Method, device and medium for constructing a viral single-gene host fitness prediction model

Abstract

The application belongs to the field of intelligent medical treatment, and particularly relates to a method for constructing a virus single-gene host adaptability prediction model, equipment and a medium. The method comprises the following steps: obtaining a training set virus single-gene data set and a host label; encoding the single genes in the virus single-gene data set in sequence to obtain an encoding feature matrix set; the encoding is as follows: using a sliding window method to traverse a gene sequence, the frequency of 64 different codons in a window is used to form a window codon frequency vector, the window codon frequency vector is used to represent a central codon of the window, and each codon of the gene sequence is represented as a window codon frequency vector in sequence; inputting the encoding feature matrix set into a residual network model to obtain a prediction label, and iteratively training based on a comparison between the prediction label and the host label to obtain a single-gene host adaptability prediction model. The model capable of predicting gene adaptability is trained by using the representation of codons, and the analysis of gene codon adaptability can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent medical treatment, and more particularly, to a method, device, medium and program product for constructing a virus single-gene host adaptability prediction model. BACKGROUND

[0002] Influenza virus, the representative species of Orthomyxoviridae, is simply called influenza virus, including human influenza virus with human host and animal influenza virus with animal host. Influenza virus is divided into four types: A, B, C and D, among which type A and type B influenza virus is the pathogen of influenza, which can cause acute respiratory infectious disease. The main symptoms after infection are cough, high fever, runny nose, myalgia, etc. A small number of cases are accompanied by severe pneumonia, and severe cases can lead to multiple organ failure and death. The highly pathogenic avian influenza virus infection has a high mortality rate. The antigenicity of influenza virus A is prone to variation, which has caused several worldwide pandemics. The segmented genome structure of influenza virus makes it easy to recombine between different viruses when co-infected. At the same time, the surface protein of influenza virus is subjected to the selective pressure of host population immunity, leading to antigenic drift, causing repeated epidemics of influenza virus A or B. Host adaptability of influenza virus is a key scientific problem for its prevention and control research.

[0003] The eight segments of the influenza virus genome, HA, NA, M, NS, NP, PB1, PB2, and PA, encode 10 proteins: hemagglutinin (HA), neuraminidase (NA), matrix proteins M2 and M1, non-structural proteins NS1 and NS2, nucleocapsid, and three polymerases. In the past two decades, in the study of viral adaptability, the definition of adaptability at least includes the virus's ability to infect, cause disease, and easily spread in a certain host. High adaptability means that the virus can spread more effectively in the host population. High infectivity, high transmissibility, and low pathogenicity of the virus match its high adaptability, and vice versa. So far, IAVs adapted to humans mainly include H3N2 and H1N1 subtypes. Avian influenza virus subtypes such as H5N1 and H7N9 occasionally infect humans, but cannot spread between humans and do not have human adaptability. The three swine influenza virus subtypes widely prevalent in the world, H1N1 (swine classic and avian-like H1N1), H3N2 (human-like and gene reassortment H3N2), and H1N2, can also occasionally infect humans, but do not have the ability to effectively spread in the human population. Avian influenza virus (AIV) has significantly enhanced replication capacity, polymerase activity, and other phenotypes after serial passage in mammalian cells or mammals, improving its adaptability to mammals (cells). SUMMARY

[0004] In view of the above problems, the application provides a method for constructing a virus single-gene host adaptability prediction model, which is trained by a characteristic coding matrix after coding the genes, so as to effectively monitor the variability of a single gene in a virus.

[0005] The application (first aspect) discloses a method for constructing a virus single-gene host adaptability prediction model, comprising:

[0006] obtaining a training set virus single-gene data set and a host label,

[0007] encoding the single genes in the virus single-gene data set in sequence to obtain a coding characteristic matrix set, wherein the encoding comprises: connecting the gene sequence of the single gene at the head and tail, then using a sliding window method to traverse the gene sequence, the frequency of 64 different codons in the window forming a window codon frequency vector, the window codon frequency vector representing the central codon of the window, and with the sliding of the window, each codon of the gene sequence is represented as the window codon frequency vector when the codon is the central codon, and after the traversal, a coding characteristic matrix composed of window codon frequency vectors is obtained;

[0008] inputting the coding characteristic matrix set into a neural network model to obtain a prediction label, and iteratively training a virus single-gene host adaptability prediction model based on the comparison between the prediction label and the host label.

[0009] Further, the method comprises: the central codon of the sliding window represents the codon located at the center position of the sliding window.

[0010] Optionally, the width of the sliding window is S codons, if S is odd, the central codon is the (S+1) / 2 codon in the window, and if S is even, the central codon is the S / 2 codon or the S / 2 codon in the window.

[0011] Optionally, when the index of the central codon in the gene sequence is less than S / 2 or, the codon at the end of the gene sequence is used to supplement the missing codon of the sliding window, and when the index of the central codon in the gene sequence is greater than n-S, the codon at the starting position of the gene sequence is used to supplement the missing codon of the sliding window, wherein n represents that the gene sequence contains n codons in total.

[0012] The 64 codons are 64 codons of the standard genetic code.

[0013] Optionally, the dimension of the coding characteristic matrix is n*64, wherein n represents that the gene sequence contains n codons in total.

[0014] Optionally, the first row of the encoding feature matrix is a window codon frequency vector when the first codon of the gene sequence is the center codon, and the last row is a window codon frequency vector when the last codon of the gene sequence is the center codon.

[0015] Optionally, the step size of the sliding window method is 1 codon.

[0016] Further, the set of encoding feature matrices is subjected to auto-encoding learning to obtain a second set of encoding feature matrices, and the encoding feature matrices in the second set of encoding feature matrices are corrected to the same dimension as the encoding feature matrices by padding 0 to obtain a third set of encoding feature matrices, and the third set of encoding feature matrices and the host label are input into a neural network model for iterative training to obtain a virus single gene host adaptability prediction model.

[0017] Optionally, the auto-encoding learning is based on one or more of the following models: LTSM, Transformer, Transformer-XL, MLP, and MLP-Mixer.

[0018] Optionally, the set of encoding feature matrices is subjected to auto-encoding learning to obtain a second set of encoding feature matrices, and the encoding feature matrices in the second set of encoding feature matrices are corrected to the same dimension as the encoding feature matrices by padding 0 to obtain a third set of encoding feature matrices, and the third set of encoding feature matrices and the host label are input into a neural network model for iterative training to obtain a virus single gene host adaptability prediction model.

[0019] Optionally, if the encoding feature matrix does not match the input dimension of the auto-encoding model, 0 is padded at the end to meet the requirements.

[0020] Optionally, the matrices in the set of encoding feature matrices and the third set of encoding feature matrices are subjected to representation effect evaluation, and the matrix with good representation effect is selected as the optimal encoding feature matrix, the optimal encoding feature matrix of the training set virus single gene constitutes an optimal encoding feature matrix set, and the optimal encoding feature matrix and the host label are input into a neural network model for iterative training to obtain a virus single gene host adaptability prediction model.

[0021] Optionally, the representation effect evaluation is one or more of the following: clustering evaluation, and umap dimensionality reduction evaluation.

[0022] Further, the training set virus single gene is any one of the following genes: HA, NA, M, NS, NP, PB1, PB2, and PA, and the corresponding virus single gene host adaptability prediction model is obtained after training.

[0023] The present application (second aspect) discloses a virus single gene host adaptability prediction method, comprising:

[0024] An encoding feature matrix of a virus single gene is obtained.

[0025] Input the coding feature matrix of the virus single gene into the virus single gene host adaptability prediction model constructed by the construction method of the virus single gene host adaptability prediction model to obtain the host adaptability of the virus single gene.

[0026] The application (third aspect) discloses a virus gene sequence codon importance ranking method, comprising:

[0027] S1: obtaining a coding feature matrix set and a host label of a test set virus single gene;

[0028] S2: inputting the coding feature matrix set of the test set into the single gene host adaptability prediction model constructed by the construction method of the virus single gene host adaptability prediction model to obtain a predicted host label, and comparing the predicted host label with the test set host label to obtain a first prediction accuracy;

[0029] S3: using a sliding window method to mask the first window of the coding feature matrix set of the test set to obtain a masked first window coding feature matrix set, the center of the first window being the first row of the coding feature matrix set, inputting the masked first window coding feature matrix set into the same single gene host adaptability prediction model to calculate the prediction accuracy of the masked first window, and repeating S3 after the sliding window traverses along the coding feature matrix along the row to obtain a second prediction accuracy, the second prediction accuracy comprising the prediction accuracies of the first window to the nth window, n representing the number of codons of the single gene;

[0030] S4: obtaining the region vector importance of the first window to the nth window based on the difference between the second prediction accuracy and the first prediction accuracy;

[0031] S5: obtaining the importance of each codon by averaging the region vector importance of each codon.

[0032] Further, the coding feature matrix set of the test set is an optimized coding feature matrix set obtained by inputting the coding feature set of the test set into the autoencoder constructed by the self-encoding learning of the coding feature matrix set of the training set.

[0033] Optionally, the coding feature matrix set of the test set is a set of optimal coding feature matrices obtained by evaluating the representation effect of the matrices in the coding feature matrix set and the optimized coding feature matrix set.

[0034] Optionally, the S3 further comprises: obtaining a Bayesian posterior accuracy of the first mask window based on the prediction accuracy of the first mask window and the probability of each window of the mask; repeating the S3 step to obtain a third prediction accuracy, the third prediction accuracy comprising Bayesian posterior accuracies of the first to the n-th windows of the mask, and obtaining a Bayesian region vector importance of the first to the n-th windows based on a difference between the third prediction accuracy and the first prediction accuracy; and obtaining a Bayesian posterior codon importance of each codon by averaging the Bayesian region vector importance of each codon.

[0035] Optionally, an optimal codon importance vector is obtained based on a weighted average of the codon importance and the Bayesian posterior codon importance.

[0036] Optionally, key codons of each fragment are screened based on the set importance threshold.

[0037] Optionally, the Bayesian posterior accuracy of the first mask window is represented as:

[0038]

[0039] wherein, P(B i |A) represents the Bayesian posterior accuracy of the i-th window of the mask; B i represents the event of the i-th window of the mask; P(B i ) represents the probability of the i-th window of the mask; and P(A|B i ) represents the prediction accuracy of the i-th window of the mask.

[0040] The fourth aspect of the present application discloses a method and system for constructing a virus single-gene host adaptability prediction model, comprising:

[0041] An acquisition module is configured to acquire a training set of virus single-gene data and host labels.

[0042] A feature encoding module is configured to encode the single genes in the virus single-gene data set in sequence to obtain an encoded feature matrix set, the encoding comprising: connecting the gene sequence of the single gene at the head and tail, and then traversing the gene sequence using a sliding window method, the frequency of 64 different codons in the window forming a window codon frequency vector, the window codon frequency vector representing the central codon of the window, and with the window sliding, each codon of the gene sequence is represented as the window codon frequency vector when the codon is the central codon, and after the traversal, an encoded feature matrix formed by the window codon frequency vectors is obtained.

[0043] The training module is configured to input the coding feature matrix set into a neural network model to obtain a predicted label, and iteratively train a virus single-gene host adaptability prediction model based on a comparison between the predicted label and a host label.

[0044] The present application (the fifth aspect) discloses a virus single-gene host adaptability prediction system, comprising:

[0045] The second acquisition module is configured to acquire a coding feature matrix of a virus single-gene.

[0046] The second prediction module is configured to input the coding feature matrix into a virus single-gene host adaptability prediction model constructed by using the construction method of the virus single-gene host adaptability prediction model to obtain host adaptability of the virus single-gene.

[0047] The present application (the sixth aspect) discloses a virus gene sequence codon importance ranking method, comprising:

[0048] The third acquisition module is configured to acquire a coding feature matrix set and a host label of a test set virus single-gene.

[0049] The first accuracy prediction module is configured to input the coding feature matrix set of the test set into a single-gene host adaptability prediction model constructed by using the construction method of the virus single-gene host adaptability prediction model to obtain a predicted host label, and obtain a first prediction accuracy by comparing the predicted host label with the host label of the test set.

[0050] The sliding window mask prediction module is configured to obtain a mask first window coding feature matrix set by using a sliding window method to mask the first window of the coding feature matrix set of the test set, and obtain a mask first window prediction accuracy by inputting the mask first window coding feature matrix set into the same single-gene host adaptability prediction model, wherein the center of the first window is the first row of the coding feature matrix set, and the sliding window is sequentially traversed along the rows of the coding feature matrix to repeatedly obtain a second prediction accuracy, wherein the second prediction accuracy comprises prediction accuracies of the mask first window to the nth window, and n represents the number of codons of the single-gene.

[0051] The region vector importance calculation module is configured to obtain region vector importance of the first window to the nth window based on a difference between the second prediction accuracy and the first prediction accuracy.

[0052] The codon importance calculation module is configured to obtain the importance of each codon by averaging the region vector importance of each codon.

[0053] The seventh aspect of the present application discloses a computer device, the device comprising: a memory and a processor; the memory is used to store program instructions; the processor is used to call the program instructions, when the program instructions are executed, the steps of the above-mentioned method are executed.

[0054] The eighth aspect of the present application discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the steps of the above-mentioned method.

[0055] The ninth aspect of the present application discloses a computer program product, comprising a computer program, which is executed by a processor to realize the steps of the above-mentioned method.

[0056] The present application has the following beneficial effects:

[0057] (1) The present application can realize the analysis of the codon adaptability of the gene by using the model trained after the characterization of the codon to predict the adaptability of the gene;

[0058] (2) The present application uses a single-gene host adaptability prediction model to analyze the codon importance of the gene, which can be used for screening adaptive gene sites, and can identify the genetic causal relationship of the existing influenza virus host adaptability. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained according to these drawings without creative labor for those skilled in the art.

[0060] Figure 1 is a method flowchart provided by the first aspect of the embodiments of the present application;

[0061] Figure 2 is a program product schematic diagram provided by the fourth aspect of the embodiments of the present application;

[0062] Figure 3 is a schematic diagram of a computer device provided by the embodiments of the present application;

[0063] Figure 4 is a schematic diagram of the architecture of an exemplary computing device provided by the embodiments of the present application;

[0064] Figure 5 is a schematic diagram of a storage medium provided by the embodiments of the present application;

[0065] Figure 6 is a Codon2Vec sliding window characterization flowchart provided by the embodiments of the present application;

[0066] Figure 7 is a Codon2Vec preprocessing and characterization evaluation flowchart provided by an embodiment of the present application;

[0067] Figure 8 is a ResNet residual network framework diagram provided by an embodiment of the present application;

[0068] Figure 9 is a codon importance calculation flowchart provided by an embodiment of the present application;

[0069] Figure 10 is a schematic diagram of a Bayesian model calculating codon importance provided by an embodiment of the present application;

[0070] Figure 11 is a schematic diagram of an ablation experiment flowchart provided by an embodiment of the present application;

[0071] Figure 12 is a simulation of reassortant data generation schematic diagram provided by an embodiment of the present application;

[0072] Figure 13 is a reassortment adaptability model construction and prediction flowchart provided by an embodiment of the present application. DETAILED DESCRIPTION

[0073] In order to enable persons skilled in the art to better understand the present application scheme, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present application.

[0074] In some of the descriptions in the specification and claims of the present application and the above-mentioned drawings, a plurality of operations appearing in a specific order are included, but it should be clearly understood that these operations can be executed or in parallel without the order in which they appear in this text. The serial numbers of the operations, such as S101, S102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes can include more or fewer operations, and the operations can be executed in sequence or in parallel. It should be noted that the "first", "second", etc. in this text are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence. Also, "first" and "second" are not different types.

[0075] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0076] Figure 1 is a flowchart of a method for constructing a virus single-gene host adaptability prediction model provided by an embodiment of the present application, comprising:

[0077] S101: obtaining a training set virus single-gene dataset and a host label,

[0078] S102: sequentially encoding single genes in the virus single-gene dataset to obtain an encoding feature matrix set, the encoding being: connecting the gene sequence of the single gene at the head and tail, then using a sliding window method to traverse the gene sequence, the frequency of 64 different codons in the window forming a window codon frequency vector, the window codon frequency vector representing the central codon of the window, with the window sliding, sequentially representing each codon of the gene sequence as the window codon frequency vector when the codon is the central codon, and after the traversal, obtaining an encoding feature matrix composed of window codon frequency vectors;

[0079] S103: inputting the encoding feature matrix set into a neural network model to obtain a prediction label, and based on the comparison of the prediction label and the host label, iteratively training to obtain a virus single-gene host adaptability prediction model.

[0080] The gene fragment of a virus refers to a nucleic acid sequence region in the viral genome that has specific functional or structural characteristics, usually corresponding to the structural protein (such as capsid protein, envelope protein) or non-structural protein (such as replicase, regulatory protein) encoding the virus. Different viruses may contain multiple functionally independent gene fragments depending on their genomic characteristics (such as segmented or non-segmented, single-stranded or double-stranded, RNA or DNA).

[0081] Some examples of viral gene fragments:

[0082] The NP (nucleocapsid protein) gene fragment of avian influenza virus (such as H5N1): encodes nucleocapsid protein, responsible for wrapping viral RNA to form ribonucleoprotein complex (RNP), a key component of viral genome replication and transcription. This gene is highly conserved and is commonly used for influenza virus typing and molecular diagnosis (such as RT-PCR detection).

[0083] The HA (hemagglutinin) and NA (neuraminidase) gene fragments of influenza virus: HA is responsible for host cell receptor binding and membrane fusion, and NA assists in virus release, both of which are the core basis for influenza virus subtype classification (such as H1N1, H5N1).

[0084] The gag, pol, and env gene fragments of HIV: respectively encode viral capsid protein, reverse transcriptase / integrase, and envelope glycoprotein, which are key targets for antiviral drugs (such as reverse transcriptase inhibitors) and vaccine research.

[0085] S (spike protein) gene segment of SARS-CoV-2: encoding the viral spike protein, mediating the binding of the host cell ACE2 receptor, is the core target for the development of new crown vaccines (such as mRNA vaccines).

[0086] The study of gene segments provides a molecular basis for virus tracing (such as analyzing the host source of avian influenza by NP gene), the development of diagnostic reagents (such as HA / NA-based antigen detection), vaccine design (such as screening of S protein antigen epitopes), and the development of antiviral drugs.

[0087] In some embodiments, the research aims to provide a method for constructing an IAV (Influenza A Virus) reassortment adaptation prediction model, and using the model to predict the human adaptation of simulated reassortment sequences of avian influenza virus (Avian Influenza Virus, AIV).

[0088] The present application first uses a self-created Codon2Vec representation method, based on sliding window codon frequency statistics and transformer encoding layer pre-training for encoding representation;

[0089] Then, based on the ResNet residual network, a single gene adaptation prediction model for the PB2, PB1, PA and NP four segments is established, and the key codon region in each gene is calculated by ablation experiment;

[0090] Finally, according to the calculated key codon region, the simulated reassortment of avian influenza virus sequences and human influenza virus sequences is generated, and based on ResNet, a reassortment host adaptation prediction model for IAV is constructed, which is finally applied to the reassortment host adaptation prediction of simulated reassortment sequences.

[0091] First, the eight segments of the influenza virus genome, HA, NA, M, NS, NP, PB1, PB2, PA, encode 10 proteins respectively: hemagglutinin (HA), neuraminidase (NA), matrix proteins M2 and M1, non-structural proteins NS1 and NS2, nucleocapsid and three polymerases. The NP protein encoded by segment 5 forms the nucleoprotein (vRNP) complex with the polymerase trimer PB1, PB2, PA and viral RNA. The influenza virus RNA is transcribed and replicated in the nucleus, and one of the key processes is the nucleocytoplasmic shuttling of the nucleoprotein. During the early stages of infection, the influenza virus nucleoprotein rapidly enters the nucleus to begin transcription, as a major component of the vRNP complex, NP mediates the nuclear import of the vRNP complex through its nuclear localization signals (NLSs). The NP protein is translated from viral mRNA in the cytoplasm and enters the nucleus to facilitate viral genome replication and transcription. In summary, the influenza virus NP protein is an important viral protein that determines the host range and pathogenicity of influenza A virus.

[0092] At present, the present application is mainly aimed at the study of polymerase-related genes, and the single-gene host adaptation and reassortant host adaptation prediction models trained are also for these four genes, but the research method in the present application can be extended to other gene fragments or even other viruses, i.e., other genes or viruses can be characterized using the method, and the corresponding adaptation prediction model can be trained. For other gene fragments of IAV, they can be generated together with the polymerase-related genes for simulated reassortment, and an AI model suitable for more gene simulated reassortment adaptation prediction can be trained.

[0093] The processing flow and specific operation steps of the present application are as follows:

[0094] I. Codon2Vec characterization method

[0095] 1.1 (step 1): input the cleaned csv file containing IAV gene sequences:

[0096] The file contains multiple lines, each line including an IAV virus sequence and its corresponding host label;

[0097] The cleaned IAV gene sequence file contains N virus fragment gene sequences and their corresponding host labels.

[0098] N = 8, including segments HA, NA, M, NS, NP, PB1, PB2, PA;

[0099] Since the adaptability needs to be predicted for different segments respectively, different segments are processed respectively,

[0100] First, for the first segment, for example, the NP segment, the method of sliding window (window size (WS1) = 128) is used to traverse the NP segment gene sequence of each row, and the center codon in the current window is encoded using the codon frequency in the window. Since the types of codons are 64, each center codon is characterized as a one-dimensional vector with 64 numerical features.

[0101] As the window slides, the encoding characterization of each codon in the sequence is obtained, that is, a two-dimensional matrix of codon number x 64 is obtained, as shown in Figure 6 .

[0102] Figure 6 The gene sequence of a certain row is represented as: ATG1 GAA2 AGA3 ATA4 AAA5 GAA6 CTA7 GAT8 … TAG n

[0103] The size of the sliding window is S, and when S contains 128 codons, S = 384;

[0104] Based on the current window, the frequency of 64-bit codons is calculated, and the frequency of each codon in the current window is counted, which is calculated as follows:

[0105]

[0106] When the window size is 128 codons, S = 384, the above formula corresponds to:

[0107]

[0108] condon j represents 64 types of codons;

[0109] Therefore, the vector representation of the codon frequency is:

[0110]

[0111] represents the codon frequency vector in the window, and the dimension is 64*1; therefore, the center codon is encoded into a 1x64 vector,

[0112] For example, the codons contained in the 64th window are ATG1 GAA2 AGA3 ATA4 AAA5 GAA6 CTA7 GAT8 … GUC 63 AUA 64 AUC 65 …UUC 128 ;

[0113] The center codon in the current window is AUA 64 ,

[0114] After the above operation, with the sliding of the window, each codon in the sequence is encoded and characterized, i.e., the central codon is characterized by the codon frequency vector, and a two-dimensional matrix of the number of codons contained in the gene sequence x 64 is obtained, i.e., assuming that the gene sequence of an NP fragment contains n codons, the dimension of the encoded feature matrix is n x 64, as shown in Figure 6 .

[0115] Since we want to encode each codon, for the central codon of the window being ATG1, the sequence of the window is obtained by splicing the codons at the end of the sequence, so as to calculate the codon frequency vector, and when the central codon is the last codon, the window sequence is supplemented by cutting the first half of the sequence to calculate the codon frequency.

[0116] The determination of the sliding window is shown in the following pseudo code:

[0117] Table 1 Logic of encoding codons in sliding window

[0118]

[0119] Due to the input size requirement of the subsequent ResNet framework, zero vectors are filled at the end of the feature matrix. For example, the size of the feature matrix on the PB2 gene is 760 x 64, while the input parameter of the subsequent ResNet model framework is 128 x 128 x 3, so the size of the feature matrix on the PB2 gene is adjusted to 768 x 64, i.e., 8 x 64 zero vectors are added at the end of the original feature matrix. Finally, the representation matrix of the same gene of multiple viral sequences is stored in the npy file.

[0120]

[0121] wherein, represents the codon frequency vector when the first codon in the sequence is the central codon; ORF represents the gene fragment sequence, and A(ORF) represents the encoded encoded feature matrix.

[0122] In some embodiments, in order to effectively encode, i.e., reduce the 0 value in the codon frequency vector, it is recommended that the window width be greater than 64.

[0123] In some embodiments, the window width of the sliding window is 128 codons.

[0124] In some embodiments, the window width of the sliding window is 100 codons.

[0125] In some embodiments, the prediction network model used replaces the RestNet network with any of the following: DenseNet, ResNeXt, Inception-ResNet, MobileNetV2, EfficientNet, Wide Residual Networks (WRN), and when applied, changes the dimension of the transformation matrix based on the requirements of the network for the input, or changes the window width of the sliding window.

[0126] 1.2 (Step 2): Input the npy file obtained in step 1, and use the transformer encoding layer to pre-train it to obtain an optimized feature encoding (with the same size as in step 1), and store it in another npy file. The transformer pre-training is performed using unsupervised training to perform an auto-encoding task to reconstruct the input data to obtain the optimized feature encoding. Figure 7 A' (ORF) is shown.

[0127] 1.3 (Step 3): Input the two npy files obtained in steps 1 and 2, and use the umap dimension reduction and hierarchical clustering methods in unsupervised learning to evaluate the representation effect of the representation matrices in the two npy files (i.e., to evaluate the representation effect of the transformer encoding layer pre-training), and use the clustering external evaluation indicators Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI) to evaluate the clustering effect, and the final evaluation result is the dimension reduction visualization result and the clustering effect score of the two representation matrices.

[0128] According to the evaluation result, the optimal npy file of the training set corresponding encoding matrix is stored, i.e., the optimal encoding feature matrix (if the transformer encoding layer pre-training effect is not good on this segment, the original representation matrix is used for representation, i.e., the npy file obtained in step 1; otherwise, the representation matrix after transformer encoding layer pre-training is used, i.e., the npy file obtained in step 2). The processing flow of steps 2-3 is shown in Figure 7 .

[0129] In some embodiments, only the clustering evaluation indicators are used to select the optimal encoding feature matrix.

[0130] In some embodiments, only the dimension reduction visualization result is used to select the optimal encoding feature matrix.

[0131] In some embodiments, the optimal encoding feature matrix is selected by combining multiple evaluation results, and the selection process is as shown in Figure 7As shown, the c value is updated according to different indicators, and A(ORF) or A'(ORF) is selected as the optimal encoding feature matrix according to the c value.

[0132] In some embodiments, the subsequent steps described below are performed based on the encoding feature matrix obtained by the sliding window.

[0133] In some embodiments, the subsequent steps described below are performed using the encoding feature matrix after the Transformer auto-encoding.

[0134] In some embodiments, the subsequent steps described below are performed using the optimal encoding feature matrix.

[0135] Unless otherwise specified, the encoding feature matrix A(ORF) described below is any one of the following: an encoding feature matrix, a Transformer output encoding feature matrix, and an optimal encoding feature matrix.

[0136] II. Establishing a ResNet single-segment adaptive prediction model

[0137] 2.1 (Step 4): Training of the single-gene host adaptation prediction model using the ResNet residual network framework (as shown in FIG. 2B) is performed. In this process, the label used for training is the original sequence host label. If the host is avian, the model label is "0"; if the host is human, the model label is "1". The trained model (i.e., the single-segment host adaptation prediction model) is saved and can be used for host adaptation prediction of unknown IAV. Figure 8

[0138] That is, steps 1-4 are repeated for different segments to obtain single-segment adaptive prediction models for different segments.

[0139] In some embodiments, steps 1-4 are independently performed for the four gene sequences (PB2, PB1, PA, and NP), and four single-gene adaptive prediction models are finally obtained.

[0140] In some embodiments, steps 1-4 are independently performed for the eight gene sequences, and eight single-gene adaptive prediction models are finally obtained.

[0141] III. Ablation experiment / Bayesian model to calculate the order of importance of codons in each gene segment

[0142] 3.1 (Step 5): The trained single-segment host adaptation prediction model is used to predict the encoding feature matrix of the test set (the test set is processed in the same way as the training set to obtain the encoding feature matrix of the test set), and the current prediction accuracy (acc1) is obtained. This will be used in the subsequent step 6 to observe the impact of the mask on the accuracy. ​

[0143] Step 6: In order to evaluate the importance of each codon in the gene fragment, the encoding feature matrix of the test set (for example, the gene fragment contains 700 codons, and the encoding feature matrix has a dimension of 700 x 64) is subjected to a sliding window mask in the row direction of the matrix (window size = 80), that is, the encoding sequences of the corresponding 80 codons in the sliding window are covered, that is, the codon frequency vectors representing the codons in the sliding window region are masked as zero vectors. Since the codon frequency vectors in the window are zero vectors, the codon mask is as shown in Figure 10 or Figure 11 The encoding feature matrix after masking is shown in, and part of the vector is represented as a zero vector. When the encoding feature matrix is subjected to a sliding window mask, the judgment of the corresponding mask region is shown in Table 2.

[0144] Table 2 Judgment logic of sliding window mask encoding feature matrix (pseudo code)

[0145]

[0146] The codon feature vector in the first window of the mask is a zero vector, and the encoding feature matrix A(ORF) of the first window of the mask is obtained mask1 ,

[0147] For the second window, A(ORF) is obtained mask2 ,

[0148] For the nth window, A(ORF) is obtained maskn ,

[0149] Finally, for a gene fragment containing n codons, n encoding feature matrices after masking are obtained. These n encoding feature matrices after masking are input into the pre-trained Transformer representation matrix to obtain optimized feature encoding. After obtaining the optimal feature matrix in the manner of step 3, it is input into the trained single-fragment host adaptability model for prediction to obtain the accuracy corresponding to each mask feature matrix (acc2, acc2 is a list). Compare acc1 with acc2, and the difference between the two can reflect the importance of the codon vector in the current mask region. If the difference between acc2 and acc1 is large, it means that the 80 codons in this part of the vector have a large impact on the accuracy after being masked, that is, the importance of the codon vector represented by this part of the vector is high.

[0150] Then the sliding window, with the sliding of the window, according to the difference between acc2 and acc1 after the prediction of the corresponding area vector mask, the importance of each block area vector is calculated in turn. Since the importance obtained at present is for the vector in a whole area mask, according to the window size (80) the importance of the codon frequency vector in the area is averaged to obtain the importance of each vector (Vector importance, VI), and then according to the sliding window size (128) in the representation process of step 1, the importance of each vector is averaged to obtain the importance of each codon in the sequence (Codon importance, CI) - CI1, the specific details are as follows Figure 9 , as shown in Table 3:

[0151] Table 3 Codon importance calculation process

[0152]

[0153]

[0154] Where acc n represents the accuracy of the n th sliding window position, VI n represents the importance of the n th vector, CI n represents the importance of the n th codon.

[0155] 3.2 (Step 7:) Calculate the importance of each codon using the Bayesian model (CI2):

[0156] For the difference in the distribution of 64 codons of human and avian influenza viruses on the specified viral gene and the specified gene site, the host prior (P(H)) of the virus, the joint distribution of the virus host and a certain codon (P(H,C) can be calculated., The host conditional probability of this kind of codon can be calculated: (P(C|H) = P(H,C) / P(H).

[0157] As Figure 10 shown, the encoded feature matrix is input into the single fragment host adaptability prediction model to obtain the accuracy acc1;

[0158] After the sliding window mask of the encoded feature matrix, the encoded feature matrix of the i th window mask is obtained in turn, and the single fragment host adaptability prediction model is input to obtain the accuracy acc2;

[0159] Where, acc2 is a list, and the elements in the list respectively represent the accuracy obtained when the i th window mask is:

[0160] acc2 = [acc21, acc22, …, acc2 i,…,acc2 n ]

[0161] Among them, acc2 i The accuracy is obtained by inputting the encoded feature matrix of the i-th window of the mask into the single-fragment host adaptive prediction model; substituting acc2 into the Bayesian model yields the Bayesian accuracy acc3, i.e., P(B i |A)

[0162]

[0163] Among them, P(B) i |A) represents B based on condition A. i The probability of occurrence, where event A represents the prediction accuracy; B i This represents the encoding feature matrix of the i-th window of the event mask;

[0164] P(B i ) represents the probability of the encoded feature matrix appearing in the i-th window of the mask. In our scenario, since we use a sliding window mask with a step size of 1 codon, therefore, n is the number of codons;

[0165] A represents the prediction accuracy, A|B i P(A|B) represents the prediction accuracy obtained at the i-th window of the mask. i ) represents event B i The probability of feature matrix A under given conditions, here acc2 i ;

[0166] Therefore, the posterior probability of acc2 (post-acc2), i.e., acc3, is expressed as:

[0167]

[0168] acc2 i The accuracy is obtained by inputting the encoded feature matrix of the i-th window of the mask into the single-fragment host adaptive prediction model.

[0169] acc3 i This represents the posterior accuracy of the single-fragment host adaptive prediction model after inputting the encoded feature matrix of the i-th window of the mask.

[0170] acc3 = [acc31, acc32, ..., acc3] i ,…,acc3 n ]

[0171] The posterior importance of the mask window is obtained based on the difference between acc3 and acc1: that is, replace acc-list in Table 3 with acc3-acc1, and then calculate CI2 based on the codon importance calculation process shown in Table 3.

[0172] Step 8: Average the codon importance calculated in steps 6 and 7, as shown in Equation 1, to obtain the final importance of each codon, as shown in Figure 10 , normalize it, and save it to a csv file.

[0173]

[0174] The ablation experiment of the 4 gene fragments requires independent running of steps 5-8, and finally obtains the codon importance results on the 4 gene fragments.

[0175] In some embodiments, only the CI calculated in step 6 is used to obtain the codon importance.

[0176] In some embodiments, the codon importance is obtained based on the Bayesian corrected CI2.

[0177] In some embodiments, the codon importance is obtained based on the mean or weighted combination of the codon importance (CI1) in step 6 and the Bayesian corrected CI2.

[0178] Four, key codon combination

[0179] Step 9: According to the evaluation results of step 3, input the optimal encoding feature matrix of the 4-fragment training set, and according to the csv file of the codon importance results obtained in step 8, the key codons important for adaptation in the 4 fragments are intercepted and combined.

[0180] That is, according to the calculated codon importance in the 4 genes, compare it with the key codons that can be found in the literature, select the codon importance threshold under the premise of containing as many key codons as possible found in the literature, intercept the vector representation of the codons greater than the importance threshold from the representation matrix, and arrange them in order according to their order in the fragment, respectively. Get the representation of the key codons of the 4 genes, and finally combine the 4 fragments in the order of PB2, PB1, PA, and NP. According to the needs, different importance thresholds can be selected, but after the threshold is selected, there is only one possibility for the splicing result.

[0181] According to the size requirements of the subsequent model, the above combined representation matrix is filled with size (synchronous step 1), and the encoding matrix of the 4-gene key codons of the representation training set is obtained, as shown in Figure 12 as shown in the upper half.

[0182] Based on this representation and the ResNet residual network framework in step 4, a reassortment adaptive prediction model is trained (here, a self-tandem representation of 4 gene fragments of the same strain of virus in the training set is used, with the same labels as the original strain), such as... Figure 13 As shown, this is used to predict the reassortment host adaptability of avian influenza viruses after 2020.

[0183] Step 10: Since this invention predicts host adaptability for reassortment sequences that do not exist in nature, it is necessary to generate representations of all possible simulated reassortment sequences of avian influenza based on existing avian influenza virus sequences. Input the 4-segment coding feature matrix of the prediction set. Based on the codon importance results CSV file obtained in Step 8, extract the key codon representations of the 4 segments respectively, and randomly replace the key codon coding matrices of the 4 sequence segments of the selected human influenza virus model strain, thereby generating the feature representation of the simulated reassortment sequence, such as... Figure 13 As shown, save it to an npy file.

[0184] Step 11: Input the npy file obtained in Step 10, and use the reassortment adaptability prediction model trained in Step 9 to perform simulated reassortment host adaptability prediction, obtaining the adaptive prediction results of the simulated reassortment, i.e., the predicted host label (adapted to humans or adapted to birds) of the simulated reassortment sequence. Analyzing the prediction results of each simulated reassortment can reveal which segments of avian influenza virus have a higher risk of reassorting to humans, and which subtypes have a higher risk of simulated reassortment to humans, providing guidance for the prevention and control of influenza A virus. Furthermore, this method can be applied to the reassortment adaptability prediction problem of other viruses, or extended to the reassortment monitoring of more gene segments.

[0185] In the reassortment adaptability prediction problem of some other viruses, if the fragments contained are different from those of IAV, after collecting data, the above method is used to predict the adaptability of the fragments, determine the codon importance ranking of the fragments, screen out the key codons, train the reassortment adaptability prediction model of the virus, and simulate the adaptability of different reassortment results after reassortment, so as to carry out targeted prevention.

[0186] Figure 3 This is a schematic diagram of a computer device provided in an embodiment of the present invention, such as... Figure 3 As shown, the device 2000 may include: one or more processors 2010 and one or more memories 2020; wherein the memories store computer-readable code that, when run by the one or more processors, can perform the methods described above.

[0187] The processor in this embodiment can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, operations, and logic block diagrams disclosed in this embodiment. The general-purpose processor can be a microprocessor or any conventional processor, and can be based on an x86 or ARM architecture.

[0188] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0189] For example, the method or apparatus according to embodiments of this disclosure can also be used by means of Figure 4 The architecture of the computing device 3000 shown is used for implementation. For example... Figure 4 As shown, the computing device 3000 may include a bus 3010, one or more CPUs 3020, a read-only memory (ROM) 3030, a random access memory (RAM) 3040, a communication port 3050 connected to a network, an input / output component 3060, a hard disk 3070, etc. The storage devices in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used for processing and / or communication of the methods provided in this disclosure, as well as program instructions executed by the CPU. The computing device 3000 may also include a user interface 3080. Of course, Figure 4 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 4 One or more components in the computing device shown.

[0190] This invention also includes a computer-readable storage medium, such as... Figure 5As shown, the computer storage medium 4020 stores computer readable instructions 4010. The computer readable instructions 4010, when executed by a processor, can implement the method according to the embodiments of the present disclosure described with reference to the above drawings. The computer readable storage medium in the embodiments of the present disclosure can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. The non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM) used as an external cache. By way of example, and not limitation, many forms of RAM can be used, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct Rambus dynamic random access memory (DR RAM). It should be noted that the memory of the method described herein is intended to include, but not be limited to, these and any other suitable types of memory. It should be noted that the memory of the method described herein is intended to include, but not be limited to, these and any other suitable types of memory.

[0191] The embodiments of the present disclosure also provide a computer program product or a computer program, which, when executed by a processor, implements the steps of the above method. Figure 2 As shown, the computer program product or the computer program includes:

[0192] The acquisition module 201 is configured to acquire a training set virus single gene dataset and a host label.

[0193] The feature encoding module 202 is configured to sequentially encode single genes in the virus single gene dataset to obtain an encoded feature matrix set. The encoding is as follows: the gene sequence of the single gene is connected at the head and tail, and then a sliding window method is used to traverse the gene sequence. The frequency of 64 different codons in the window forms a window codon frequency vector. The window codon frequency vector represents the central codon of the window. With the sliding of the window, each codon of the gene sequence is represented as the window codon frequency vector when the codon is the central codon. After the traversal, an encoded feature matrix formed by the window codon frequency vectors is obtained.

[0194] The training module 203 is configured to input the coding feature matrix set into a neural network model to obtain a predicted label, and iteratively train a single-gene host adaptability prediction model based on a comparison between the predicted label and a host label.

[0195] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible architectural, functional, and operational scenarios of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowcharts or block diagrams can represent a module, a program segment, or a portion of code that contains one or more executable instructions for implementing the specified functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in a different order than that noted in the accompanying drawings. For example, two blocks that are shown in succession can actually be executed substantially in parallel, or they can be executed in reverse order according to the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0196] Generally, the various example embodiments of the present disclosure can be implemented in hardware or special-purpose circuits, software, firmware, logic, or any combination thereof. Certain aspects can be implemented in hardware, while other aspects can be implemented in firmware or software which can be executed by a controller, microprocessor or other computing device. While aspects of the present disclosure have been described above, it is to be understood that person skilled in the art, on consideration of this description, can make modifications and / or improvements within the scope and spirit of the present disclosure. When aspects of the present disclosure are illustrated or described as a flowchart, a flow diagram, or using some other graphical representation, it is to be understood that the described blocks, devices, systems, techniques or methods can be implemented in hardware, software, firmware, special-purpose circuits or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0197] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, devices and units can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0198] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the described device embodiments are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0199] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments of the present application.

[0200] In addition, each functional unit in the various embodiments of the present application can be integrated in a processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware, or in the form of a software functional unit.

[0201] The example embodiments of the present disclosure described in detail above are merely illustrative, rather than limiting. Those skilled in the art should understand that various modifications and combinations can be made to the embodiments or features thereof without departing from the principles and spirits of the present disclosure, and such modifications should fall within the scope of the present disclosure.

Claims

1. A method for constructing a single-gene host adaptability prediction model for influenza virus, characterized in that, The method includes: Obtain the training set viral single-gene dataset and host labels. The single genes in the viral single gene dataset are encoded sequentially to obtain a set of encoding feature matrices. The encoding is as follows: the gene sequences of the single genes are concatenated end to end and the gene sequences are traversed using a sliding window method. The frequencies of 64 different codons within the window constitute a window codon frequency vector. The window codon frequency vector represents the central codon of the window. As the window slides, each codon of the gene sequence is represented by the window codon frequency vector when that codon is the central codon. After the traversal is completed, the encoding feature matrix composed of the window codon frequency vectors is obtained. The encoded feature matrix set is input into a neural network model to obtain a predicted label. Based on the comparison between the predicted label and the host label, an iterative training is conducted to obtain a viral single-gene host adaptability prediction model. The method includes: the center codon of the sliding window represents the codon located at the center of the sliding window; The width of the sliding window is S codons. If S is an odd number, the center codon is the (S+1) / 2th codon in the window. If S is an even number, the center codon is the S / 2th codon or the S / 2nd codon in the window. When the index of the central codon in the gene sequence is less than S / 2 or less, the missing codon in the sliding window is filled with a codon from the end of the gene sequence. When the index of the central codon in the gene sequence is greater than nS, the missing codon in the sliding window is filled with a codon from the beginning of the gene sequence, where n represents the total number of codons in the gene sequence. The 64 different codons are the 64 codons of the standard genetic code. The second coding feature matrix set is obtained by performing autoencode learning on the coding feature matrix set. The coding feature matrices in the second coding feature matrix set are corrected to the same dimension as the coding feature matrices by padding with zeros to obtain the third coding feature matrix set. The third coding feature matrix set and the host label are input into the neural network model for iterative training to obtain the virus single gene host adaptive prediction model. After evaluating the representation effect of the matrices in the coding feature matrix set and the third coding feature matrix set, the matrix with the best representation effect is selected as the optimal coding feature matrix. The optimal coding feature matrix of the viral single gene in the training set constitutes the optimal coding feature matrix set. The optimal coding feature matrix and the host label are input into the neural network model for iterative training to obtain the viral single gene host adaptability prediction model. The training set of viral single genes consists of any one of the following genes: HA, NA, M, NS, NP, PB1, PB2, PA. After training, the corresponding viral single gene host adaptability prediction model is obtained.

2. The method for constructing the influenza virus single-gene host adaptability prediction model according to claim 1, characterized in that, The dimension of the encoding feature matrix is ​​n×64, where n represents the total number of codons in the gene sequence.

3. The method for constructing the influenza virus single-gene host adaptability prediction model according to claim 1, characterized in that, The first row of the encoding feature matrix is ​​the window codon frequency vector when the first codon of the gene sequence is the center codon, and the last row is the window codon frequency vector when the last codon of the gene sequence is the center codon.

4. The method for constructing the influenza virus single-gene host adaptability prediction model according to claim 1, characterized in that, The step size of the sliding window method is one codon.

5. The method for constructing the influenza virus single-gene host adaptability prediction model according to claim 1, characterized in that, Autoencoder learning is based on one or more of the following models: LTSM, Transformer, Transformer-XL, MLP, and MLP-Mixer.

6. The method for constructing the influenza virus single-gene host adaptability prediction model according to claim 1, characterized in that, The second set of encoded feature matrices is obtained by performing autoencoder learning on the set of encoded feature matrices, and at the same time, the trained autoencoder is obtained.

7. The method for constructing the influenza virus single-gene host adaptability prediction model according to claim 1, characterized in that, If the encoded feature matrix does not match the input dimension of the autoencoder model, it is padded with zeros at the end to meet the requirements.

8. The method for constructing the influenza virus single-gene host adaptability prediction model according to claim 1, characterized in that, The evaluation of the representation effect can be one or more of the following: clustering evaluation, UMAP dimensionality reduction evaluation.

9. A method for predicting the host adaptability of a single gene in influenza virus, characterized in that, Obtain the coding feature matrix of a single viral gene; The host fitness of the viral single gene is obtained by inputting the encoded feature matrix into a single-gene host fitness prediction model constructed using the method described in any one of claims 1-8.

10. A method for sequencing the importance of codons in an influenza virus gene sequence, characterized in that, The method includes: S1: Obtain the coding feature matrix set of the virus single gene in the test set and the host tag; S2: Input the coding feature matrix set of the test set into the viral single-gene host adaptability prediction model constructed using the method described in any one of claims 1-8 to obtain the predicted host label, and obtain the first prediction accuracy by comparing the predicted host label with the host label of the test set. S3: After masking the first window of the coding feature matrix set of the test set using the sliding window method, the first window of the masked coding feature matrix set is obtained. The center row of the first window is the first row of the coding feature matrix set. The prediction accuracy of the first window of the masked coding feature matrix set is calculated after inputting the first window of the masked coding feature matrix set into the same single-gene host adaptability prediction model. S4: After the sliding window slides along the rows of the encoded feature matrix and repeats S3, the second prediction accuracy is obtained. The second prediction accuracy includes the prediction accuracy of the masked first window to the nth window, where n represents the number of codons in a single gene. S5: Obtain the importance of the region vectors from the first to the nth window based on the difference between the second prediction accuracy and the first prediction accuracy; S6: The importance of each codon is obtained by averaging the importance of the region vectors containing each codon.

11. The method for ordering the importance of codons in the influenza virus gene sequence according to claim 10, characterized in that, The coding feature matrix set of the test set is the optimized coding feature matrix set obtained by inputting the coding feature set of the test set into the autoencoder constructed through autoencoder learning of the coding feature matrix set of the training set.

12. The method for ordering the importance of codons in the influenza virus gene sequence according to claim 10, characterized in that, The set of coding feature matrices for the test set is the set of optimal coding feature matrices obtained by evaluating the representation effect of the matrices in the set of coding feature matrices and the optimized set of coding feature matrices.

13. The method for ordering the importance of codons in the influenza virus gene sequence according to claim 10, characterized in that, S3 further includes: obtaining the Bayesian posterior accuracy of the first mask window based on the prediction accuracy of the first mask window and the probability of each mask window; repeating step S3 to obtain a third prediction accuracy, the third prediction accuracy including the Bayesian posterior accuracy of the first to nth mask windows, obtaining the Bayesian region vector importance of the first to nth windows based on the difference between the third prediction accuracy and the first prediction accuracy; and averaging the Bayesian region vector importance of each codon to obtain the Bayesian posterior codon importance of each codon.

14. The method for ordering the importance of codons in the influenza virus gene sequence according to claim 10, characterized in that, The optimal codon importance vector is obtained by weighted averaging of codon importance and Bayesian posterior codon importance.

15. The method for ordering the importance of codons in the influenza virus gene sequence according to claim 10, characterized in that, The key codons for each segment are obtained by filtering based on the set importance threshold.

16. The method for ordering the importance of codons in the influenza virus gene sequence according to claim 10, characterized in that, The Bayesian posterior accuracy of the first window of the mask, based on the prediction accuracy of the i-th window and the probability of each window in the mask, is expressed as follows: in, This represents the Bayesian posterior accuracy of the i-th window of the mask; This represents the i-th window of the event mask; This represents the probability of the i-th window in the mask; This represents the prediction accuracy of the i-th window of the mask.

17. A computer device, characterized in that, The device includes: a memory and a processor; the memory is used to store a computer program; the processor executes the computer program to implement the steps of the method according to any one of claims 1-16.

18. A computer-readable storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the steps of the method as described in any one of claims 1-16.

19. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-16.

Images