Method for generating antiviral functional material composition based on deep generative model

Abstract

The invention discloses a deep generative model-based antiviral functional material composition generation method, which comprises the following steps of: constructing a generative adversarial network which is a diversity-sensitive conditional generative adversarial network and comprises a generator and a discriminator; constructing a diversity sensitive loss function, performing distributed iterative training on the generative adversarial network based on the diversity sensitive loss function and a federated learning architecture to obtain an antiviral functional material composition generation model, and dynamically adjusting weighting coefficients of a generator and a discriminator in each round of training; and obtaining the antiviral functional material composition based on the antiviral functional material composition generation model. The method provided by the invention is suitable for high efficiency, diversity and privacy protection requirements on molecular generation in the development process of antiviral drugs, and can significantly improve the efficiency and quality of drug research and development.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and in particular relates to a method for generating an antiviral functional material composition based on a deep generative model. Background Art

[0002] With the intensification of the global problem of viral infections, the research and development of antiviral drugs has become a core task in the biomedical field. However, the traditional drug development process faces problems such as long cycles, high costs, and low data utilization efficiency. In recent years, artificial intelligence technology has been widely used in the field of drug research and development. Among them, molecular generation technology based on generative adversarial networks has attracted much attention due to its ability to efficiently generate potential drug molecules. However, existing technologies still have many limitations in the process of generating antiviral functional materials. On the one hand, centralized data training methods do not provide sufficient privacy protection. The large amount of molecular data involved in drug research and development is usually highly sensitive, and centralized storage and processing may lead to data leakage risks. On the other hand, traditional molecular generation technology has limited performance in terms of the diversity of generated results and the optimization of target properties. The generated molecules often lack the diversity of chemical space, which makes it difficult to meet the requirements of antiviral drugs for broad-spectrum effects and multi-property balance. In addition, the single-property optimization mode commonly used in multi-objective optimization methods leads to biased generation results and fails to take into account the balance between drug activity, toxicity and other key properties. These problems significantly limit the application effect and efficiency of existing molecular generation technology in actual drug research and development. In summary, the existing technology has the following shortcomings:

[0003] 1. Existing centralized model training methods usually require the collection and centralized storage of data, which easily leads to the risk of data leakage in the field of drug research and development and is difficult to meet privacy protection requirements.

[0004] 2. Traditional generative adversarial networks are difficult to effectively control the structural diversity of generated molecules and are prone to produce highly similar molecules, resulting in insufficient coverage of the chemical space and limiting the design space of antiviral functional materials.

[0005] 3. Existing generation methods usually optimize a single property and ignore the balance between multi-dimensional properties, resulting in the generated molecules being difficult to meet various performance requirements in practical applications.

[0006] 4. Existing methods lack targeted design in the optimization of generators and discriminators, resulting in low training efficiency, low authenticity and property consistency of generated results, and difficulty in meeting the research and development needs of antiviral functional materials.

[0007] Therefore, there is an urgent need to propose a solution to improve the efficiency, quality and safety of antiviral drug development by addressing issues such as privacy protection, generation diversity and multi-attribute optimization in the generation process of antiviral functional materials. Summary of the Invention

[0008] In order to solve the above technical problems, the present invention proposes a method for generating an antiviral functional material composition based on a deep generative model to solve the problems existing in the above-mentioned prior art.

[0009] To achieve the above objectives, the present invention provides a method for generating an antiviral functional material composition based on a deep generative model, comprising:

[0010] Constructing a generative adversarial network, wherein the generative adversarial network is a diversity-sensitive conditional generative adversarial network, including a generator and a discriminator;

[0011] Constructing a diversity-sensitive loss function, and performing distributed iterative training on the generative adversarial network based on the diversity-sensitive loss function and the federated learning architecture to obtain a generation model of an antiviral functional material composition, wherein the weighted coefficients of the generator and the discriminator are dynamically adjusted in each round of training;

[0012] An antiviral functional material composition is obtained based on an antiviral functional material composition generation model.

[0013] Optionally, the generator receives a latent vector and target drug attribute inputs to generate a molecular sample that meets the attribute constraints; the discriminator performs a multi-dimensional evaluation of the authenticity, target attribute consistency, and structural similarity of the generated molecules.

[0014] Optionally, the generator generation process is expressed as follows:

[0015]

[0016] Where, is the generated drug molecule sample, z c is a random vector in the latent space, R ac is the drug property condition, is the weight of the generator, is the mapping function of the generator network to the potential vector,

[0017] is the mapping function of the generator to the drug attributes, f G () is the activation function.

[0018] Optionally, the diversity-sensitive loss function is:

[0019]

[0020] Where, L div is the diversity-sensitive loss function, λ d is the diversity sensitive weight, δ d To adjust the weighted coefficient of the influence of drug attribute differences, and are the attribute values of the i-th and i-1-th generated molecular samples respectively; The structural difference measure between the i-th and j-th generated molecular samples is expressed as: in, is the characteristic vector of the molecular structure, and ||||2 is the L2 norm.

[0021] Optionally, when performing distributed iterative training on the generative adversarial network, the loss function of the generator is:

[0022]

[0023] Where, L G is the objective function of the generator, Denotes the expectation of the latent vector and drug attributes, D c is the authenticity score of the discriminator output, L div is the diversity loss term, λ d is the diversity sensitive weight, G c () is a generator function.

[0024] Optionally, when performing distributed iterative training on the generative adversarial network, the loss function of the discriminator is:

[0025]

[0026] Where, L D is the objective function of the discriminator, X c For real drug molecules, To generate molecules, f mol () is the molecular structure characteristic vector, represents expectation, λ sim is the weighting coefficient to adjust the influence of molecular similarity.

[0027] Optionally, the training process of the generator further includes multi-attribute optimization, which performs weighted constraints on multiple drug attributes. The calculation method of the multi-attribute optimization is expressed as:

[0028]

[0029] Where, L attr is the multi-attribute loss function, R m,ac is the true value of the mth target attribute, is the predicted value of the generated ith molecule on the mth attribute, is the predicted value of the generated molecule i-1 on the mth attribute, λ1 is the weighting coefficient of the corresponding attribute, λ3 is the weighting coefficient of the property difference, N is the number of generated samples, M is the number of target conditions, and λ2 is the weighting coefficient for adjusting the molecular structure difference. represents the i-th generated molecule The characteristic of the jth biophysical and chemical properties, P j (X c ) represents the true sample X c Characteristics of the same physical and chemical properties.

[0030] The present invention also provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0031] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0032] The present invention also provides a computer program product, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0033] Compared with the prior art, the present invention has the following advantages and technical effects:

[0034] The present invention provides a method for generating antiviral functional material compositions based on a deep generative model. It constructs a federated learning architecture, holds local data on multiple clients, and achieves model optimization through local generative model training and parameter interaction without exchanging original data, thereby protecting data privacy. The generative adversarial network adopts a diversity-sensitive conditional generative adversarial network. By introducing a diversity loss function and a multi-objective optimization mechanism, it improves the diversity of generated molecular structures while ensuring that the generated molecules achieve an optimal balance in multiple properties such as antiviral activity, toxicity, and solubility. The generator generates molecules that meet the target attribute constraints by inputting latent vectors and target drug attributes. The discriminator evaluates the authenticity of the generated molecules and the consistency of the target attributes, and conducts adversarial training with the generator, ultimately achieving a significant improvement in the quality and attribute control capabilities of the generated molecules. The present invention also adopts a method of dynamically adjusting weighting coefficients to optimize the performance of the generator and discriminator in multiple rounds of iterative training, avoid pattern collapse, and ensure a high match between the diversity of the generated molecules and the target attributes. The method of the present invention is applicable to the requirements of high efficiency, diversity, and privacy protection for molecular generation in the development of antiviral drugs, and can significantly improve the efficiency and quality of drug research and development. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0036] Figure 1 This is a diagram of the model training architecture of an embodiment of the present invention;

[0037] Figure 2 is a flow chart of a method according to an embodiment of the present invention;

[0038] Figure 3 Schematic diagram comparing the antiviral activity success rates of different models according to the embodiments of the present invention. DETAILED DESCRIPTION

[0039] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0040] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0041] Example 1

[0042] This embodiment provides a method for generating an antiviral functional material composition based on a deep generative model, such as Figure 2 As shown, including:

[0043] The antiviral functional material composition is generated using a generative model, specifically a generative adversarial network, and the generative model training architecture adopts a federated learning architecture. The generative model is trained on multiple distributed devices or servers, and the devices or servers hold local data samples without exchanging these data.

[0044] Specifically, the federated learning architecture proposed in this invention is as follows Figure 1 As shown, there are multiple clients, labeled as client 0, client 1, client 2, and client 3, etc. These clients represent independent nodes in the federated learning network, and each client has its own local data;

[0045] Specifically, local data collection of each client comes from the following channels:

[0046] 1. Public chemical databases: such as PubChem, ChEMBL, ZINC, etc. These databases contain a large number of known drug molecules and their related properties, such as molecular structure, activity, toxicity and other information.

[0047] 2. Laboratory data: Molecular data generated during the synthesis and performance testing of antiviral functional materials in the laboratory. These data are obtained through high-throughput screening, molecular experimental synthesis, etc., and are more targeted and potentially valuable.

[0048] 3. Public literature data: including data on antiviral drugs and related materials disclosed through scientific research articles, academic reports and other channels. These data are timely and innovative.

[0049] Data collection uses automated data capture tools. These tools capture relevant data from designated databases and literature based on web crawler technology. By setting crawler rules, they extract structural information of drug molecules (such as SMILES strings, molecular images), relevant pharmacological data (such as EC50 values, inhibition rates, etc.) and experimental conditions.

[0050] In terms of storage format, in order to more efficiently manage data and train models, the present invention can adopt the following storage formats:

[0051] 1. SMILES format: used to store the structural representation of drug molecules. Each molecule is represented by a string of character codes, which is convenient for processing and conversion in the model.

[0052] 2. CSV format: used to store drug experimental results data and its corresponding characteristics, such as activity, toxicity, etc.

[0053] 3. JSON format: used to store metadata of drug data, including basic information, structural description and other related attributes of the drug, which is easy to interact with other systems.

[0054] Furthermore, the collected data is labeled. The labeling categories include:

[0055] 1. Drug activity labeling: The inhibitory activity of the drug against viruses is characterized by experimental data such as EC50 value or IC50 value. The lower the value, the higher the activity.

[0056] 2. Drug toxicity labeling: The toxicity level of the drug, including acute toxicity, long-term toxicity, cytotoxicity, etc., can be based on existing in vitro experimental data or cell experiment results.

[0057] 3. Molecular property annotation: physical and chemical properties related to drug molecules, such as molecular weight, solubility, lipid solubility, polarity, etc.

[0058] 4. Antiviral spectrum labeling: Label the drug's ability to inhibit different virus strains, including its effectiveness against a single virus strain and its broad-spectrum antiviral potential.

[0059] Each client has a local generative model, which is trained on the data of each client. The interaction between the local generative model and the central generative model is the update flow of the generative model parameters. fed ' represents the updated generative model parameters after local training, θ fed represents the parameters received from the central generative model;

[0060] Furthermore, the central generative model aggregates updates from local generative models. The update of the global generative model is a way for the federated learning process to aggregate individual updates to create a new improved global generative model. After the global generative model update is completed, the final trained generative model is obtained.

[0061] In one embodiment, the model parameters θ are generated fed0 to θ fed3 exchanging between the local generative model and the central generative model in a manner that allows the central generative model to aggregate updates and the local generative model to receive new, aggregated parameters;

[0062] Furthermore, through multiple iterations, the local generative model is trained, the updates are sent to the central generative model, a global update is performed, and then the updated parameters are sent back to the local generative model;

[0063] Based on this federated learning architecture, data privacy protection can be achieved. During the training process, the original data is not shared between clients or with the central server, but only the generation model parameters are exchanged for update. That is, the parameters are synchronized to each client through the central generation model. fed ' Assign a value to each client.

[0064] Furthermore, the generation model of each client uses a diversity-sensitive conditional generative adversarial network to generate an anti-virus functional material composition;

[0065] To address the problem of chemical space generation of drug molecules, a diversity-sensitive conditional generative adversarial network (CGN) adopts a diversity loss term in the generation process and combines multi-objective optimization of drug properties to enhance the structural diversity of the generated molecules while maintaining molecular realism. On the basis of traditional CGN, it adopts a diversity-sensitive mechanism and combines the idea of drug multi-attribute optimization to improve the activity, synthesizability and diversity of the generated molecules by optimizing the adversarial loss function of the generator and discriminator.

[0066] Specifically, the training process of the diversity-sensitive conditional generative adversarial network is as follows:

[0067] Initialize the model parameters of the generator and discriminator of the generative adversarial network. The generator accepts the potential vector and the drug target attributes to generate samples. The discriminator evaluates the authenticity of the generated samples and whether the target attributes are satisfied, thereby laying the foundation for subsequent adversarial training, solving the problem of rationality of the network parameters of the generator and discriminator in the initial stage, and ensuring that subsequent training can converge efficiently. The initialization calculation method is expressed as:

[0068]

[0069] In the formula, ~ means obeying a specific distribution, is the weight matrix of the generator at initialization, is the weight matrix of the discriminator at initialization, σ 2 Initialize the variance of the parameters of the generative adversarial network, Indicates that the mean is 0 and the variance is σ 2 Gaussian distribution. Preferably, σ 2 Set to 0.01.

[0070] The generator generates preliminary data based on the latent vector and the target drug properties. By combining the latent vector with the drug property information, it realizes the molecular construction with drug property constraints, solves the technical problem of coupling the chemical molecular structure with the target property, and makes the generated molecules conform to the real molecular distribution while having preliminary property consistency. The calculation method of the generator is expressed as:

[0071]

[0072] Where, is the generated drug molecule sample, z c is a random vector in the latent space, R ac is the drug property condition, is the weight of the generator, is the mapping function of the generator network to the potential vector, is the mapping function of the generator to the drug attributes, f G () is the activation function.

[0073] Furthermore, the calculation method of the generator's mapping function to the drug attributes is represented as:

[0074]

[0075] Where, is the weight matrix associated with the attribute, is the bias term.

[0076] A diversity-sensitive evaluation is performed on the generated drug molecules. The diversity-sensitive loss function measures the differences between molecular structures to avoid generating molecules with too similar structures and enhance the coverage of molecules in the chemical space. It also constrains the problem of excessive similarity in generated molecules and improves the structural diversity and practical value of the generated results. The calculation method of the diversity metric is expressed as:

[0077]

[0078] Where, is the structural difference measure between the i-th and j-th generated molecular samples, is the characteristic vector of the molecular structure, and ||||2 is the L2 norm.

[0079] Furthermore, the calculation method of diversity-sensitive loss is expressed as:

[0080]

[0081] Where, L div is the diversity-sensitive loss function, λ d is the diversity sensitive weight, δ d To adjust the weighted coefficient of the influence of drug attribute differences, and are the property values of the i-th and i-1-th generated molecules respectively. Preferably, λ d Set to 0.2, δ d Set to 0.5.

[0082] The discriminator needs to judge the authenticity of the generated drug molecules and evaluate whether they meet the target drug properties. This provides a discriminant signal for the generator update, screens the authenticity and property compliance of the generated samples, and accurately distinguishes between real data and generated data, thereby improving the generation effect and the accuracy of the discriminator. The calculation method of the discrimination probability is expressed as:

[0083]

[0084] Where, is the true probability of the discriminator output, D c () is the discriminator function, is the molecular structure characteristic vector, R ac is the drug attribute information, α c is the attribute weight coefficient, Sig() is the Sigmoid activation function. Preferably, α c Set to 0.2.

[0085] Through adversarial training, the generator and discriminator are optimized simultaneously. The generator aims to generate realistic molecules that meet the target properties, while the discriminator strives to distinguish the generated molecules from real molecules. The adversarial game between the generation quality and the discriminator's discrimination ability can improve the generator's property control ability and molecular diversity in continuous iteration. The generator's loss function is calculated as follows:

[0086]

[0087] Where, L G is the objective function of the generator, Denotes the expectation of the latent vector and drug attributes, D c is the authenticity score of the discriminator output, L div is the diversity loss term, λ d is the diversity sensitive weight, G c () is a generator function. Preferably, λ d Set to 0.2.

[0088] Furthermore, the loss function of the discriminator is calculated as follows:

[0089]

[0090] Where, L D is the objective function of the discriminator, X c For real drug molecules, To generate molecules, f mol () is the molecular structure characteristic vector, represents expectation, λ sim is a weighted coefficient for adjusting the influence of molecular similarity. Preferably, λ sim Set to 0.2.

[0091] The multi-objective optimization module of drug properties is used to impose weighted constraints on multiple drug properties (such as antiviral activity, toxicity, solubility, etc.) to ensure that the generated molecules can achieve a better balance in multiple properties, solve the problem of biased generation results that may be caused by single attribute optimization, and make the generated results more in line with the actual drug development needs. At the same time, the property difference constraint is used to impose comprehensive constraints on the molecular structure, target properties (such as antiviral activity) and biophysical and chemical properties of the generated samples to ensure that the generated results not only meet the target properties but also have actual physical and chemical feasibility. The calculation method of multi-attribute optimization is expressed as:

[0092]

[0093] Where, L attr is the multi-attribute loss function, R m,ac is the true value of the mth target attribute, is the predicted value of the generated ith molecule on the mth attribute, is the predicted value of the generated molecule i-1 on the mth attribute, λ1 is the weighting coefficient of the corresponding attribute, λ3 is the weighting coefficient of the property difference, N is the number of generated samples, M is the number of target conditions, and λ2 is the weighting coefficient for adjusting the molecular structure difference. represents the i-th generated molecule Characteristics of the jth biophysical and chemical properties (such as solubility, stability, etc.), P j (X c ) represents the true sample X c Characteristics of the same physical and chemical properties.

[0094] Furthermore, in each round of training, the weight coefficients are adaptively adjusted as follows:

[0095]

[0096] Where β1 is the first dynamic adjustment coefficient, β2 is the second dynamic adjustment coefficient, and β3 is the third dynamic adjustment coefficient. and f G (X c ,R ac ) represent the target attribute adaptability metrics of generated samples and real samples respectively. The generator can adjust the focus at different training stages to ensure that the generated drug molecules have biophysical feasibility while meeting the antiviral activity.

[0097] By continuously iterating the parameter updates of the generator and discriminator, the loss function gradually converges. The generator gradually masters how to generate drug molecules that conform to the real data distribution and have the advantages of multiple target attributes. The discriminator can also more accurately distinguish between authenticity and falsehood, avoiding the problems of model collapse or pattern collapse. Through multiple rounds of confrontation and attribute optimization iterations, we can finally obtain a generative model that has both diversity and target attribute matching.

[0098] This example compares the success rate performance of four molecular generation algorithms in the antiviral activity screening scenario, including the diversity-sensitive conditional generative adversarial network, conditional generative adversarial network, Wasserstein generative adversarial network with gradient penalty, and variational autoencoder proposed in the present invention.

[0099] Figure 3The horizontal axis sets four progressively higher thresholds for antiviral activity (labeled as >0.7, >0.75, >0.8, and >0.85), representing the minimum activity levels required for a molecule to achieve. The vertical axis shows the proportion of molecules generated at these thresholds that meet these thresholds, reflecting the model's ability to generate highly active molecules. The colored bars correspond to the four algorithms, with the red bar representing our technology and the blue, orange, and green bars representing the other three comparison algorithms, respectively.

[0100] By comparing the success rates under different activity thresholds, the comprehensive advantages of the present invention in key drug discovery indicators were verified. Four gradually increasing antiviral activity thresholds were set on the horizontal axis, and the vertical axis represented the proportion of molecules reaching the threshold. The success rate curve of the present invention showed a gentle downward trend and still maintained a significant advantage in the high threshold range. The success rates of traditional generative adversarial networks and variational autoencoders decayed sharply with the increase of the threshold, indicating the synergistic effect of the attribute coupling generation mechanism and the diversity-sensitive constraints. The former ensures that the generated molecules accurately match the target attributes, and the latter increases the probability of discovering highly active molecules by expanding the search range of the chemical space.

[0101] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for generating an antiviral functional material composition based on a deep generative model, characterized in that: The following steps are involved: Constructing a generative adversarial network, wherein the generative adversarial network is a diversity-sensitive conditional generative adversarial network, including a generator and a discriminator; Constructing a diversity-sensitive loss function, and performing distributed iterative training on the generative adversarial network based on the diversity-sensitive loss function and the federated learning architecture to obtain a generation model of an antiviral functional material composition, wherein the weighted coefficients of the generator and the discriminator are dynamically adjusted in each round of training; An antiviral functional material composition is obtained based on an antiviral functional material composition generation model.

2. The method for generating an antiviral functional material composition based on a deep generative model according to claim 1, characterized in that: The generator receives the latent vector and the target drug attribute input and generates a molecular sample that meets the attribute constraints; the discriminator performs a multi-dimensional evaluation on the authenticity, target attribute consistency and structural similarity of the generated molecules.

3. The method for generating an antiviral functional material composition based on a deep generative model according to claim 2, wherein: The generator generation process is expressed as follows: Where, is the generated drug molecule sample, z c is a random vector in the latent space, R ac is the drug property condition, is the weight of the generator, is the mapping function of the generator network to the potential vector, is the mapping function of the generator to the drug attributes, f G () is the activation function.

4. The method for generating an antiviral functional material composition based on a deep generative model according to claim 3, characterized in that: The diversity-sensitive loss function is: Where, L div is the diversity-sensitive loss function, λ d is the diversity sensitive weight, δ d To adjust the weighted coefficient of the influence of drug attribute differences, and are the attribute values of the i-th and i-1-th generated molecular samples respectively; The structural difference measure between the i-th and j-th generated molecular samples is expressed as: in, is the characteristic vector of the molecular structure, and || ||2 is the L2 norm.

5. The method for generating an antiviral functional material composition based on a deep generative model according to claim 4, characterized in that: When the generative adversarial network is trained iteratively, the loss function of the generator is: Where, L G is the objective function of the generator, Denotes the expectation of the latent vector and drug attributes, D c is the authenticity score of the discriminator output, L div is the diversity loss term, λ d is the diversity sensitive weight, G c () is a generator function.

6. The method for generating an antiviral functional material composition based on a deep generative model according to claim 5, characterized in that: When the generative adversarial network is trained iteratively, the loss function of the discriminator is: Where, L D is the objective function of the discriminator, X c For real drug molecules, To generate molecules, f mol () is the molecular structure characteristic vector, represents expectation, λ sim is the weighting coefficient to adjust the influence of molecular similarity.

7. The method for generating an antiviral functional material composition based on a deep generative model according to claim 6, characterized in that: The training process of the generator also includes multi-attribute optimization, which performs weighted constraints on multiple drug attributes. The calculation method of the multi-attribute optimization is expressed as: Where, L attr is the multi-attribute loss function, R m,ac is the true value of the mth target attribute, is the predicted value of the generated ith molecule on the mth attribute, is the predicted value of the generated molecule i-1 on the mth attribute, λ1 is the weighting coefficient of the corresponding attribute, λ3 is the weighting coefficient of the property difference, N is the number of generated samples, M is the number of target conditions, and λ2 is the weighting coefficient for adjusting the molecular structure difference. represents the i-th generated molecule The characteristic of the jth biophysical and chemical properties, P j (X c ) represents the true sample X c Characteristics of the same physical and chemical properties.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images