A dominant strain prediction model training method, application method and related device

Abstract

The present invention discloses a training method, application method and related device for a dominant strain prediction model, which relates to the field of bioinformatics, wherein the training method for the dominant strain prediction model comprises: obtaining sample data; dividing the sample data from small to large according to the year of the flu season, and obtaining an amino acid sequence subset corresponding to each flu season; determining the corresponding time series set of strain evolution according to the amino acid sequence subset; taking the time series set of the current strain evolution as input, and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next flu season as output, training a neural network model based on seq2seq to obtain a dominant strain prediction model. The present invention can predict the amino acid sequence of the hemagglutinin protein of the dominant strain prevalent in the future flu season, and buy time for the design and generation of vaccines.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and in particular to a dominant strain prediction model training method, an application method and related devices. Background Art

[0002] Vaccines to prevent seasonal influenza infections save many lives every year. In order to evade host immunity, the main antigen of influenza strains, hemagglutinin (HA), changes very quickly. This continuous and cumulative change produces new antigenic strains, which leads to a decrease in the effectiveness of the vaccine. Therefore, vaccines must be frequently updated to remain effective. However, the screening and production of vaccines takes at least 6 months.

[0003] Modern sequencing technology reduces the cost of obtaining HA sequences of influenza strains, promotes the accumulation of HA sequences, and lays the foundation for exploring strain evolution based on sequences. Existing studies have established single-site mutation prediction methods for key sites in HA sequences, or used the frequency of site mutations to calculate strain evolution. However, the mutation rate of influenza viruses is very high, and the sites of mutations in the next generation of strains may not be key sites or single-site mutations; and there may be nonlinear interactions between sites.

[0004] Therefore, how to design a method to predict the dominant influenza strains in the future flu season to buy time for vaccine design and development has become a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] The purpose of the present invention is to provide a dominant strain prediction model training method, application method and related devices. The present invention can predict the amino acid sequence of the hemagglutinin protein of the dominant strain prevalent in the future influenza season, thereby buying time for the design and generation of vaccines.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] In a first aspect, the present invention provides a method for training a dominant strain prediction model, comprising:

[0008] Obtain sample data; the sample data is a set of amino acid sequences of hemagglutinin proteins of dominant influenza strains in several consecutive influenza seasons.

[0009] The sample data is divided according to the years of the flu season from small to large to obtain an amino acid sequence subset corresponding to each flu season.

[0010] According to the amino acid sequence subset, a corresponding time series set of strain evolution is determined.

[0011] Taking the time series set of the current strain evolution as input and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output, the seq2seq-based neural network model is trained to obtain the dominant strain prediction model.

[0012] Optionally, determining a time series set corresponding to the strain evolution according to the amino acid sequence subset specifically includes:

[0013] Select A t An amino acid sequence in b t , for i from t to 2, use BLAST to calculate b i With A i-1 The distance between each amino acid sequence in i-1 Middle and b i A sequence with a distance less than β is taken as b i-1 ;

[0014] Among them, i ranges from 1 to t, A i is the ith amino acid sequence subset, b i is the ith amino acid sequence in the evolutionary time series of a strain, t is a positive integer greater than 1, and β is a real number greater than 0.

[0015] Optionally, the seq2seq-based neural network model includes:

[0016] Embedding module, used to map the input strain evolution time series collection into a vector.

[0017] The encoding module is used to convert the vector into an abstract strain feature.

[0018] A decoding module is used to generate a prediction vector of the amino acid sequence of the hemagglutinin protein of the influenza strain corresponding to the next influenza season according to the strain characteristics.

[0019] The output module is used to map the predicted vector to the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season.

[0020] Optionally, the method uses the time series set of the current strain evolution as input and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output to train the seq2seq-based neural network model to obtain the dominant strain prediction model, specifically including:

[0021] Determine the loss function.

[0022] The time series set of the current strain evolution is input into the seq2seq-based neural network model to obtain the prediction output of the seq2seq-based neural network model.

[0023] The loss value is determined according to the predicted output of the seq2seq-based neural network model, the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season and the loss function.

[0024] The parameters of the seq2seq-based neural network model are optimized according to the loss value to obtain a dominant strain prediction model.

[0025] In a second aspect, the present invention provides an application method of a dominant strain prediction model, comprising:

[0026] Get a collection of time series of strain evolution for the current flu season.

[0027] The time series set of the current strain evolution is input into the dominant strain prediction model to obtain the amino acid sequence of the hemagglutinin protein of the dominant influenza strain in the next influenza season.

[0028] The dominant strain prediction model is a model trained according to the training method of the dominant strain prediction model described in the first aspect.

[0029] In a third aspect, the present invention provides a training device for a dominant strain prediction model, comprising:

[0030] The sample data acquisition unit is used to acquire sample data; the sample data is a set of amino acid sequences of hemagglutinin proteins of dominant influenza strains in several consecutive influenza seasons.

[0031] The partitioning unit is used to partition the sample data according to the years of the flu season from small to large, so as to obtain an amino acid sequence subset corresponding to each flu season.

[0032] The strain evolution time series set determination unit is used to determine the corresponding strain evolution time series set based on the amino acid sequence subset.

[0033] The training unit is used to train the seq2seq-based neural network model with the time series set of the current strain evolution as input and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output to obtain a dominant strain prediction model.

[0034] In a fourth aspect, the present invention provides an application device of a dominant strain prediction model, comprising:

[0035] A data acquisition unit, used to obtain a time series set of the current strain evolution;

[0036] The prediction unit inputs the time series set of the current strain evolution into the dominant strain prediction model to obtain the amino acid sequence of the hemagglutinin protein of the dominant influenza strain in the next influenza season; the dominant strain prediction model is a model trained according to the training method of the dominant strain prediction model described in the first aspect.

[0037] In a fifth aspect, the present invention provides an electronic device, comprising:

[0038] Processor; and

[0039] a memory storing computer readable program instructions,

[0040] Wherein, when the computer-readable program instructions are executed by the processor, the training method of the dominant strain prediction model as described in the first aspect is executed, or the application method of the dominant strain prediction model as described in the second aspect is executed.

[0041] In a sixth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the training method of the dominant strain prediction model described in the first aspect, or implements the steps of the application method of the dominant strain prediction model described in the second aspect.

[0042] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0043] The present invention provides a training method, an application method and related devices for a dominant strain prediction model. The dominant strain prediction model of the present invention is a seq2seq-based neural network structure model. The model takes the time series set of the current strain evolution as input and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output. It can predict the amino acid sequence of the hemagglutinin protein of the dominant strain prevalent in the future influenza season, thereby buying time for the design and generation of vaccines. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0045] Figure 1 A flowchart of a method for training a dominant strain prediction model provided in Example 1 of the present invention;

[0046] Figure 2 A flowchart of an application method of a dominant strain prediction model provided in Example 2 of the present invention;

[0047] Figure 3 A schematic diagram of the structure of a training device for a dominant strain prediction model provided in Example 3 of the present invention;

[0048] Figure 4 A schematic diagram of the structure of an application device of a dominant strain prediction model provided in Example 4 of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The purpose of the present invention is to provide a dominant strain prediction model training method, application method and related devices, through which the dominant strain prediction model can predict the dominant strains prevalent in the future influenza season, thereby buying time for the design and generation of vaccines.

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Embodiment 1:

[0053] like Figure 1 As shown, this embodiment provides a training method for a dominant strain prediction model, comprising:

[0054] A1: Obtain sample data; the sample data is a set of amino acid sequences of hemagglutinin proteins of dominant influenza strains in t consecutive influenza seasons, where t is a positive integer greater than 1.

[0055] In practical applications, the amino acid sequence of the hemagglutinin protein of one influenza subtype, such as H3N2, H1N1, H5N1 or B, covering multiple consecutive influenza seasons can be downloaded through the National Center for Biotechnology Information (NCBI) or the Global Initiative on Sharing Influenza Data (GISAID).

[0056] In this embodiment, a set of H3N2 hemagglutinin protein amino acid sequences covering 10 consecutive influenza seasons between 2013 and 2022 is downloaded from GISAID, and each amino acid sequence contains 566 amino acids.

[0057] A2: Divide the sample data according to the years of the flu season from small to large, and obtain an amino acid sequence subset corresponding to each flu season.

[0058] Specifically, for the amino acid sequence set of hemagglutinin proteins of dominant influenza strains in t consecutive influenza seasons, the strains are divided into t non-overlapping subsets according to the influenza seasons in which the strains were collected, and the amino acid sequence subsets A1, A2, ..., A3 are obtained in order from small to large according to the years of the influenza seasons. t .

[0059] In this embodiment, the amino acids of the H3N2 hemagglutinin protein downloaded from 2013 to 2022 are divided into 10 subsets, which correspond to the 10 influenza seasons A1, A2, ..., A 10 .

[0060] A3: Determine the corresponding time series set of strain evolution based on the amino acid sequence subset.

[0061] A4: Take the time series set of the current strain evolution as input and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output, train the seq2seq-based neural network model, and obtain the dominant strain prediction model.

[0062] As an optional implementation method provided in this embodiment, step A3 specifically includes:

[0063] According to the amino acid sequence subsets A1, A2, ..., A t , construct a set of time series of strain evolution; each strain evolution time series in the set (b1, b2, ..., b t ) is obtained by:

[0064] Select A t An amino acid sequence in b t , for i from t to 2, use BLAST to calculate b i With A i-1 The distance between each amino acid sequence in i-1 Middle and b i A sequence with a distance less than β is taken as b i-1 ;

[0065] Among them, i ranges from 1 to t, A i is the ith amino acid sequence subset, b i is the ith amino acid sequence in the evolutionary time series of a strain, t is a positive integer greater than 1, and β is a real number greater than 0.

[0066] In this embodiment, β=0.0001, and the time series of each strain evolution is (b1, b2, ..., b 10 ), b 10 Yes A 10 An amino acid sequence in b was calculated using BLAST 10 The distance from all amino acid sequences in A9, b9 is an amino acid sequence in A9, and b 10 The distance from b9 is less than β; using the same method, determine b8, b7, ..., b1 in turn; finally, obtain a time series set of strain evolution.

[0067] As an optional implementation provided by this embodiment, the seq2seq-based neural network model includes:

[0068] Embedding module, used to map the input strain evolution time series collection into a vector.

[0069] The encoding module is used to convert the vector into an abstract strain feature.

[0070] A decoding module is used to generate a prediction vector of the amino acid sequence of the hemagglutinin protein of the influenza strain corresponding to the next influenza season according to the strain characteristics.

[0071] The output module is used to map the prediction vector to the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season to complete the prediction process of the entire model.

[0072] Among them, the encoding module and the decoding module can adopt a stacked recurrent neural network (RNN), a long short-term memory network (LSTM) or a transformer network (Transformer).

[0073] In this embodiment, a transformer network with one layer is used to complete the encoding and decoding of the time series. The embedding module vectorizes each amino acid sequence in the strain evolution time series based on prot2vec. A known time series (b1, b2, ..., b 10 ) is a training sample, which is vectorized using prot2vec to obtain vectors u1, u2, ..., u9, u 10 The strain sequences (b1, b2, ..., b9) of the 1st to 9th influenza seasons can be used as the observed values ​​of the sample, and the strain b of the 10th influenza season 10 As the label value of the sample. For a known time series (b1, b2, ..., b 10 ), concatenate u1, u2, ..., u9 in sequence to obtain the observation vector o, u 10 As the label vector r of this sample.

[0074] The encoding module completes the following calculation on the observation vector o through the self-attention mechanism:

[0075]

[0076]

[0077]

[0078]

[0079] in, are the three parameter matrices of the model, and ⊙ is the dot product of the matrices. e represents the query vector, K e Represents the key vector, V e represents the value vector. The three vectors are obtained by multiplying the weight matrix and the observation vector o element by element, and are used for the query, key and value calculation of the self-attention mechanism. σ represents the activation function softmax, K e T is the matrix K e The transpose operation, d k It's K e The dimension is used as a scaling factor to prevent the softmax function from having too large an input, which can lead to the gradient vanishing problem. Finally, the self-attention output score is obtained.

[0080] Further calculate a hidden state vector:

[0081] h=V e ☉M;

[0082] The decoding module uses the same attention to calculate a state vector, and can accept the vector u9 of strain b9 as the prompt input p, and calculate as follows:

[0083]

[0084]

[0085]

[0086]

[0087] in, are the three parameter matrices of the model, σ represents the activation function softmax, ⊙ is the dot product of the matrix; K d T is the matrix K d The transpose operation, d k It's K d Dimension; Qd represents the query vector, K d Represents the key vector, V d Represents the value vector; the query vector is obtained by multiplying the weight matrix with the prompt vector p, and the key vector and value vector are obtained by multiplying the weight matrix with the hidden vector obtained by the encoder, and are used for the query, key and value calculation of the self-attention mechanism. The decoder finally outputs the predicted strain vector y.

[0088] The output module uses an amino acid encoding dictionary to map the predicted strain vector y into an amino acid sequence.

[0089] As an optional implementation method provided in this embodiment, step A4 specifically includes:

[0090] Determine the loss function, which is as follows:

[0091]

[0092] Among them, n represents the number of training samples, r i is the label vector of the i-th training sample, y i is the prediction vector of the i-th training sample calculated by the prediction model of the dominant strain, and ||.|| is the L1 norm.

[0093] The time series set of the current strain evolution is input into the seq2seq-based neural network model to obtain the prediction output of the seq2seq-based neural network model.

[0094] The loss value is determined according to the predicted output of the seq2seq-based neural network model, the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season and the loss function.

[0095] The parameters of the seq2seq-based neural network model are optimized according to the loss value to obtain a dominant strain prediction model.

[0096] Specifically, a randomized method is used to initialize the model parameters, back propagation and stochastic gradient descent are used to minimize the loss function to train the model, and all the model parameters are updated.

[0097] In this embodiment, the surface glycoprotein hemagglutinin of the influenza strain can be predicted instead of the mutation prediction of a single key site; at the same time, the dominant strains in the future influenza season can be predicted, buying time for the design and generation of the vaccine.

[0098] Embodiment 2:

[0099] like Figure 2As shown, this embodiment provides an application method of a dominant strain prediction model, comprising:

[0100] B1: Get the time series set of the current strain evolution.

[0101] B2: Input the time series set of the current strain evolution into the dominant strain prediction model to obtain the amino acid sequence of the hemagglutinin protein of the dominant influenza strain in the next influenza season.

[0102] The dominant strain prediction model is a model trained according to the dominant strain prediction model training method described in Example 1.

[0103] It should be noted that the step of obtaining the time series set of the current strain evolution is the same as the step of determining the corresponding time series set of the strain evolution according to the amino acid sequence subset in Example 1, and will not be repeated here.

[0104] In practical applications, multi-step prediction can be used to predict the amino acid sequence of the hemagglutinin protein of the dominant strain that will be prevalent after the next k influenza seasons, where k is a positive integer.

[0105] In this embodiment, a known strain evolution time series is (c1, ..., c9), which is input into the dominant strain prediction model, wherein c1, ..., c9 are from 9 consecutive influenza seasons from 2014 to 2022 respectively; through the following process, after predicting 3 influenza seasons, the hemagglutinin protein amino acid sequence of the dominant strain in the 2025 influenza season can be known:

[0106] First, the dominant strains after the next flu season are predicted, and the amino acid sequence of the hemagglutinin protein of the dominant strains in the 10th flu season, that is, the 2023 flu season, is obtained. Then, the amino acid sequence c of the dominant strains in the 2023 flu season can be determined. 10 ; (c2, ..., c9, c 10 ) is input into the dominant strain prediction model, and then the dominant strain after the next influenza season is predicted to obtain the amino acid sequence of the hemagglutinin protein of the dominant strain in the 11th influenza season, and then the amino acid sequence of the dominant strain in the 2024 influenza season can be determined. 11 ; Then (c3, ..., c9, c 10 , c 11 ) is input into the dominant strain prediction model, and the amino acid sequence of the hemagglutinin protein of the dominant strain three influenza seasons later can be obtained, that is, the amino acid sequence of the hemagglutinin protein of the dominant strain prevalent in the 2025 influenza season.

[0107] Embodiment 3:

[0108] like Figure 3As shown, this embodiment provides a training device for a dominant strain prediction model, comprising:

[0109] The sample data acquisition unit M1 is used to acquire sample data; the sample data is a set of amino acid sequences of hemagglutinin proteins of dominant influenza strains in several consecutive influenza seasons.

[0110] The division unit M2 is used to divide the sample data according to the year of the flu season from small to large, so as to obtain an amino acid sequence subset corresponding to each flu season.

[0111] The strain time series set determination unit M3 is used to determine the corresponding strain evolution time series set based on the amino acid sequence subset.

[0112] The training unit M4 is used to train the seq2seq-based neural network model with the time series set of the current strain evolution as input and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output to obtain a dominant strain prediction model.

[0113] Embodiment 4:

[0114] like Figure 4 As shown, this embodiment provides an application device of a dominant strain prediction model, comprising:

[0115] Data acquisition unit N1, used to obtain the time series set of the current strain evolution;

[0116] The prediction unit N2 inputs the time series set of the current strain evolution into the dominant strain prediction model to obtain the amino acid sequence of the hemagglutinin protein of the dominant influenza strain in the next influenza season; the dominant strain prediction model is a model trained according to the training method of the dominant strain prediction model described in Example 1.

[0117] Embodiment 5:

[0118] This embodiment provides an electronic device, including:

[0119] Processor; and

[0120] a memory storing computer readable program instructions,

[0121] Wherein, when the computer-readable program instructions are executed by the processor, the training method of the dominant strain prediction model as described in Example 1 is executed, or the application method of the dominant strain prediction model as described in Example 2 is executed.

[0122] Embodiment 6:

[0123] The present invention provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the training method of the dominant strain prediction model described in Example 1 are implemented, or the steps of the application method of the dominant strain prediction model described in Example 2 are implemented.

[0124] The program part of the technology can be considered as a "product" or "manufactured product" in the form of executable code and / or related data, which is participated in or realized by computer-readable media. Tangible and permanent storage media can include any memory or storage used by any computer, processor, or similar device or related module. For example, various semiconductor memories, tape drives, disk drives or any similar devices that can provide storage functions for software.

[0125] All or part of the software may sometimes communicate over a network, such as the Internet or other communications network. Such communications can load software from one computer device or processor to another. For example: from a server or host computer of a video target detection device to a hardware platform of a computer environment, or other computer environment that implements the system, or a system with similar functions related to providing information required for target detection. Therefore, another medium capable of transmitting software elements can also be used as a physical connection between local devices, such as light waves, radio waves, electromagnetic waves, etc., transmitted through cables, optical cables or air. Physical media used to carry carriers, such as cables, wireless connections or optical cables and the like, can also be considered as media that carry software. As used herein, unless limited to tangible "storage" media, other terms referring to computer or machine "readable media" refer to media involved in the process of the processor executing any instructions.

[0126] In addition, those skilled in the art will appreciate that various aspects of the present invention may be illustrated and described by a number of patentable categories or situations, including any new and useful process, machine, product or combination of substances, or any new and useful improvements thereto. Accordingly, various aspects of the present invention may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may all be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". In addition, various aspects of the present invention may be represented as a computer product located in one or more computer-readable media, which includes computer-readable program code.

[0127] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the invention belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined as such herein.

[0128] The above is an explanation of the present invention and should not be considered as a limitation thereof. Although several exemplary embodiments of the present invention have been described, it will be readily appreciated by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined in the claims. It should be understood that the above is an explanation of the present invention and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

[0129] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0130] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A training method for a dominant strain prediction model, characterized in that: include: Obtaining sample data; the sample data is a set of amino acid sequences of hemagglutinin proteins of dominant influenza strains in several consecutive influenza seasons; Dividing the sample data according to the year of the flu season from small to large, to obtain an amino acid sequence subset corresponding to each flu season; Determining a corresponding time series set of strain evolution according to the amino acid sequence subset; Taking the time series set of the current strain evolution as input and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output, the seq2seq-based neural network model is trained to obtain a dominant strain prediction model; According to the amino acid sequence subset, a time series set corresponding to the strain evolution is determined, specifically including: Select A t An amino acid sequence in b t , for i from t to 2, use BLAST to calculate b i With A i-1 The distance between each amino acid sequence in i-1 Middle and b i A sequence with a distance less than β is taken as b i-1 ; Among them, i ranges from 1 to t, A i is the ith amino acid sequence subset, b i is the ith amino acid sequence in the evolutionary time series of a strain, t is a positive integer greater than 1, and β is a real number greater than 0; The seq2seq-based neural network model includes: The embedding module is used to map the input time series of strain evolution into a vector; An encoding module, used for converting the vector into an abstract strain feature; A decoding module, for generating a prediction vector of the amino acid sequence of the hemagglutinin protein of the influenza strain corresponding to the next influenza season according to the strain characteristics; An output module, used to map the predicted vector to an amino acid sequence of a hemagglutinin protein of an influenza strain in the next influenza season; The method uses the time series set of the current strain evolution as input and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output to train the seq2seq-based neural network model to obtain a dominant strain prediction model, specifically including: Determine the loss function; the loss function is as follows: Among them, l represents the loss function, n represents the number of training samples, and r i is the label vector of the i-th training sample, y i is the prediction vector of the ith training sample calculated by the prediction model of the dominant strain, and ||.|| is the L1 norm; Inputting the time series set of the current strain evolution into the seq2seq-based neural network model to obtain the prediction output of the seq2seq-based neural network model; Determining a loss value according to the predicted output of the seq2seq-based neural network model, the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season, and the loss function; The parameters of the seq2seq-based neural network model are optimized according to the loss value to obtain a dominant strain prediction model.

2. A method for applying a dominant strain prediction model, characterized in that: include: Get the time series set of the current strain evolution; Inputting the time series set of the current strain evolution into the dominant strain prediction model to obtain the amino acid sequence of the hemagglutinin protein of the dominant influenza strain in the next influenza season; The dominant strain prediction model is a model trained according to the dominant strain prediction model training method according to claim 1.

3. A training device for a dominant strain prediction model, characterized in that: include: A sample data acquisition unit, used to acquire sample data; the sample data is a set of amino acid sequences of hemagglutinin proteins of dominant influenza strains in several consecutive influenza seasons; A division unit, used to divide the sample data according to the year of the flu season from small to large, to obtain an amino acid sequence subset corresponding to each flu season; A strain evolution time series set determination unit, used to determine the corresponding strain evolution time series set according to the amino acid sequence subset; A training unit is used to train a seq2seq-based neural network model using a time series set of the current strain evolution as input and an amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output to obtain a dominant strain prediction model; According to the amino acid sequence subset, a time series set corresponding to the strain evolution is determined, specifically including: Select A t An amino acid sequence in b t , for i from t to 2, use BLAST to calculate b i With A i-1 The distance between each amino acid sequence in i-1 Middle and b i A sequence with a distance less than β is taken as b i-1 ; Among them, i ranges from 1 to t, A i is the ith amino acid sequence subset, b i is the ith amino acid sequence in the evolutionary time series of a strain, t is a positive integer greater than 1, and β is a real number greater than 0; The seq2seq-based neural network model includes: The embedding module is used to map the input time series of strain evolution into a vector; An encoding module, used for converting the vector into an abstract strain feature; A decoding module, for generating a prediction vector of the amino acid sequence of the hemagglutinin protein of the influenza strain corresponding to the next influenza season according to the strain characteristics; An output module, used to map the predicted vector to an amino acid sequence of a hemagglutinin protein of an influenza strain in the next influenza season; The method uses the time series set of the current strain evolution as input and the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season as output to train the seq2seq-based neural network model to obtain a dominant strain prediction model, specifically including: Determine the loss function; the loss function is as follows: Among them, l represents the loss function, n represents the number of training samples, and r i is the label vector of the i-th training sample, y i is the prediction vector of the ith training sample calculated by the prediction model of the dominant strain, and ||.|| is the L1 norm; Inputting the time series set of the current strain evolution into the seq2seq-based neural network model to obtain the prediction output of the seq2seq-based neural network model; Determining a loss value according to the predicted output of the seq2seq-based neural network model, the amino acid sequence of the hemagglutinin protein of the influenza strain of the next influenza season, and the loss function; The parameters of the seq2seq-based neural network model are optimized according to the loss value to obtain a dominant strain prediction model.

4. An application device for a dominant strain prediction model, characterized in that: include: A data acquisition unit, used to obtain a time series set of the current strain evolution; The prediction unit inputs the time series set of the current strain evolution into the dominant strain prediction model to obtain the amino acid sequence of the hemagglutinin protein of the dominant influenza strain in the next influenza season; the dominant strain prediction model is a model trained according to the training method of the dominant strain prediction model according to claim 1.

5. An electronic device comprising: processor; as well as a memory storing computer readable program instructions, Wherein, when the computer-readable program instructions are executed by the processor, the training method of the dominant strain prediction model as claimed in claim 1 is executed, or the application method of the dominant strain prediction model as claimed in claim 2 is executed.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the training method of the dominant strain prediction model described in claim 1 are implemented, or the steps of the application method of the dominant strain prediction model described in claim 2 are implemented.

Images