.
2. The drug target prediction method based on drug meta-learning according to claim 1, characterized in that: The drug meta-learning module is used to use a drug meta-learner containing multiple graph convolutional layers to realize high-order graph feature learning of drugs and obtain the meta-representation of drugs, including: in each graph convolution layer, the learned node representation matrix, by averaging the node representation matrix embeddings, and through a fully connected layer to obtain the meta-representation of the drug in the current task.
3. The drug target prediction method based on drug meta-learning according to claim 2, characterized in that: In each graph volume base layer, the learned node representation matrix includes: first converting the drug molecules into SMILES format, and using the RDKit library to convert the SMILES format drug molecules into molecular graphs; based on the atomic type features and chiral label features, the initial feature representation of the atoms is calculated; based on the adjacency matrix of the molecular graph and the initial feature representation, the node representation matrix is learned.
4. The drug target prediction method based on drug meta-learning according to claim 3, characterized in that: For the amino acid sequence of the protein, Conjoint Triad coding is used to convert it into a fingerprint representation, and then a protein feature extractor consisting of a two-layer fully connected neural network is used to effectively extract protein features.
5. The drug target prediction method based on drug meta-learning according to claim 4, characterized in that: The node representation matrix X obtained by the lth convolutional layer l for: Where D represents the Laplace matrix calculated from the drug adjacency matrix A, W l-1 represents the parameter matrix of the l-1th convolutional layer, X l-1 represents the representation matrix of the l-1th convolutional layer, and σ represents the sigmoid activation function.
6. The drug target prediction method based on drug meta-learning according to claim 5, characterized in that: The meta-representation C of the drug in the current task is obtained Tc , as shown below: Among them, W and b are parameter matrices, N represents the number of atoms; X i L Represents the representation vector of the i-th atom in the L-th convolutional layer.
7. The drug target prediction method based on drug meta-learning according to claim 6, characterized in that: The fully connected layer then outputs the predicted score of the relationship between the drug and the target, including: h l =σ(W l h l-1 +b l ) s(d,p)=W L h L-1 +b L Among them, MLP represents a multi-layer perceptron network. represents vector concatenation, σ is the sigmoid activation function; e p The fingerprint representation of the amino acid sequence of protein p, h 0 represents the input features of the fully connected layer, h l represents the output vector of the lth fully connected layer, and s(d,p) represents the predicted score values of drug d and protein p.
8. The drug target prediction method based on drug meta-learning according to claim 7, characterized in that: The cross entropy loss function L for the drug-target relationship binary classification task is: y i Represents the sample label, if drug d and protein p i If there is a relationship between i =1, if there is no relationship, y i =0; N represents the total number of samples in the support set under this task, s(d,p i ) represents drug d and protein p i The scoring value between .
9. A drug target prediction system based on drug meta-learning, characterized in that: include: An acquisition module is used to obtain the chemical molecular formula information of the drug to be predicted and the amino acid sequence information of the target protein; A prediction module is configured to process the obtained molecular formula of the drug to be predicted and the amino acid sequence of the target protein using a pre-trained drug target prediction model to obtain a structure indicating whether a relationship exists between the drug to be predicted and the target. The drug target prediction model includes a drug meta-learning module, a protein feature extraction module, and a prediction scoring module. The drug meta-learning module is configured to learn high-order graph features of the drug using a drug meta-learner comprising multiple graph convolutional layers to obtain a meta-representation of the drug. The protein feature extraction module is configured to extract protein features based on the amino acid sequence of the target protein. The prediction scoring module is used to splice the meta-representation of the drug and the protein features, and output the predicted score value of the relationship between the drug and the target through the fully connected layer.
10. An electronic device, characterized in that: include: A processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions for implementing the drug target prediction method based on drug meta-learning as described in any one of claims 1 to 8.