Antiviral drug screening method, system and medium based on inductive hypergraph learning

Abstract

The present invention provides an antiviral drug screening method, system, and medium based on inductive hypergraph learning, belonging to the interdisciplinary field of bioinformatics, computational biology, and artificial intelligence. The method includes: S1. constructing a virus-drug adjacency matrix; S2. calculating a virus integration similarity matrix and a drug integration similarity matrix; S3. extracting statistical eigenvectors of viruses and drugs, graph theory eigenvectors, and potential eigenvectors of virus-drug pairs, and merging them row by row to obtain a total virus feature matrix and a total drug feature matrix; S4. solving the objective function using a dual coordinate descent method to obtain an optimized projection parameter matrix; S5. calculating prediction scores from the virus and drug perspectives, averaging the results to obtain a final virus-drug association prediction score matrix; S6. screening the scores of the target virus row and sorting them to obtain the final prediction result. The present invention not only improves prediction performance but also has good adaptability and generalizability.

Description

Technical Field

[0001] The present invention relates to the technical field of antiviral drug screening, and in particular to an antiviral drug screening method, system and medium based on inductive hypergraph learning. Background Art

[0002] With the increasing number of emerging and recurrent viral infections, the global disease burden caused by viral infections is increasing. Highly pathogenic viruses, in particular, often trigger global public health events due to their strong transmissibility, ability to mutate, and severe clinical harm. In recent years, the continuous emergence of viral mutations and epidemic strains has increasingly challenged the effectiveness of traditional antiviral drugs. The emergence of new variants, in particular, has made the development of new antiviral drugs a critical issue that needs to be addressed urgently.

[0003] The existing antiviral drug development process typically includes drug discovery, preclinical research, and clinical research. This process is lengthy, costly, and requires high-level biosafety laboratories and rigorous experimental procedures. Traditional drug development models are unable to meet the urgent need for rapid screening of antiviral drugs, leading to a growing interest in drug repurposing strategies. By uncovering new potential relationships between drugs and viruses or diseases, new uses for existing drugs can be developed, significantly shortening the time-to-market and reducing development costs.

[0004] In recent years, methods based on networks and computational models have been widely used in the field of virus-drug association prediction, greatly improving the efficiency of drug discovery. Network methods typically construct heterogeneous networks of drugs, viruses (or diseases), and their interactions, combining multi-source heterogeneous information such as chemical structure, genome, or protein interactions to achieve complex relationship modeling and feature propagation between molecular entities. Typical methods include constructing heterogeneous networks, algorithms based on similarity propagation, and the use of new-generation artificial intelligence methods such as matrix decomposition, random walks, and graph neural networks. In particular, with the rise of deep learning in recent years, more and more methods have used deep neural networks to encode, fuse, and predict the multidimensional features of drugs and viruses, improving the expressive power of low-dimensional embedded features and the generalization performance of models on new drugs and new viruses.

[0005] However, existing mainstream algorithms often face three technical challenges. First, many models perform calculations based solely on surface similarities such as drug chemical structures, rarely capturing deeper functional relationships and network information between drugs and viruses, thus impacting the reliability of the results. Second, most prediction models still rely on simple graph structure modeling, making it difficult to fully leverage higher-order structural relationships or complex interaction patterns between objects. Third, supervised learning often relies on limited and biased positive and negative samples, which can easily lead to overfitting risks and reduced generalization capabilities. Existing research (CN202310910294.0 Hypergraph Learning-Based Antiviral Drug Screening Method, System, and Storage Medium) has not fully explored the potential of hypergraph learning in antiviral drug screening. Therefore, developing a novel virus-drug association prediction method that integrates multi-source information, possesses high-order representation capabilities, and can still achieve efficient and reliable predictions under sparse sample conditions has become a technical bottleneck and core demand in the current field of viral drug screening. Summary of the Invention

[0006] The present invention provides an antiviral drug screening method, system and medium based on inductive hypergraph learning. In line with the trend of artificial intelligence integrating multi-source biological data, and in response to the urgent need for effective integration of complex object relationships, high-order network features and external knowledge in current virus-drug prediction scenarios, a new method for the initial screening of antiviral drugs based on hypergraph learning and inductive matrix completion mechanism is provided, which not only improves the prediction performance but also has good adaptability and promotion capabilities.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] The antiviral drug screening method based on inductive hypergraph learning includes the following steps:

[0009] S1. Construct a virus-drug adjacency matrix based on known virus-drug associations;

[0010] S2. Calculate the virus integration similarity matrix and the drug integration similarity matrix based on the virus-drug adjacency matrix;

[0011] S3. Construct a hypergraph feature set and extract statistical eigenvectors, graph-theoretic eigenvectors, and potential eigenvectors of virus and drug pairs from the virus integration similarity matrix, drug integration similarity matrix, and virus-drug adjacency matrix, respectively. Merge them row by row to obtain the total virus feature matrix and the total drug feature matrix.

[0012] S4. Based on inductive hypergraph learning, the objective function is constructed using the total feature matrix of the virus and the total feature matrix of the drug. The dual coordinate descent method is used to solve the objective function to obtain the optimized projection parameter matrix.

[0013] S5. Based on the optimized projection parameter matrix, the prediction scores are calculated from the virus perspective and the drug perspective respectively, and the results of the two are averaged to obtain the final virus-drug association prediction score matrix;

[0014] S6. Based on the final virus-drug association prediction score matrix, filter out the scores of the row where the target virus is located, and obtain the final prediction results after sorting.

[0015] In this specification, the statistical feature vector in S3 includes:

[0016] For a virus or drug, the sum of the number of associations observed in the corresponding i-th row or j-th column of the virus-drug adjacency matrix;

[0017] The mean of the similarity scores, i.e., the average of the i-th row or j-th column of the virus integration similarity matrix or the drug integration similarity matrix;

[0018] The range [0,1] is divided into n intervals and the proportion of virus or drug similarity scores falling into each interval is calculated.

[0019] In this specification, the graph theory feature vector in S3 includes:

[0020] the number of neighbors of a node in the unweighted graph of the viral integration similarity matrix or the drug integration similarity matrix;

[0021] The similarity values ​​of the node’s k nearest neighbors;

[0022] The average value of the eigenvector of the statistic among the k nearest neighbors;

[0023] The average value of the statistic feature vectors of the k nearest neighbors of a node in the similarity weighted graph;

[0024] Node betweenness, proximity, and eigenvector centrality;

[0025] The Page-Rank score of the node.

[0026] In this specification, the potential feature vectors of the virus-drug pair in S3 include:

[0027] The virus-drug adjacency matrix is ​​constructed by non-negative matrix factorization to obtain the latent vectors of viruses and drugs;

[0028] betweenness, proximity, and eigenvector centrality of the associated pairs of nodes;

[0029] Latent vectors extracted using an autoencoder.

[0030] In this specification, the objective function is constructed based on inductive hypergraph learning combined with kernel methods and elastic network regularization. The objective function includes:

[0031] The kernelized feature hypergraph regularization term is used to maintain the structural consistency of the feature space. The kernel mapping matrix of the feature matrix is ​​calculated using the kernel technique. The rows or columns of the corresponding feature matrices from the virus and drug perspectives are taken respectively, and the semi-positive definite matrix associated with the hypergraph is constructed.

[0032] The empirical loss term is used to measure the error between the projection prediction and the known label;

[0033] L 2,1 Regularization term, with sparse selection capability;

[0034] F-norm term, used to prevent overfitting;

[0035] Elastic network regularization, controlling L through μ1 and μ2 respectively 2,1 and L F The weight of the regularization term.

[0036] In this specification, the hypergraph construction parameter is set to have the number of neighbors k=5, and the hyperedge weight adopts the Gaussian kernel function.

[0037] In this specification, the method for calculating the semi-positive definite matrix related to the hypergraph is: using the K-nearest neighbor method based on the hypergraph feature set to generate hyperedges, and the calculation involves the unit matrix, the hypergraph weight matrix, the hypergraph association matrix, the vertex degree diagonal matrix and the matrix transpose.

[0038] In this specification, the kernel mapping matrix of the characteristic matrix in the objective function is calculated by the kernel technique, wherein the rows or columns of the corresponding characteristic matrix are taken for calculation from the virus perspective and the drug perspective respectively.

[0039] An antiviral drug screening system based on inductive hypergraph learning, applying any of the above-mentioned antiviral drug screening methods based on inductive hypergraph learning, the antiviral drug screening system based on inductive hypergraph learning comprises:

[0040] Adjacency matrix building module, used to construct the adjacency matrix of virus-drug association;

[0041] The similarity calculation module is used to calculate the Jaccard similarity and cosine distance similarity of viruses and drugs based on the adjacency matrix, and fuse them to obtain the virus integration similarity matrix and the drug integration similarity matrix;

[0042] The feature extraction and synthesis module is used to construct a hypergraph feature set, extract the statistical features of viruses and drugs, graph theory features, and potential features of virus-drug pairs from the virus integration similarity matrix and the virus-drug association adjacency matrix, and merge them to generate the total virus feature matrix and the total drug feature matrix;

[0043] The loss function optimization module is used to construct the objective function based on inductive hypergraph learning using the total virus feature matrix and the total drug feature matrix. The objective function is differentiated with respect to the projection parameter matrix to obtain an analytical solution, that is, the optimized projection parameter matrix.

[0044] The prediction score matrix generation module is used to calculate the prediction scores from the virus perspective and the drug perspective based on the optimized projection parameter matrix, and average them to generate the final virus-drug association prediction score matrix;

[0045] The prediction module is used to filter out the scores of the target virus row based on the final virus-drug association prediction score matrix, and obtain the final prediction results after sorting.

[0046] A computer-readable storage medium storing computer instructions. When a computer reads the computer instructions, the computer executes any one of the above-mentioned antiviral drug screening methods based on inductive hypergraph learning.

[0047] In summary, the present invention has at least the following beneficial effects:

[0048] 1. Hypergraph learning can preserve high-order relationships between multiple samples and enhance the model's generalization ability for sparse associations;

[0049] 2. The closed-form solution design makes the model efficient and easy to expand to larger samples.

[0050] In summary, the present invention provides an efficient, accurate, and easily scalable technical solution in the field of antiviral drug screening, which significantly improves the initial screening hit rate and reduces experimental costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] 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.

[0052] Figure 1 Schematic diagram of the steps of the antiviral drug screening method based on inductive hypergraph learning involved in the present invention.

[0053] Figure 2 Schematic diagram of the process of antiviral drug screening method based on inductive hypergraph learning involved in the present invention.

[0054] Figure 3 Schematic diagram of the comparison results between the performance of the present invention and the baseline method. DETAILED DESCRIPTION

[0055] Hereinafter, only certain exemplary embodiments are briefly described. As will be appreciated by those skilled in the art, the described embodiments may be modified in various ways without departing from the spirit or scope of the embodiments of the present invention. Therefore, the drawings and description are to be regarded as illustrative in nature and not restrictive.

[0056] The disclosure below provides many different embodiments or examples for implementing different structures of the embodiments of the present invention. In order to simplify the disclosure of the embodiments of the present invention, the components and configurations of specific examples are described below. Of course, these are merely examples and are not intended to limit the embodiments of the present invention. In addition, the embodiments of the present invention may repeat reference numerals and / or reference letters in different examples. Such repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or configurations discussed.

[0057] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0058] like Figure 1 As shown, this embodiment provides an antiviral drug screening method based on inductive hypergraph learning, comprising the following steps:

[0059] S1. Construct a virus-drug adjacency matrix based on known virus-drug associations;

[0060] S2. Calculate the virus integration similarity matrix and the drug integration similarity matrix based on the virus-drug adjacency matrix;

[0061] S3. Construct a hypergraph feature set and extract statistical eigenvectors, graph-theoretic eigenvectors, and potential eigenvectors of virus and drug pairs from the virus integration similarity matrix, drug integration similarity matrix, and virus-drug adjacency matrix, respectively. Merge them row by row to obtain the total virus feature matrix and the total drug feature matrix.

[0062] S4. Based on inductive hypergraph learning, the objective function is constructed using the total feature matrix of the virus and the total feature matrix of the drug. The dual coordinate descent method is used to solve the objective function to obtain the optimized projection parameter matrix.

[0063] S5. Based on the optimized projection parameter matrix, the prediction scores are calculated from the virus perspective and the drug perspective respectively, and the results of the two are averaged to obtain the final virus-drug association prediction score matrix;

[0064] S6. Based on the final virus-drug association prediction score matrix, filter out the scores of the row where the target virus is located, and obtain the final prediction results after sorting.

[0065] In some embodiments, the statistical feature vector in S3 includes:

[0066] For a virus or drug, the sum of the number of associations observed in the corresponding i-th row or j-th column of the virus-drug adjacency matrix;

[0067] The mean of the similarity scores, i.e., the average of the i-th row or j-th column of the virus integration similarity matrix or the drug integration similarity matrix;

[0068] The range [0,1] is divided into n intervals and the proportion of virus or drug similarity scores falling into each interval is calculated.

[0069] In some embodiments, the graph theory feature vector in S3 includes:

[0070] the number of neighbors of a node in the unweighted graph of the viral integration similarity matrix or the drug integration similarity matrix;

[0071] The similarity values ​​of the node’s k nearest neighbors;

[0072] The average value of the eigenvector of the statistic among the k nearest neighbors;

[0073] The average value of the statistic feature vectors of the k nearest neighbors of a node in the similarity weighted graph;

[0074] Node betweenness, proximity, and eigenvector centrality;

[0075] The Page-Rank score of the node.

[0076] In some embodiments, the potential feature vectors of the virus-drug pairs in S3 include:

[0077] The virus-drug adjacency matrix is ​​constructed by non-negative matrix factorization to obtain the latent vectors of viruses and drugs;

[0078] betweenness, proximity, and eigenvector centrality of the associated pairs of nodes;

[0079] Latent vectors extracted using an autoencoder.

[0080] In some embodiments, an objective function is constructed based on inductive hypergraph learning combined with kernel methods and elastic net regularization, and the objective function includes:

[0081] The kernelized feature hypergraph regularization term is used to maintain the structural consistency of the feature space. The kernel mapping matrix of the feature matrix is ​​calculated using the kernel technique. The rows or columns of the corresponding feature matrices from the virus and drug perspectives are taken respectively, and the semi-positive definite matrix associated with the hypergraph is constructed.

[0082] The empirical loss term is used to measure the error between the projection prediction and the known label;

[0083] L 2,1 Regularization term, with sparse selection capability;

[0084] F-norm term, used to prevent overfitting;

[0085] Elastic network regularization, controlling L through μ1 and μ2 respectively 2,1 and L F The weight of the regularization term.

[0086] In some embodiments, the number of neighbors k of the hypergraph construction parameters is set to 5, and the hyperedge weight adopts the Gaussian kernel function.

[0087] In some embodiments, the method for calculating the semi-positive definite matrix related to the hypergraph is: using the K-nearest neighbor method to generate hyperedges based on the hypergraph feature set, and the calculation involves the unit matrix, the hypergraph weight matrix, the hypergraph association matrix, the vertex degree diagonal matrix and the matrix transpose.

[0088] In some embodiments, a kernel mapping matrix of a feature matrix in an objective function is calculated by a kernel technique, wherein the calculation is performed by taking rows or columns of the corresponding feature matrix from the virus perspective and the drug perspective, respectively.

[0089] In some embodiments, an adjacency matrix is ​​constructed using known virus-drug associations:

[0090] Record known viruses as a set , drugs are recorded as a set Then, the virus-drug adjacency matrix is ​​constructed using known virus-drug associations ,in Indicates the number of viruses, Indicates the amount of medicine, The meanings of the elements are as follows:

[0091] ;

[0092] Using the adjacency matrix, we calculate the drug Jaccard similarity and drug cosine distance similarity, and then calculate the drug integrated similarity matrix. and The Jaccard similarity is calculated as follows:

[0093] ;

[0094] in and The virus-drug adjacency matrix is The p and q columns of represents the number of intersections between the two. Indicates the number of the union of the two; drug and The cosine similarity calculation method is:

[0095] ;

[0096] in is the dot product, is the L2 norm. Calculate the drug integration similarity matrix Chinese elements:

[0097] ;

[0098] Each element in the range satisfies .

[0099] Similar to the above method, the columns corresponding to the two viruses in the virus-drug adjacency matrix A are used to calculate the virus Jaccard similarity and virus cosine distance similarity, and then calculate the virus integrated similarity matrix S v .

[0100] Virus and The Jaccard similarity is calculated as follows:

[0101] ;

[0102] in and The virus-drug adjacency matrix is No. 、 List, represents the number of intersections between the two. Indicates the number of the two sets; virus and The cosine similarity calculation method is:

[0103] ;

[0104] in is the dot product, is the L2 norm. Calculate the virus integration similarity matrix Sv Chinese elements:

[0105] ;

[0106] Each element in the range satisfies .

[0107] In some embodiments, a hypergraph feature set is constructed by integrating similarity matrices from viruses S v , drug integration similarity matrix S d and the virus-drug adjacency matrix A Extract and calculate the eigenvectors and synthesize the feature matrix:

[0108] The Class I characteristics X1 (or X1') of each virus (or drug) include:

[0109] ①For virus v(i) (or drug d(j)), the sum of the number of associations observed in the corresponding i-th row (or j-th column) of the virus-drug adjacency matrix A;

[0110] ②The mean of the similarity scores, i.e. S v (or S d )’s average value in the i-th row (or j-th column):

[0111] ③ Divide the range [0,1] into n intervals and calculate the proportion of v(i) (or d(j)) similarity scores that fall into each interval.

[0112] The Class II characteristics X2 (or X2') of each virus (or drug) include:

[0113] ①S v (or S d ) in an unweighted graph;

[0114] ②The similarity values ​​of the node’s k nearest neighbors;

[0115] ③The average value of type I features among the k nearest neighbors;

[0116] ④ The average value of the type I features among the k nearest neighbors of the node in the similarity value weighted graph;

[0117] ⑤ Node betweenness, proximity, and eigenvector centrality;

[0118] ⑥Page-Rank score of the node.

[0119] The Class III features X3 (or X3') for each virus-drug association pair include:

[0120] ① The latent vectors of viruses and drugs obtained by non-negative matrix factorization of the virus-drug adjacency matrix A;

[0121] ② The betweenness, proximity, and eigenvector centrality of the associated nodes;

[0122] ③ Latent vector extracted using autoencoder.

[0123] Merge the three types of virus features by row to obtain the total feature matrix of the virus, namely X v =[X1X2 X3];

[0124] Merge the three types of drug features by row to obtain the total feature matrix of the drug, namely X d =[X1' X2' X3'].

[0125] According to inductive hypergraph learning, combined with kernel methods and elastic network regularization, the loss function is constructed and the objective function is solved:

[0126] ;

[0127] Where P is the projection parameter matrix to be determined; is the kernelized feature hypergraph regularization term, which maintains the structural consistency of the feature space. is the kernel mapping matrix of the feature matrix X, φ represents the mapping function, through the kernel technique calculate, represents the kernel function, x i 、 x j Taking X from the perspective of viruses v Rows or columns, take X from the drug perspective d The rows or columns of Q are the semi-positive definite matrix associated with the hypergraph, and the calculation method is , hyperedges are generated using the K-nearest neighbor method based on the hypergraph feature set. I represents the identity matrix, W represents the hypergraph weight matrix, H represents the hypergraph incidence matrix, and the value 1 indicates that the vertex belongs to the hyperedge, otherwise it is 0; D represents the vertex degree diagonal matrix, and T represents the matrix transpose; is the empirical loss term, which measures the error between the projection prediction and the known label; For L 2,1 The regular term has sparse selection capability. is the F norm term to prevent overfitting. μ1 and μ2 control L 2,1 and L F The weight of the regularization term forms the elastic net regularization.

[0128] The dual coordinate descent method is used to solve the problem, and the analytical solution of the projection coefficient is: , where R is a diagonal matrix with diagonal elements , initially let R be the identity matrix and iteratively update until convergence; are the diagonal elements of the matrix P; the introduction of the kernel matrix enables the model to capture nonlinear feature relationships, and the elastic network regularization balances feature selection and group effects.

[0129] Output prediction score matrix:

[0130] Based on the optimized projection parameter matrix obtained through learning, the final prediction score matrix is ​​calculated:

[0131] ;

[0132] The virus perspective , drug perspective , and are the optimized projection parameter matrices obtained after optimization.

[0133] The corresponding elements in the prediction score matrix S represent the association probability between the corresponding drug and the virus. The element value range is [0,1]. The higher the score, the greater the possibility of association.

[0134] In some embodiments, in actual applications, the following parameters are set: Hypergraph construction parameters Neighbor number k =5, the hyperedge weight adopts Gaussian kernel function; is a hyperparameter, and its empirical value is The number of intervals, n, was set to 8. Iteration termination criteria were: the change in the objective function value between two consecutive iterations was less than 10-6 or the maximum number of iterations, 1000, was reached. Five-fold cross-validation was used to evaluate model performance, and the AUC was used to measure prediction accuracy.

[0135] Verification of the effectiveness of the present invention:

[0136] like Figure 2 The antiviral drug screening method based on inductive hypergraph learning (IHL) described here uses five-fold cross-validation to evaluate predictive performance. Specifically, all known virus-drug associations are randomly divided into five groups. Each of these five groups is then used as a test sample, with the remaining groups serving as training samples. The method then uses the training samples as input to generate predictions. Finally, the prediction score for each test sample in the group is compared with the candidate score. To minimize the impact of random partitioning during the test sample generation process, 50 five-fold cross-validations are performed.

[0137] The following data were obtained after calculation using Matlab: Figure 3 The figure shows a comparison of the AUROC (area under the ROC curve) values ​​between this method and several previously reported virus-drug screening models. This method achieved an AUROC value of 0.8816±0.0042 in a 5-fold cross-validation, demonstrating superior predictive performance compared to several classical models.

[0138] On the other hand, this method is used to predict a specific virus, such as the new coronavirus (SARS-CoV-2). By screening the row corresponding to SARS-CoV-2 in the scoring matrix S, the prediction scores of related drugs are obtained. After arranging them in descending order, 17 of the top 20 drugs can be supported by reported literature.

[0139] The second and third columns in the table below represent the names of the top 20 drugs in the predicted results and the PMID numbers of the supporting literature.

[0140]

[0141] An antiviral drug screening system based on inductive hypergraph learning, applying any of the above-mentioned antiviral drug screening methods based on inductive hypergraph learning, the antiviral drug screening system based on inductive hypergraph learning comprises:

[0142] Adjacency matrix building module, used to construct the adjacency matrix of virus-drug association;

[0143] The similarity calculation module is used to calculate the Jaccard similarity and cosine distance similarity of viruses and drugs based on the adjacency matrix, and fuse them to obtain the virus integration similarity matrix and the drug integration similarity matrix;

[0144] The feature extraction and synthesis module is used to construct a hypergraph feature set, extract the statistical features of viruses and drugs, graph theory features, and potential features of virus-drug pairs from the virus integration similarity matrix and the virus-drug association adjacency matrix, and merge them to generate the total virus feature matrix and the total drug feature matrix;

[0145] The loss function optimization module is used to construct the objective function based on inductive hypergraph learning using the total virus feature matrix and the total drug feature matrix. The objective function is differentiated with respect to the projection parameter matrix to obtain an analytical solution, that is, the optimized projection parameter matrix.

[0146] The prediction score matrix generation module is used to calculate the prediction scores from the virus perspective and the drug perspective based on the optimized projection parameter matrix, and average them to generate the final virus-drug association prediction score matrix;

[0147] The prediction module is used to filter out the scores of the target virus row based on the final virus-drug association prediction score matrix, and obtain the final prediction results after sorting.

[0148] A computer-readable storage medium storing computer instructions. When a computer reads the computer instructions, the computer executes any one of the above-mentioned antiviral drug screening methods based on inductive hypergraph learning.

[0149] The above embodiments are intended to illustrate the present invention, not to limit the present invention. Therefore, changes in illustrative values ​​or substitutions of equivalent components should still fall within the scope of the present invention.

[0150] From the above detailed description, it will be clear to those skilled in the art that the present invention can indeed achieve the aforementioned objectives and is in compliance with the provisions of the Patent Law.

[0151] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as covering the preferred embodiments and all changes and modifications that fall within the scope of the invention. The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

[0152] It should be noted that the above description of the relevant processes is for illustration and purpose only and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the processes under the guidance of this specification. However, such modifications and changes are still within the scope of this specification.

[0153] The basic concepts have been described above. It will be apparent to those skilled in the art after reading this application that the above disclosures are merely illustrative and do not constitute limitations on this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and amendments to this application. Such modifications, improvements, and amendments are suggested in this application and remain within the spirit and scope of the exemplary embodiments of this application.

[0154] At the same time, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment," "an embodiment," and / or "some embodiments" refer to a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that "one embodiment," "an embodiment," or "an alternative embodiment" mentioned twice or more in different places in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application may be appropriately combined.

[0155] Furthermore, those skilled in the art will appreciate that various aspects of the present application may be illustrated and described in terms of a number of patentable categories or situations, including any new and useful process, machine, product, or combination of substances, or any new and useful improvement thereof. Thus, various aspects of the present application may be implemented entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software. Each of the above hardware and software may be referred to as a "unit," "module," or "system." Furthermore, various aspects of the present application may take the form of a computer program product embodied in one or more computer-readable media, with computer-readable program code embodied therein.

[0156] The computer program code required for the operation of each part of this application can be written in any one or more programming languages, including object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages ​​such as C programming language, Visual Basic, Fortran2103, Perl, COBOL2102, PHP, ABAP, dynamic programming languages ​​such as Python, Ruby and Groovy, or other programming languages. The program code can be run entirely on the user's computer, or as a standalone software package on the user's computer, or partly on the user's computer and partly on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer via any network form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., via the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0157] In addition, unless expressly stated in the claims, the order of the processing elements and sequences described in this application, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this application. Although the above disclosure discusses some embodiments of the invention that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this application. For example, although the implementation of the various components described above can be embodied in a hardware device, it can also be implemented as a pure software solution, for example, installation on an existing server or mobile device.

[0158] Similarly, it should be noted that in order to simplify the presentation of this disclosure and thereby facilitate understanding of one or more of the invention's embodiments, the foregoing descriptions of the embodiments of this disclosure sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this approach should not be interpreted as reflecting an intention that the claimed subject matter requires more features than expressly recited in each claim. Rather, the subject matter of the invention may possess fewer features than the single embodiment described above.

Claims

1. An antiviral drug screening method based on inductive hypergraph learning, characterized in that: The steps include: S1. Construct a virus-drug adjacency matrix based on known virus-drug associations; S2. Calculate the virus integration similarity matrix and the drug integration similarity matrix based on the virus-drug adjacency matrix; S3. Construct a hypergraph feature set and extract statistical eigenvectors, graph-theoretic eigenvectors, and potential eigenvectors of virus and drug pairs from the virus integration similarity matrix, drug integration similarity matrix, and virus-drug adjacency matrix, respectively. Merge them row by row to obtain the total virus feature matrix and the total drug feature matrix. S4. Based on inductive hypergraph learning, the objective function is constructed using the total feature matrix of the virus and the total feature matrix of the drug. The dual coordinate descent method is used to solve the objective function to obtain the optimized projection parameter matrix. S5. Based on the optimized projection parameter matrix, the prediction scores are calculated from the virus perspective and the drug perspective respectively, and the results of the two are averaged to obtain the final virus-drug association prediction score matrix; S6. Based on the final virus-drug association prediction score matrix, filter out the scores of the row where the target virus is located, and sort them to obtain the final prediction result; The objective function is constructed based on inductive hypergraph learning combined with kernel methods and elastic net regularization. The objective function includes: a kernelized feature hypergraph regularization term to maintain the structural consistency of the feature space, which is constructed in conjunction with the hypergraph-related semi-positive definite matrix; The empirical loss term is used to measure the error between the projection prediction and the known label; L 2,1 Regularization term, with sparse selection capability; F-norm term, used to prevent overfitting; Elastic network regularization, controlling L through μ1 and μ2 respectively 2,1 and L F The weight of the regularization term.

2. The antiviral drug screening method based on inductive hypergraph learning according to claim 1, characterized in that: The statistical feature vector in S3 includes: For a virus or drug, the sum of the number of associations observed in the corresponding i-th row or j-th column of the virus-drug adjacency matrix; The mean of the similarity scores, i.e., the average of the i-th row or j-th column of the virus integration similarity matrix or the drug integration similarity matrix; The range [0,1] is divided into n intervals and the proportion of virus or drug similarity scores falling into each interval is calculated.

3. The antiviral drug screening method based on inductive hypergraph learning according to claim 1, characterized in that: The graph theory feature vectors in S3 include: the number of neighbors of a node in the unweighted graph of the viral integration similarity matrix or the drug integration similarity matrix; The similarity values ​​of the node’s k nearest neighbors; The average value of the eigenvector of the statistic among the k nearest neighbors; The average value of the statistic feature vectors of the k nearest neighbors of a node in the similarity weighted graph; Node betweenness, proximity, and eigenvector centrality; The Page-Rank score of the node.

4. The antiviral drug screening method based on inductive hypergraph learning according to claim 1, characterized in that: The potential feature vectors of the virus-drug pair in S3 include: The virus-drug adjacency matrix is ​​constructed by non-negative matrix factorization to obtain the latent vectors of viruses and drugs; betweenness, proximity, and eigenvector centrality of the associated pairs of nodes; Latent vectors extracted using an autoencoder.

5. The antiviral drug screening method based on inductive hypergraph learning according to claim 1, characterized in that: The number of neighbors in the hypergraph construction parameters is set to k = 5, and the hyperedge weight adopts the Gaussian kernel function.

6. The antiviral drug screening method based on inductive hypergraph learning according to claim 1, characterized in that: The method for calculating the semi-positive definite matrix related to the hypergraph is: based on the hypergraph feature set, the K-nearest neighbor method is used to generate hyperedges. The calculation involves the unit matrix, the hypergraph weight matrix, the hypergraph incidence matrix, the vertex degree diagonal matrix, and the matrix transpose.

7. The antiviral drug screening method based on inductive hypergraph learning according to claim 1, characterized in that: The kernel mapping matrix of the characteristic matrix in the objective function is calculated by the kernel technique, where the rows or columns of the corresponding characteristic matrix are taken for calculation from the virus perspective and the drug perspective respectively. 8.An antiviral drug screening system based on inductive hypergraph learning, characterized by: The antiviral drug screening method based on inductive hypergraph learning according to any one of claims 1 to 7 is applied, wherein the antiviral drug screening system based on inductive hypergraph learning comprises: Adjacency matrix building module, used to construct the adjacency matrix of virus-drug association; The similarity calculation module is used to calculate the Jaccard similarity and cosine distance similarity of viruses and drugs based on the adjacency matrix, and fuse them to obtain the virus integration similarity matrix and the drug integration similarity matrix; The feature extraction and synthesis module is used to construct a hypergraph feature set, extract the statistical features of viruses and drugs, graph theory features, and potential features of virus-drug pairs from the virus integration similarity matrix and the virus-drug association adjacency matrix, and merge them to generate the total virus feature matrix and the total drug feature matrix; The loss function optimization module is used to construct the objective function based on inductive hypergraph learning using the total virus feature matrix and the total drug feature matrix. The objective function is differentiated with respect to the projection parameter matrix to obtain an analytical solution, that is, the optimized projection parameter matrix. The prediction score matrix generation module is used to calculate the prediction scores from the virus perspective and the drug perspective based on the optimized projection parameter matrix, and average them to generate the final virus-drug association prediction score matrix; The prediction module is used to filter out the scores of the target virus row based on the final virus-drug association prediction score matrix, and obtain the final prediction results after sorting.

9. A computer-readable storage medium, characterized in that The medium stores computer instructions, and when a computer reads the computer instructions, the computer executes the antiviral drug screening method based on inductive hypergraph learning according to any one of claims 1 to 7.

Images