Method and device for constructing antigen variation prediction model and predicting influenza virus antigen variation

Abstract

The invention relates to a method and a device for constructing an antigen variation prediction model and predicting influenza virus antigen variation. The method for constructing the antigen variation prediction model comprises the following steps: acquiring an antigen similarity score and an antigen distance of a known target virus pair; constructing a preset function based on the antigen similarity score and the antigen distance of the known target virus pair, and performing adaptive Fourier decomposition on the preset function by using a maximum selection principle to obtain an important antigen point set, an orthogonal system and an important antigen point similarity score set obtained by multiple times of maximum selection; and constructing an antigen variation prediction model based on the important antigen point set, the orthogonal system, the important antigen point similarity score set and the antigen distance of the known target virus pair. According to the method, the optimization principle and the reordering thought of adaptive Fourier decomposition are combined, the prediction model is established with less linear combinations, the method has wide applicability and data selection flexibility, and the model precision and convergence are improved.

Description

Technical Field

[0001] The present application relates to the field of influenza virus technology, and in particular to a method and device for constructing an antigen variation prediction model and predicting influenza virus antigen variation. Background Art

[0002] Influenza is a contagious respiratory disease caused by influenza viruses that infect the nose, throat and lungs. In the past 100 years, influenza viruses have caused four pandemics, namely H1N1 in 1918, H2N2 in 1957, H3N2 in 1968 and H1N1 in 2009. So far, according to WHO reports, there are about 1 billion cases of seasonal influenza each year, including 3-5 million severe cases, and influenza causes 290,000 to 650,000 deaths from respiratory diseases each year.

[0003] The World Health Organization (WHO) reviews the world epidemiological situation twice a year and recommends new vaccine strains when necessary based on existing evidence. The current gold standard for WHO antigen detection is mainly through laboratory hemagglutination inhibition (HI) tests. However, since this test is very time-consuming and labor-intensive, the HI test for emerging virus strains lags far behind the rapid accumulation of new virus strains spreading globally. The continuous genetic changes of influenza viruses are mainly due to point mutations and recombination events in the viral genome, resulting in the continuous emergence of new virus variants. This phenomenon is called "antigenic drift." In these viral genomes, HA and NA are the main surface antigen proteins. Compared with other genes, the HA1 subunit in the HA protein mutates the fastest. The HA1 subunit contains the antigenic determinant region and the key binding site for the virus to attach to the host cell. Its mutation affects the antigenicity and receptor binding affinity of the virus, which is also the main reason for the continuous changes in antigens.

[0004] The viral genome is routinely detected through high-throughput sequencing, so sequence alignment provides extremely valuable information on the antigenic variation of influenza strains. The influenza prediction model established on this basis will help monitor the emergence of new variants, reduce the detection time of new antigen types, and improve the efficiency of vaccine development and preparation.

[0005] The traditional method of establishing influenza prediction models is mainly based on one-dimensional information to deal with changes in amino acid sequences, while ignoring the nonlinear relationship between the heterotopic effects of different amino acid changes in the protein sequence and the spatial position of amino acids. In addition, the above models usually only consider limited key sites in the amino acid sequence. If the influenza virus mutates and the mutation site is not at the key site set by the original prediction model, the prediction ability of the above model will be affected. Summary of the invention

[0006] Based on this, it is necessary to provide a method and device for constructing an antigen variation prediction model and predicting influenza virus antigen variation to improve the accuracy of predicting antigen variation.

[0007] The first object of the present application is to provide a method for constructing an antigen variation prediction model, comprising:

[0008] Obtain antigenic similarity scores and antigenic distances for known target virus pairs;

[0009] A preset function is constructed based on the antigenic similarity score and antigenic distance of known target virus pairs, and the preset function is adaptively Fourier decomposed using the maximum selection principle to obtain a set of important antigenic sites, an orthogonal system, and a set of similarity scores of important antigenic sites obtained by multiple maximum selections;

[0010] An antigenic variation prediction model was constructed based on a set of important antigenic sites, an orthogonal system, a set of similarity scores for important antigenic sites, and antigenic distances of known target virus pairs.

[0011] The present application combines the optimization principle and reordering idea of ​​adaptive Fourier decomposition to achieve the establishment of a prediction model with fewer linear combinations, has wide applicability and flexibility in data selection, and improves the accuracy and convergence of the model.

[0012] In one embodiment, obtaining the antigenic similarity score and antigenic distance of the known target virus pair includes:

[0013] Acquire multiple antigen protein sequences of known target viruses, and obtain antigen similarity scores of known target virus pairs based on similarity scores of each amino acid site of the multiple antigen protein sequences;

[0014] A hemagglutination inhibition titer set of known target viruses is obtained, wherein the hemagglutination inhibition titer set of known target viruses includes hemagglutination inhibition titer data of known target viruses and different sera, and the antigenic distance of the known target virus pair is determined based on the hemagglutination inhibition titer set of known target viruses.

[0015] In one embodiment, the preset function is subjected to adaptive Fourier decomposition using the maximum selection principle to obtain a set of important antigenic sites, an orthogonal system, and a set of similarity scores of important antigenic sites obtained by multiple maximum selections, including:

[0016] The value of a preset function is calculated using the antigenic similarity score of the known target virus pair and the antigenic distance of the known target virus pair, and the important antigenic sites, orthogonal bases, and similarity scores of important antigenic sites that maximize the value of the preset function are screened;

[0017] After obtaining the important antigenic sites, orthogonal bases and similarity scores of the important antigenic sites, reconstructing the preset function based on the orthogonal bases obtained each time, and selecting the important antigenic sites, orthogonal bases and similarity scores of the important antigenic sites that maximize the reconstructed preset function value;

[0018] When the maximum value of the reconstructed preset function is less than the preset threshold, the screening is stopped to obtain an important antigenic site set, an orthogonal system and an important antigenic site similarity score set based on multiple maximum selections.

[0019] In one embodiment, constructing an antigenic variation prediction model based on a set of important antigenic sites, an orthogonal system, a set of similarity scores for important antigenic sites, and antigenic distances of known target virus pairs comprises:

[0020] The conversion coefficients were calculated based on the orthogonal system and the similarity scores of important antigenic sites;

[0021] Determine training parameters based on the orthogonal system and antigenic distances of known target virus pairs;

[0022] The model parameters are determined based on the conversion coefficients and the training parameters.

[0023] The second object of the present application is to provide a method for predicting influenza virus antigen variation, comprising:

[0024] Obtaining antigenic similarity scores between the target virus to be analyzed and the known target virus;

[0025] The antigenic similarity scores of the target virus to be analyzed and the known target virus are input into the antigenic variation prediction model constructed by the above method to obtain the predicted antigenic distance between the target virus to be analyzed and the known target virus.

[0026] The third object of the present application is to provide a device for constructing an antigen variation prediction model, comprising:

[0027] The first data acquisition module is used to obtain the antigen similarity score and antigen distance of the known target virus pair;

[0028] Maximum selection module: used to construct a preset function based on the antigenic similarity score and antigenic distance of known target virus pairs, and perform adaptive Fourier decomposition of the preset function using the maximum selection principle to obtain a set of important antigenic sites, an orthogonal system, and a set of similarity scores of important antigenic sites obtained by multiple maximum selections;

[0029] Model parameter calculation module: used to construct an antigen variation prediction model based on a set of important antigenic sites, an orthogonal system, a set of similarity scores for important antigenic sites, and antigenic distances of known target virus pairs.

[0030] The fourth object of the present application is to provide a device for predicting influenza virus antigen variation, comprising:

[0031] The second data acquisition module is used to obtain the antigenic similarity score between the target virus to be analyzed and the known target virus;

[0032] Prediction module: used to input the antigenic similarity scores of the target virus to be analyzed and the known target virus into the antigenic variation prediction model constructed by the above method to obtain the predicted antigenic distance between the target virus to be analyzed and the known target virus.

[0033] The fifth objective of the present application is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0034] The sixth object of the present application is to provide a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when the computer program is executed by a processor.

[0035] The seventh objective of the present application is to provide a computer program product, comprising a computer program, characterized in that the steps of the above method are implemented when the computer program is executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 A flowchart of a method for predicting influenza virus antigen variation provided in one embodiment of the present application;

[0038] Figure 2 A flowchart of a method for constructing an antigen variation prediction model provided in one embodiment of the present application;

[0039] Figure 3 A structural block diagram of a device for predicting influenza virus antigen variation provided in one embodiment of the present application;

[0040] Figure 4 A structural block diagram of a device for constructing an antigen variation prediction model provided in one embodiment of the present application;

[0041] Figure 5 An internal structure diagram of a computer device provided in one embodiment of the present application;

[0042] Figure 6 A schematic diagram of a method for predicting influenza virus antigen variation provided in Example 1 of the present application;

[0043] Figure 7 A schematic diagram of a method for constructing a Matrix-POAFD model provided in Example 1 of the present application;

[0044] Figure 8 The performance evaluation results of different models based on the influenza virus training set provided in Example 2 of the present application;

[0045] Fig. 9 The performance evaluation results of different models based on the influenza virus validation set provided in Example 2 of the present application;

[0046] Fig.10 The performance evaluation results of different models based on the H5N1 virus training set provided in Example 2 of the present application;

[0047] Fig.11 The performance evaluation results of different models based on the H5N1 virus validation set provided in Example 2 of the present application;

[0048] Fig.12 A bar graph showing the important residues provided in Example 3 of the present application at different antigenic epitopes;

[0049] Fig.13 A bar graph showing the important residues provided in Example 4 of the present application at different antigenic epitopes;

[0050] Fig.14 A bar graph showing the important residues provided in Example 5 of the present application at different antigenic epitopes;

[0051] Fig.15 A network diagram showing the two-site interaction provided in Example 5 of the present application. DETAILED DESCRIPTION

[0052] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present application are described in detail below. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0053] In this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0055] In order to solve at least one of the above problems, the first aspect of the present application provides a method for constructing an antigen variation prediction model, comprising:

[0056] S10: Obtaining antigenic similarity scores and antigenic distances of known target virus pairs;

[0057] Specifically, the target virus refers to a virus with similar antigenicity, such as influenza B virus, H1N1 influenza A virus, etc. The antigenic similarity score of the known target virus pair refers to the antigenic sequence similarity score between different known target viruses.

[0058] In some embodiments, the antigen of the target virus refers to hemagglutinin (HA) on the surface of the virus that can cause agglutination of red blood cells of humans or certain mammals.

[0059] In some embodiments, the known target virus antigen sequence can be obtained from databases, literature, and any public information. Specifically, for the antigen sequence of influenza B virus, the HA sequence of influenza B virus over the years can be downloaded from Genbank, Influenza Virus Resource, Influenza Research Database, and GISAID, respectively. According to the different functions of the HA sequence, the signal peptide in the HA1 fragment is selected as the starting position, and the HA1 and HA2 cleavage sites are selected as the ending position, which are uniformly used as the intercepted HA sequence, i.e., the target virus antigen sequence.

[0060] In some embodiments, the antigenic similarity score of the known target virus pair can be obtained by scoring the amino acid similarity of the same position of the known target virus antigen pair according to the conventional amino acid similarity scoring rules in the art. In some specific embodiments, it can refer to scoring the aligned HA sequences from different target viruses using the pattern induced multiple sequence alignment (PIMA coding) function to generate a similarity score matrix X. In a similarity score matrix X generated in the PIMA function, each element X [i][j] represents the similarity score of each amino acid site between the ith HA sequence and the jth HA sequence. Exemplarily, X can be an m×n matrix, also called a core matrix, where n represents the number of amino acid sites in the antigen sequence, and m represents the number of pairs of known target virus pairs.

[0061] The antigenic distance of the known target virus pair refers to the antigenic distance between different known target viruses. The antigenic distance of the known target virus pair is obtained based on the hemagglutination inhibition test (HI) data of different target viruses and each serum. Exemplarily, the hemagglutination inhibition test (HI) data of different target viruses and each serum are downloaded from official websites such as published articles, WHO, various national and regional disease control centers, and the US FDA.

[0062] Furthermore, the matrix of HI reaction values ​​of different target viruses and each serum was supplemented with missing data and low reaction values ​​using the low-rank matrix completion method, and the antigenic distance vector Y between each virus was calculated according to the following formula:

[0063]

[0064] Exemplarily, Y is a column matrix (vector) containing m components, called target vectors, where m represents the number of pairs of known target virus pairs.

[0065] S20: constructing a preset function based on the antigenic similarity score and antigenic distance of the known target virus pair, and performing adaptive Fourier decomposition on the preset function using the maximum selection principle to obtain an important antigenic site set, an orthogonal system, and an important antigenic site similarity score set obtained by multiple maximum selections;

[0066] Specifically, adaptive Fourier decomposition (AFD) is an extension of traditional Fourier decomposition, which generates an adaptive basis through matching pursuit and achieves fast energy convergence in the decomposition process.

[0067] Let u and v be any two n-column vectors, ‖u‖ = 1. Use Q v represents the Gram-Schmidt orthogonalization of v: Q u (v)=v–<v,u> u;

[0068] Using the projection operator: For vector A collection of Indicates that v is Orthogonal projection in. For l vectors u1,…,u l , are not necessarily orthogonal to each other, it can be proved that:

[0069]

[0070] Therefore, an inner product function is constructed based on the antigen similarity score and antigen distance of the known target virus pairs. When performing adaptive Fourier decomposition, important antigenic sites are selected at each maximal step to avoid overfitting, which can well characterize the relationship between the similarity score of important antigenic sites and the antigen distance, that is, the target vector.

[0071] For example, according to Construct a preset function, where k = 1, ..., n; further, when performing the maximum selection for the first time, use the following formula to select Thus, the important antigenic sites obtained by the first step of maximum selection are determined according to k1, and at the same time Thus we obtain the orthogonal basis maximally selected in the first step.

[0072] In some embodiments, the preset function is subjected to adaptive Fourier decomposition using the maximum selection principle to obtain a set of important antigenic sites, an orthogonal system, and a set of similarity scores of important antigenic sites obtained by multiple maximum selections, including:

[0073] The value of a preset function is calculated using the antigenic similarity score of the known target virus pair and the antigenic distance of the known target virus pair, and the antigenic site, orthogonal base, and antigenic site similarity score that maximize the value of the preset function are selected;

[0074] After obtaining the important antigenic sites, orthogonal bases and similarity scores of the important antigenic sites, reconstructing the preset function based on the orthogonal bases obtained each time, and selecting the important antigenic sites, orthogonal bases and similarity scores of the important antigenic sites that maximize the reconstructed preset function value;

[0075] When the maximum value of the reconstructed preset function is less than the preset threshold, the screening is stopped to obtain an important antigenic site set, an orthogonal system and an important antigenic site similarity score set based on multiple maximum selections.

[0076] Specifically, each time an important antigenic site k is obtained l Then, the orthogonal basis u obtained by the lth maximum selection is used l right Perform Gram-Schmidt pre-orthogonalization: Thus, the inner product function is reconstructed using the orthogonalized results of the antigen similarity scores of the known target virus pairs and the antigen distance, and then the l+1th maximum selection is performed based on the reconstructed inner product function to obtain the antigenic site, orthogonal basis and antigenic site similarity score that maximizes the reconstructed inner product function value. Among them, the pre-orthogonalization process can eliminate redundancy and noise in the data, allowing the model to focus more on extracting key features and relationships.

[0077] For example, when performing the second maximum selection, the orthogonal basis u1 is used to select v k After Gram-Schmidt orthogonalization, we get remember Further, according to the following formula, select Naturally, k2≠k1. If the maximum value of the above formula is less than the preset threshold, the maximum selection stops at this step. Otherwise, the important antigenic sites obtained by the second step maximum selection are determined according to k2, and at the same time,

[0078] When performing the third maximum selection, use the orthogonal basis u2 to After Gram-Schmidt orthogonalization, we get The next maximum selection is performed to obtain k3 and u3, and so on, and finally the important antigen site k is obtained. l =(k1...,k n1 ), orthogonal system u l =(u1...,u n1 ), similarity score of important antigenic sites in,

[0079] This application adopts the maximum selection principle in the sense of Euclidean norm, so that the model is not constrained by outliers during training, and the series representation of the model has a fast convergence effect. It is a sparse representation prediction method with stability.

[0080] S30: Construct an antigenic variation prediction model based on the set of important antigenic sites, the orthogonal system, the similarity score set of important antigenic sites, and the antigenic distance of known target virus pairs.

[0081] In some embodiments, constructing an antigen variation prediction model based on an important antigenic site set, an orthogonal system, an important antigenic site similarity score set, and an antigenic distance of a known target virus pair comprises:

[0082] The conversion coefficients were calculated based on the orthogonal system and the similarity scores of important antigenic sites;

[0083] Determine training parameters based on the orthogonal system and antigenic distances of known target virus pairs;

[0084] The model parameters are determined based on the conversion coefficients and the training parameters.

[0085] Specifically, use Represented by a column vector The m×n1 matrix is ​​formed in the order listed, The conversion coefficient is represented by Λ, which means and The transformation matrix is

[0086] According to Y and u l The relationship between the training parameters can be obtained remember

[0087] Further, according to and relationship, and and , perform feature extraction based on the training set to obtain a new subset matrix, and multiply the subset matrix with the parameter matrix to obtain a nonlinear Matrix-POAFD model: Among them, W is based on The representative coefficient vector of .

[0088] In summary, the model building method of the present application adopts an adaptive regression algorithm, which can perform adaptive learning according to the characteristics of the input data, so that the model can better handle the nonlinear relationship in antigen variation, thereby improving the prediction accuracy, while avoiding the problem of insufficient adaptability of traditional models to complex data. Secondly, the model building method of the present application also introduces a pre-orthogonalization processing algorithm, which can eliminate redundancy and noise in the data by rearranging and optimizing data input, so that the model can focus more on extracting key features and relationships, further improving the performance of the model.

[0089] In addition, the model of the present application utilizes a sparse representation method, which establishes a prediction model by using fewer linear combinations of important antigenic sites, which can effectively reduce the complexity of the model and improve the prediction speed, while avoiding the problems of overfitting and insufficient generalization ability.

[0090] The model construction method of the present application obtains important amino acid sites and accurate predicted antigenic distances through the combination of the above steps, successfully solving the problem of low accuracy in predicting antigenic variations, providing strong support for research in fields such as bioinformatics and vaccine design, and helping to better understand and respond to the challenges brought about by viral variations.

[0091] The second aspect of the present application provides a method for predicting influenza virus antigen variation, such as Figure 2 As shown, including:

[0092] S100: Obtaining the antigenic similarity score between the target virus to be analyzed and the known target virus;

[0093] S200: Input the antigenic similarity score between the target virus to be analyzed and the known target virus into the antigenic variation prediction model to obtain the predicted antigenic distance between the target virus to be analyzed and the known target virus.

[0094] Specifically, when a new X (HA sequence matrix) is obtained, according to the important antigenic site k l =(k1...,k n1 ) to obtain the corresponding important antigenic site similarity score set The above model can be used to obtain a similarity score set based on important antigenic sites. The predicted target vector Y is the antigenic distance between the new target virus and the known target virus.

[0095] The third aspect of the present application provides a device for constructing an antigen variation prediction model, such as Figure 3 As shown, including:

[0096] The first data acquisition module 10 is used to obtain the antigen similarity score and antigen distance of the known target virus pair;

[0097] Maximum selection module 20: used to construct a preset function based on the antigen similarity score and antigen distance of the known target virus pair, and perform adaptive Fourier decomposition of the preset function using the maximum selection principle to obtain an important antigen site set, an orthogonal system and an important antigen site similarity score set obtained by multiple maximum selections;

[0098] Model building module 30: used to build an antigen variation prediction model based on a set of important antigenic sites, an orthogonal system, a set of similarity scores of important antigenic sites, and antigenic distances of known target virus pairs.

[0099] The fourth aspect of the present application provides a device for predicting influenza virus antigen variation, such as Figure 4 As shown, including:

[0100] The second data acquisition module 100 is used to obtain the antigenic similarity score between the target virus to be analyzed and the known target virus;

[0101] Prediction module 200: used to input the antigenic similarity scores between the target virus to be analyzed and the known target virus into the antigenic variation prediction model constructed by the above method to obtain the predicted antigenic distance between the target virus to be analyzed and the known target virus.

[0102] The specific limitation of the device for predicting influenza virus antigen variation and / or building an antigen variation prediction model can be found in the above for the limitation of the method for predicting influenza virus antigen variation and / or building an antigen variation prediction model, which will not be repeated here. Each module in the device for predicting influenza virus antigen variation and / or building an antigen variation prediction model can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor calls and executes the operations corresponding to the above modules.

[0103] In some embodiments, a computer device is provided, which may be a server 104 or a terminal 102, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory and a communication interface connected via a system bus. When the computer device is a terminal, it also includes a display screen and an input device connected to the system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting influenza virus antigen variation and / or constructing an antigen variation prediction model is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0104] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0105] The present application also provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for predicting influenza virus antigen variation and / or constructing an antigen variation prediction model are implemented.

[0106] The present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned method for predicting influenza virus antigen variation and / or constructing an antigen variation prediction model.

[0107] The present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the method for predicting influenza virus antigen variation and / or building an antigen variation prediction model are implemented.

[0108] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0109] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0110] The implementation scheme of the present application will be described in detail below with reference to examples.

[0111] Example 1

[0112] This embodiment provides a Matrix-POAFD model for establishing an antigen variation prediction model. The model combines the optimization principle of AFD with the idea of ​​Cholesky decomposition and reordering. It has wide applicability and flexibility in data set selection. Its adaptability is reflected in the selection of model parameters, and strives to establish a prediction model with fewer linear combinations to achieve high accuracy and fast convergence. The main processing flow of Matrix-POAFD is as follows Figure 6 As shown:

[0113] The processing flow is divided into two processes: training and prediction. In the training phase, training samples are first obtained and then preprocessed. Next, the Matrix-POAFD algorithm is trained using these preprocessed samples. During this process, the relevant information of the model will be saved for subsequent use. After the training, the performance of the model is evaluated to ensure its accuracy and reliability. By using the trained Matrix-POAFD model, these new sequences can be analyzed and calculated to obtain the antigenic distance between different sequences and the characteristic amino acid positions.

[0114] Taking influenza B virus as an example, the specific steps of the Matrix-POAFD model training process are as follows:

[0115] 1. Collection of flub HA sequences and serological data

[0116] (1) Collection of HA sequences

[0117] Download the HA gene sequences of influenza viruses from Genbank, Influenza Virus Resource, Influenza Research Database and GISAID. According to the different functions of HA sequences, the signal peptide in the HA1 segment is selected as the starting position, and the HA1 and HA2 cleavage sites are selected as the ending positions, which are uniformly used as the sequence fragments intercepted by HA1.

[0118] (2) Serological data collection

[0119] Download official HI experimental data from published articles, WHO, various national and regional disease control centers, US FDA and other official websites.

[0120] 2. Data preprocessing

[0121] (1) The aligned HA sequences are evaluated using the pattern-induced multiple sequence alignment (PIMA coding) function to generate a similarity score matrix X. The effect of the HA sequence preprocessing method in the prediction model is finally determined after evaluation. In the PIMA function, a similarity score matrix X is generated based on the similarity of amino acids, where each element X[i][j] represents the similarity score between the i-th HA sequence and the j-th HA sequence. For example, the similarity score from one amino acid (e.g., "V") to another amino acid (e.g., "L") is calculated as 6 minus the base of the most recent ancestor of the two amino acids (e.g., "c"). Therefore, the similarity score between "V" and "L" is 4, which is 6 minus 2.

[0122] (2) Calculation of antigenic distance between HI data and viruses. The matrix of different viruses and serum HI reaction values ​​is first supplemented with missing data and low reaction values ​​using the low-rank matrix completion method. The antigenic distance vector Y between each virus is calculated according to the following formula:

[0123]

[0124] 3. Model training (1) Establishing the Matrix-POAFD methodology Assume that there is a pair of matrices (X, Y), where X is an m×n matrix, also called a kernel matrix, and Y is a column matrix (vector) containing m components, called the target vector. Let v k , k=1,…,n represents the column vector of X, let V=(v k ). The purpose is to find v k A reordered subset of and the row vector Make

[0125] ‖Y-WV t ‖

[0126] Minimum, where V t is the transpose of V.

[0127] If m≤Rank(V), then at most for some n1≤m, the above model will reach a minimum of zero. To summarize in terms of operator theory is to find the generalized inverse of Y in the domain of the kernel matrix operator.

[0128] Let u and v be any two n-column vectors, ‖u‖ = 1. Use Q v represents the Gram-Schmidt orthogonalization of v:

[0129] Q u (v)=v–<v,u> u;

[0130] Using the projection operator: For vector A collection of Indicates that v is Orthographic projection in .

[0131] For l vectors u1,…,u l , are not necessarily orthogonal to each other, it can be proved

[0132]

[0133] Select according to the following formula

[0134] make as well as remember Next, select according to the following formula

[0135] Naturally, k2≠k1. If the maximum value of the above formula is equal to zero, then the maximum selection stops at this step. Otherwise, let and By analogy, we finally get a standard orthogonal system There are two cases: all Both Orthogonal, or Y is perpendicular to all Orthogonal. In this case, ‖Y-Projυ(Y)‖ cannot be reduced further.

[0136] There are two possibilities here:

[0137] ‖Y-Projυ(Y)‖>0 and ‖Y-Projυ(Y)‖=0;

[0138] n1=n and ‖Y-Projυ(Y)‖>0.

[0139] In simple terms, the vector capacity in V is exhausted before approaching Y. Case (ii) usually occurs when m is much larger than n. If the data in X follows a spatial distribution, case (i) may occur, with m much larger than n.

[0140] In short, there are

[0141]

[0142] use Represented by a column vector An m×n1 matrix is ​​formed in the order listed, and Let Λ be the transformation matrix between two bases and in is orthogonal.

[0143] Therefore, the following Matrix-POAFD model is established:

[0144] Where W is based on The representative coefficient vector of .

[0145] In addition, for the HA matrix sparse problem, the ε-regularization method is used, that is, the ε-regularization matrix is ​​established, which is recorded as

[0146]

[0147] Replacing the original matrix X with X+ε is general and is denoted as X. Then the new X matrix is ​​a column / row full rank matrix. At this time, the value of ε in the ε matrix becomes particularly important. When ε is very small, it does not change the properties of the matrix X, but the full rank effect is not strong. When ε is large, it will change the performance of the X matrix, because the design of the ε matrix must be derived from the experience of the training set and becomes one of the training set parameters.

[0148] (2) Establishing an antigen variation prediction model

[0149] like Figure 7 As shown in the flowchart, input X is the HA sequence matrix, Y is the HI vector (antigenic distance between two viruses), first, let l = 1, screen the first important antigenic site, and use the maximum selection principle Obtain important antigenic sites k1, model process orthogonalization result u1, and similarity scores of important antigenic sites

[0150] Next, l=2 was used to screen the second important antigenic site, using maximum selection The important antigenic site k2 is obtained, where It is an orthogonalization process, which ensures that k2≠k1. The orthogonalization result of the model process is u2, and the similarity score of important antigenic sites is

[0151] When ε<0.01, stop selecting, and select kl=(k1...,kn1), ul=(u1...,un1), Using the important antigenic site similarity score set Represented by a column vector The m×n1 matrix is ​​formed in the order listed, and U=(u1...,u n1 ), and get the transformation matrix

[0152] Finally, the model parameters are ( are training parameters).

[0153] When a new X (HA sequence matrix) is obtained, a predicted Y (antigenic distance) can be obtained by constructing the completed model.

[0154] Among them, the maximum selection principle By setting the minimum ε, the model selects important antigenic sites at each step to avoid overfitting, which can well characterize the relationship between features and targets, better handle nonlinear relationships in antigenic variation, and thus improve prediction accuracy.

[0155] It is an orthogonalization process that ensures k2≠k1. This pre-orthogonalization process can eliminate redundancy and noise in the data, allowing the model to focus more on extracting key features and relationships.

[0156] The HA matrix sparse representation method is used to limit the threshold of ε, so that better prediction results can be obtained by summing a small number of ∑WX.

[0157] Example 2

[0158] In order to evaluate the performance of the antigen variation prediction model in Example 1, this example sets a series of parameters for Matrix-POAFD model training, and divides the influenza B virus samples collected in Example 1 into a training set and a test set at a ratio of 80% and 20%. Further, the Matrix-POAFD model is compared and analyzed with four machine learning algorithms: random forest (RF), LASSO, gradient boosting regression (GBR), and elastic network regression (ENR). CORR (correlation coefficient), RMSE, accuracy index (Precision) and F1-Score are used to evaluate the model performance.

[0159] The significance of the CORR indicator is to measure the strength and direction of the linear relationship between the model's predicted value and the actual observed value. By calculating the correlation coefficient between the predicted value and the actual value, we can understand the degree of correlation between the model's predicted results and the actual situation. RMSE evaluates the deviation between the predicted value and the actual value, reflecting the accuracy of the prediction result. A lower RMSE value usually means that the model has better prediction performance. The accuracy index (Precision) evaluates the model's ability to correctly identify the antigenic similarity between two influenza viruses. The F1-Score is the harmonic mean of Precision and Recall, which comprehensively considers the accuracy and comprehensiveness of the model in identifying antigens, and provides a comprehensive indicator for evaluating model performance.

[0160] The performance of the models based on the above indicators is shown in Table 1 and Figure 8-Figure 9 As shown, Matrix-POAFD always obtains the highest accuracy and F score in CORR, RMSE, Precision and F1-Score. The experimental results show that the Matrix-AFD model constructed in Example 1 is superior to other models in predicting the antigenicity of influenza B virus.

[0161] Similarly, the H5N1 influenza virus samples were collected and divided into a training set and a test set at a ratio of 80% and 20% using the same method as in Example 1, and the Matrix-POAFD model was compared with four machine learning algorithms: random forest (RF), LASSO, gradient boosting regression (GBR), and elastic network regression (ENR). CORR (correlation coefficient), RMSE, accuracy index (Precision) and F1-Score were used to evaluate the model performance.

[0162] The performance of the models based on the above indicators is shown in Table 2 and Figure 10-11 As shown, Matrix-POAFD always obtains the highest accuracy and F score in CORR, RMSE, Precision and F1-Score. The experimental results show that the Matrix-AFD model constructed in Example 1 is superior to other models in predicting the antigenicity of H5N1 influenza virus.

[0163] Table 1

[0164]

[0165] Table 2

[0166]

[0167] Example 3

[0168] In this example, the Matrix-POAFD model is applied to the HI data and HA sequence of the fluB virus. When ε<0.01, the model stops selecting and selects (k1...,k 18 ), that is, a total of 28 important antigenic sites with variations were screened. According to literature reports, four major antigenic epitopes on influenza B virus HA have been identified, namely, 120 loop, 150 loop, 160 loop and 190 helix, which are close in space to form large, continuous antigenic sites. Among them, 9 of the 28 amino acid sites screened in this embodiment are located on these four antigenic epitopes. These 9 important antigenic sites include 2 120loop antigenic sites, 2 1501oop antigenic sites, 4 1601oop antigenic sites, and 1 190loop antigenic site, as shown in Table 3. The bar graph of important residues in different antigenic epitopes is shown in Table 3. Fig.12 The results suggest that changes at these sites may significantly alter antigenicity and may result in antigenic drift.

[0169] Example 4

[0170] This example is based on the same Matrix-POAFD model construction method as that in Example 3, and is applied to the HI data and HA sequence of the H5N1 virus, and 68 important antigenic sites of H5N1 with mutations are screened out. Like human influenza virus, the H5N1 avian influenza virus mainly changes the main antigenic sites of its HA protein to escape antibody attack. According to literature reports, the antigenic epitopes of H5N1 can be divided into five (AE), and 29 of the 68 amino acids screened out in this example are located on these five antigenic epitopes, as shown in Table 4, including 7 A antigenic sites, 10 B antigenic sites, 2 C antigenic sites, 4 D antigenic sites, and 6 E antigenic sites. The bar graph of important residues in different antigenic epitopes is shown in Table 4. Fig.13 The results suggest that changes at these sites may significantly alter antigenicity and may result in antigenic drift.

[0171] Table 3

[0172] Antigenic epitope (Flub) Number of amino acid residues Specific amino acid residue positions 120loop 2 75，116 150loop 2 146，150 160loop 4 162，165，166，167 190loop 1 230

[0173] Table 4

[0174] Antigenic epitope (H5N1) Number of amino acid residues Specific amino acid residue positions A 7 123,124,129,133,136,140,141 B 10 154,155,156,159,162,163,165,188,189,195 C 2 282,309 D 4 94,97,219,235 E 6 53,66,71,83,86,263

[0175] Example 4

[0176] Considering the synergistic effect between amino acids, the screening of single sites is expanded to screening of double sites. This embodiment is based on the Matrix-POAFD model establishment method in Example 3, applied to the HI data and HA sequence of fluB virus, and the training set is feature enhanced in the same way as the polynomial regression idea, such as setting X = (v1…, v n ), the second-order enhancement is denoted by X (2) ,

[0177]

[0178] In this embodiment, a total of 33 antigenic sites with mutations were screened by constructing a Matrix-POAFD model using feature enhancement on the training set, of which 9 antigenic sites were related to the antigenicity of the virus, including 2 120loop antigenic sites, 2 150loop antigenic sites, 4 160loop antigenic sites, and 1 1901oop antigenic site. Table 5 shows the important single sites and double sites selected by the Matrix-POAFD model. Among these mutations, residues 15, 58, 87, 166, 252, and 230 have higher weights in single site mutations and pair site mutations, as shown in Table 5. Fig.14 and Fig.15 As shown, Fig.14 Bar graph showing important residues in different epitopes. Fig.15 A network diagram representing two-site interactions. The results suggest that changes in these sites may drastically alter antigenicity and may lead to antigenic drift.

[0179] Table 5

[0180]

[0181] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0182] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims, and the description and drawings may be used to interpret the contents of the claims.

Claims

1. A method for constructing an antigen variation prediction model, characterized in that: include: Obtain antigenic similarity scores and antigenic distances for known target virus pairs; Constructing a preset function based on the antigenic similarity score and antigenic distance of the known target virus pair, and performing adaptive Fourier decomposition on the preset function using the maximum selection principle to obtain an important antigenic site set, an orthogonal system, and an important antigenic site similarity score set obtained by multiple maximum selections; An antigen variation prediction model is constructed based on the important antigenic site set, the orthogonal system, the important antigenic site similarity score set, and the antigenic distance of known target virus pairs.

2. The method according to claim 1, characterized in that The obtaining of the antigenic similarity score and antigenic distance of the known target virus pair comprises: Acquire multiple antigen protein sequences of known target viruses, and obtain antigen similarity scores of known target virus pairs based on similarity scores of each amino acid site of the multiple antigen protein sequences; A hemagglutination inhibition titer set of known target viruses is obtained, wherein the hemagglutination inhibition titer set of known target viruses includes hemagglutination inhibition titer data of known target viruses and different sera, and the antigenic distance of the known target virus pair is determined based on the hemagglutination inhibition titer set of known target viruses.

3. The method according to claim 1 or 2, characterized in that: The method of using the maximum selection principle to perform adaptive Fourier decomposition on the preset function to obtain a set of important antigenic sites, an orthogonal system and a similarity score set of important antigenic sites obtained by multiple maximum selections includes: Calculate the value of the preset function using the antigenic similarity score of the known target virus pair and the antigenic distance of the known target virus pair, and select the antigenic site, orthogonal base, and antigenic site similarity score that maximize the value of the preset function; After obtaining the important antigenic sites, orthogonal bases and similarity scores of the important antigenic sites, reconstructing the preset function based on the orthogonal bases obtained each time, and selecting the important antigenic sites, orthogonal bases and similarity scores of the important antigenic sites that maximize the value of the reconstructed preset function; When the maximum value of the reconstructed preset function is less than a preset threshold, the screening is stopped to obtain an important antigen site set, an orthogonal system and an important antigen site similarity score set based on multiple maximum selections.

4. The method according to claim 1, characterized in that: The antigen variation prediction model is constructed based on the important antigen site set, the orthogonal system, the important antigen site similarity score set and the antigen distance of the known target virus pair, including: Calculating a conversion coefficient based on the orthogonal system and the important antigenic site similarity score set; Determining training parameters based on the antigenic distance between the orthogonal system and known target virus pairs; Model parameters are determined based on the conversion coefficients and training parameters.

5. A method for predicting influenza virus antigen variation, characterized in that: include: Obtaining antigenic similarity scores between the target virus to be analyzed and the known target virus; The antigenic similarity score between the target virus to be analyzed and the known target virus is input into the antigenic variation prediction model constructed by the method described in any one of claims 1 to 4 to obtain the predicted antigenic distance between the target virus to be analyzed and the known target virus.

6. A device for constructing an antigen variation prediction model, characterized in that: include: The first data acquisition module is used to obtain the antigen similarity score and antigen distance of the known target virus pair; Maximum selection module: used to construct a preset function based on the antigen similarity score and antigen distance of the known target virus pair, and perform adaptive Fourier decomposition on the preset function using the maximum selection principle to obtain an important antigen site set, an orthogonal system and an important antigen site similarity score set obtained by multiple maximum selections; Model parameter calculation module: used to construct an antigen variation prediction model based on the important antigen site set, the orthogonal system, the important antigen site similarity score set and the antigen distance of the known target virus pairs.

7. A device for predicting influenza virus antigen variation, characterized in that: include: The second data acquisition module is used to obtain the antigenic similarity score between the target virus to be analyzed and the known target virus; Prediction module: used to input the antigenic similarity score between the target virus to be analyzed and the known target virus into the antigenic variation prediction model constructed by the method described in any one of claims 1 to 4 to obtain the predicted antigenic distance between the target virus to be analyzed and the known target virus.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

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 4 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 4 are implemented.

Images