Method and device for rapid identification of antigenicity of novel coronavirus strains

Abstract

The present invention discloses a method and device for rapid antigenicity identification of a novel coronavirus strain, which is used to solve the technical problems of high time and economic cost and insufficient timely and accurate mutation analysis in the existing related technologies. The method comprises: obtaining multiple groups of strain pair sequences of the novel coronavirus, and multi-bit antigen descriptors corresponding to each group of strain pair sequences; based on multiple multi-bit antigen descriptors, converting multiple groups of strain pair sequences into antigen description data sets, and obtaining an antigen relationship prediction model through model training based on the antigen description data sets; obtaining strain sequence information, and performing hierarchical sampling on the strain sequence information to obtain sequence information to be tested; using the antigen relationship prediction model to perform antigen similarity prediction on the sequence information to be tested, and obtaining the prediction results of the antigen similarity relationship between multiple strains in the sequence information to be tested.

Description

Technical Field

[0001] The present invention relates to the field of virus research technology, and in particular to a method for rapid antigenicity identification of a novel coronavirus strain, a device for rapid antigenicity identification of a novel coronavirus strain, an electronic device and a storage medium. Background Art

[0002] At present, vaccination is the main prevention and control method to avoid infection and death from the new coronavirus (COVID-19). However, due to the continuous mutation of the new coronavirus, the virus's immune escape ability and its ability to bind to the ACE2 receptor (Angiotensin-Converting Enzyme 2, angiotensin-converting enzyme 2 / a transmembrane protein) have changed greatly, resulting in certain changes in its antigenic pattern.

[0003] Although there were few amino acid mutations in the virus in the early stage, new mutants began to appear at the end of 2020. The virus's immune escape ability and receptor binding ability have changed significantly. Based on the risk assessment of the transmission, pathogenesis and immune escape of the virus variants, the virus variants are classified into different categories.

[0004] At present, the Omicron strain is still the main epidemic strain. Compared with the wild strain, the Omicron strain has undergone more than 30 mutations. A considerable part of them are amino acid replacements in the receptor binding domain (RBD). The receptor binding domain is the main target of neutralizing antibodies. Mutations in this region are likely to lead to changes in immune escape ability. At present, many studies have pointed out that vaccines related to the original strain have little effect on mutant strains, and even have no neutralizing effect on some strains. Therefore, there is an urgent need for accurate and quantitative analysis of the antigenicity of the new coronavirus in order to better select suitable vaccine strains.

[0005] At present, most of the related research is mainly based on the data of neutralization experiments to analyze the antigenic evolution pattern of the new coronavirus. That is, through a large number of neutralization experiments, the antigenic distance between strains is calculated according to the ratio of neutralization titers to perform antigen mapping. However, this method is not only time-consuming and economically expensive, but also not timely enough when new mutants appear, and it is difficult to conduct large-scale analysis. Summary of the invention

[0006] The present invention provides a method for rapid antigenicity identification of a novel coronavirus strain, a device for rapid antigenicity identification of a novel coronavirus strain, an electronic device and a storage medium, which are used to solve or partially solve the technical problems existing in the existing related technologies, such as high time economic cost and insufficiently timely and accurate mutation analysis.

[0007] The present invention provides a method for rapid antigenicity identification of a novel coronavirus strain, comprising:

[0008] Obtain multiple sets of strain pair sequences of the novel coronavirus, and multiple antigen descriptors corresponding to each set of strain pair sequences;

[0009] Based on the multiple multi-bit antigen descriptors, the multiple groups of strain pair sequences are converted into antigen description data sets, and according to the antigen description data sets, an antigen relationship prediction model is obtained through model training;

[0010] Obtaining strain sequence information, and performing hierarchical sampling on the strain sequence information to obtain sequence information to be tested;

[0011] The antigenic relationship prediction model is used to predict the antigenic similarity of the sequence information to be tested, and the prediction results of the antigenic similarity relationship between multiple strains in the sequence information to be tested are obtained.

[0012] The present invention also provides a device for rapid identification of antigenicity of a novel coronavirus strain, comprising:

[0013] A data acquisition module, used to obtain multiple sets of strain pair sequences of the novel coronavirus, and multiple antigen descriptors corresponding to each set of strain pair sequences;

[0014] A prediction model training module, used for converting the plurality of groups of strain pair sequences into antigen description data sets based on the plurality of multi-bit antigen descriptors, and obtaining an antigen relationship prediction model through model training according to the antigen description data sets;

[0015] A hierarchical sampling module is used to obtain strain sequence information and perform hierarchical sampling on the strain sequence information to obtain sequence information to be tested;

[0016] The antigen similarity prediction module is used to use the antigen relationship prediction model to predict the antigen similarity of the sequence information to be tested, and obtain the prediction results of the antigen similarity relationship between multiple strains in the sequence information to be tested.

[0017] The present invention also provides an electronic device, the device comprising a processor and a memory:

[0018] The memory is used to store program code and transmit the program code to the processor;

[0019] The processor is used to execute the method for rapid antigenicity identification of the new coronavirus strain as described in any of the above items according to the instructions in the program code.

[0020] The present invention also provides a computer-readable storage medium, which is used to store program code, and the program code is used to execute the method for rapid antigenicity identification of the new coronavirus strain as described in any of the above items.

[0021] It can be seen from the above technical solutions that the present invention has the following advantages:

[0022] A method for rapid antigenic identification of a novel coronavirus strain is provided.

[0023] For the construction and training of the prediction model, we first obtain multiple sets of strain pair sequences of the new coronavirus and the multi-bit antigen descriptors corresponding to each set of strain pair sequences; then, based on multiple multi-bit antigen descriptors, we convert the multiple sets of strain pair sequences into antigen description data sets, and obtain the antigen relationship prediction model through model training based on the antigen description data sets. Therefore, by providing this more reference multi-bit antigen descriptor, the model can quickly capture more critical and effective feature information based on the multi-bit antigen descriptor when using the data set for prediction training, thereby improving the prediction accuracy, timeliness and effectiveness of the model prediction.

[0024] In order to process the information to be tested, the strain sequence information is obtained, and the strain sequence information is hierarchically sampled to obtain the sequence information to be tested. Thus, by hierarchically sampling the sequence information to be predicted, a sequence to be tested that can better reflect the overall sampling amount balance effect can be obtained, further making the prediction results obtained based on the model prediction more accurate.

[0025] For the prediction of antigenic relationships, the antigenic relationship prediction model is used to predict the antigenic similarity of the sequence information to be tested, and the prediction results of the antigenic similarity relationship between multiple strains in the sequence information to be tested are obtained. Since the entire prediction process is automatically implemented through machine learning algorithms, there is no need for a large number of neutralization experiments and additional mapping, which can also greatly reduce time and economic costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] 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 or the description of the prior art 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 work.

[0027] Figure 1 A flowchart of the steps of a method for rapid antigenic identification of a novel coronavirus strain;

[0028] Figure 2 A schematic diagram of an antigen epitope obtained through clustering;

[0029] Figure 3 It is a schematic diagram of the corresponding silhouette coefficient size under different K values;

[0030] Figure 4 A schematic diagram of an antigen epitope screening process;

[0031] Figure 5 A schematic diagram of the screening process for a physical and chemical property item;

[0032] Figure 6 A schematic diagram of Markov clustering results under different expansion coefficients;

[0033] Figure 7 This is a schematic diagram of the overall process of a method for rapid antigenic identification of a novel coronavirus strain;

[0034] Figure 8 This is a structural block diagram of a device for rapid antigenic identification of a new coronavirus strain. DETAILED DESCRIPTION

[0035] The embodiments of the present invention provide a method for rapid antigenicity identification of a novel coronavirus strain, a device for rapid antigenicity identification of a novel coronavirus strain, an electronic device, and a storage medium, which are used to solve or partially solve the technical problems in the existing related technologies, such as high time economic cost and insufficient and inaccurate mutation analysis.

[0036] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below 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.

[0037] As an example, at present, most related studies mainly analyze the antigenic evolution pattern of the new coronavirus based on the data of neutralization experiments. That is, through a large number of neutralization experiments, the antigenic distance between strain pairs is calculated according to the ratio of neutralization titers to perform antigen mapping. However, this method not only has high time and economic costs, but also is not timely enough when new mutants appear, and it is difficult to conduct large-scale analysis.

[0038] In addition, some related studies are not based on human data, but on animal experiments, which are essentially different from human data.

[0039] Therefore, one of the core invention points of the embodiment of the present invention is: in view of the defects and deficiencies in the prior art, a method for constructing a novel coronavirus antigen prediction model and its application are provided. (1) For the construction and training of the prediction model, firstly, multiple groups of strain pair sequences of the novel coronavirus and the multi-bit antigen descriptors corresponding to each group of strain pair sequences are obtained; then, based on the multiple multi-bit antigen descriptors, the multiple groups of strain pair sequences are converted into antigen description data sets, and the antigen relationship prediction model is obtained through model training based on the antigen description data sets. By providing such a more reference multi-bit antigen descriptor, the model can quickly capture more critical and effective feature information based on the multi-bit antigen descriptor when using the data set for prediction training, thereby improving the prediction accuracy, timeliness and effectiveness of the model prediction. Thus, by providing a more reference multi-bit antigen descriptor, the model can quickly capture more critical and effective feature information based on the multi-bit antigen descriptor when using the data set for prediction training, thereby improving the prediction accuracy, timeliness and effectiveness of the model prediction. (2) For the processing of the information to be tested, the strain sequence information is obtained, and the strain sequence information is hierarchically sampled to obtain the sequence information to be tested. Thus, by performing hierarchical sampling on the sequence information to be predicted, a sequence to be tested that can better reflect the overall sampling balance effect can be obtained, further making the prediction results obtained based on model prediction more accurate. (3) For the prediction of antigenic relationships, the antigenic relationship prediction model is used to predict the antigenic similarity of the sequence information to be tested, and the prediction results of the antigenic similarity relationship between multiple strains in the sequence information to be tested are obtained. By adopting a prediction model constructed based on antigen-related features, the antigenic relationship between any two new coronaviruses can be accurately and efficiently predicted. In practical applications, only two sequences need to be input, and the antigenic similarity prediction results are quickly given through model prediction. In addition, since the entire prediction process is automatically implemented through machine learning algorithms, there is no need for a large number of neutralization experiments and additional mapping, which can greatly reduce time and economic costs. (4) Furthermore, based on a large amount of sequence data, an antigenic similarity network is constructed, and new antigenic clusters are quickly identified through clustering methods. Based on the antigenic clusters, the constructed antigenic similarity network is described as an antigenic cluster, and the dynamics of the antigenic clusters are tracked in real time. On this basis, by improving the vaccine strain recommendation strategy, efficient and rapid recommendation of vaccine strains can be achieved.

[0040] Reference Figure 1 , shows a flowchart of a method for rapid antigenicity identification of a novel coronavirus strain provided by an embodiment of the present invention, which may specifically include the following steps:

[0041] Step 101, obtaining multiple groups of strain pair sequences of the novel coronavirus, and multiple antigen descriptors corresponding to each group of strain pair sequences;

[0042] Specifically, obtaining multiple sets of strain pair sequences of the novel coronavirus can be achieved by executing the following steps S01 to S03:

[0043] Step S01: Obtain in vitro serum from people who are initially infected with different lineages of novel coronaviruses, conduct a neutralization experiment based on the in vitro serum from the people, and obtain titer values ​​corresponding to multiple groups of original strains;

[0044] Among them, the neutralization experiment (also known as the neutralization test) refers to diluting the serum containing antibodies according to a certain gradient. During the dilution process, each concentration gradient takes the same volume. The higher the dilution factor, the lower the antibody concentration and the less antibody it contains. The diluted serum is contacted with the same amount of virus for reaction. The dilution factor that can neutralize 50% of the virus is the neutralization titer, which is the titer value referred to in the embodiments of the present invention.

[0045] In this step, sera from people who were infected with different lineages of the new coronavirus for the first time were used to conduct neutralization experiments on different new coronaviruses. Among them, using sera from people who were infected for the first time can reflect the antigenicity of the new coronavirus to the greatest extent, so as to exclude the influence of cross-protection caused by multiple infections.

[0046] Step S02: for each group of original strain pairs, the antigenic relationship of the original strain pairs is determined according to a preset similarity threshold and a titer value, wherein the antigenic relationship indicates that the original strain pairs are antigenically similar or antigenically dissimilar;

[0047] For the definition of antigenic similarity, the embodiment of the present invention refers to the research method in influenza virus and selects four times the titer as the threshold of similarity and dissimilarity. Specifically, multiple neutralization experiments were performed on a serum. Among them, the strains used in the experiment include the same strains as the infectious strains, and the results obtained are homologous titer values. The strains used in the experiment also include other strains that do not infect the population, and the results obtained are heterologous titer values. Using the "homologous titer value / heterologous titer value" method, if the ratio of the two is greater than 4, the antigenic relationship between this group of strain pairs is considered to be dissimilar.

[0048] Take the new coronavirus strain BA.1 as an example. Assuming that the homologous titer value of BA.1 is about 960, and the heterologous titer value of its heterologous titer BA.4 / BA.5 is about 60, then the ratio of homologous titer value / heterologous titer value is 960 / 60=16, which is greater than the preset similarity threshold of 4. Therefore, it can be considered that the antigenic relationship between the strain pairs BA.1 and BA.4 / BA.5 is dissimilar.

[0049] Step S03: Preprocess multiple groups of original strain pairs to obtain multiple groups of strain pair sequences of the new coronavirus.

[0050] Taking the original strains BA.1 and BA.2 as an example, based on the neutralization experiment of the serum of the initial infection with BA.1, we can get a similarity relationship between BA.1 and BA.2 (similar or dissimilar). Similarly, we can also get a similarity relationship between BA.1 and BA.2 in the serum of the initial infection with BA.2, so the experimental results are repeated.

[0051] In this case, data preprocessing can be performed on multiple groups of original strains determined after the neutralization experiment, that is, sequence data without homologous titers, duplications, and contradictions can be deleted to obtain multiple groups of strain pair sequences as strain pair data for modeling. In the embodiment of the present invention, after data preprocessing, a total of 135 groups of strain pair sequences can be obtained for subsequent modeling.

[0052] In the embodiment of the present invention, for each group of strain pair sequences, their antigenic descriptors are generated by fully considering their antigenic epitopes, physicochemical properties, and differences in the distances from the mutation sites to the receptor binding domains.

[0053] Among them, antigen epitope refers to a special sequence on the viral protein, which is the site of contact with the antibody. The sequence of the viral protein is represented by letters, but it needs to be calculated with numbers. The numbers used for calculation are antigen descriptors.

[0054] Specifically, for each group of strain pair sequences, the step of generating a multi-bit antigen descriptor in the embodiment of the present invention may mainly include the following steps S11 to S16:

[0055] Step S11: predicting and screening the antigenic sites of the S protein of the strain sequence to obtain multiple initial antigenic sites;

[0056] In terms of antigen epitope screening, since the new coronavirus mainly binds to and infects cells based on the S protein (spike protein), ScanNet (a tool that can predict antigen sites based on protein structure) is first used to predict the antigen sites on the S protein. Then, antigen sites with a screening prediction probability greater than 0.15 are retained. FreeSASA (a program for calculating solvent accessible surface area (SASA)) is then used to calculate the solvent accessible surface area of ​​these antigen sites, and only the initial antigen sites with a solvent accessible surface area greater than 0.3 are retained, a total of 164.

[0057] Among them, the solvent accessible surface area is the surface area of ​​biological molecules (residues) that can be contacted by the solvent. This indicator can be used to determine whether the antigen site is exposed. The larger the solvent accessible surface area, the greater the degree of exposure of the residue, and the more likely it is to become a site for antibody or receptor binding.

[0058] It is known that antigen epitopes are exposed amino acid sets with strong immunogenicity. Since the current tools for predicting antigen epitopes are not mature and accurate enough, if only ScanNet tools are used for prediction, it is very likely that some unexposed amino acids will be predicted. Theoretically, the possibility that these unexposed amino acids belong to antigen epitopes is extremely small.

[0059] Therefore, in order to better define the antigenic epitope, the embodiment of the present invention further determines whether the antigenic site screened by the prediction tool ScanNet is exposed by combining the solvent accessible surface area. If it is considered that there are unexposed antigenic sites after further determination, they are deleted. Thus, through the preliminary prediction screening of the prediction tool and the secondary screening confirmation of the solvent accessible surface area, the prediction of the antigenic epitope can be made more accurate, and a more referenceable antigenic epitope can be obtained.

[0060] Step S12: performing cluster calculation based on the multiple initial antigen sites to determine multiple antigen epitopes that meet the preset site screening conditions;

[0061] Then, the K-means algorithm was used for clustering calculation, wherein the features used for clustering calculation were the three-dimensional coordinates of each amino acid in 7DZW.pdb (a data file for storing protein three-dimensional structure information).

[0062] In the embodiment of the present invention, the K values ​​of 5 to 10 are mainly considered, and the K value corresponding to the maximum silhouette coefficient is set as the site screening condition. First, the clustering results of each K value are calculated to obtain the silhouette coefficient. Then, the K value corresponding to the maximum silhouette coefficient is selected as the final selected K value. Based on the experimental data used, it can be calculated that when K=8, the silhouette coefficient is the largest.

[0063] Among them, the silhouette coefficient is an indicator used to evaluate the quality of clustering. It can be understood as an indicator that describes the clarity of the outline of each category after clustering. It includes two factors: cohesion and separation. Cohesion can be understood as reflecting the closeness between a sample point and the elements within the class. Separation can be understood as reflecting the closeness between a sample point and the elements outside the class.

[0064] Through the above clustering calculation, the embodiment of the present invention preliminarily defines 8 antigen epitopes (corresponding to the K value corresponding to the maximum silhouette coefficient finally selected), such as Figure 2 The colored balls in the schematic diagram of antigen epitopes obtained after clustering.

[0065] Figure 3 It shows a line diagram of the corresponding silhouette coefficient size under different K values. Figure 3In the figure, the horizontal axis is the K value selected during clustering calculation, and the vertical axis represents the silhouette coefficient obtained after clustering calculation. It can be seen that when K=8, the silhouette coefficient is the largest, close to 0.48.

[0066] Step S13: performing secondary screening on the multiple antigen epitopes to obtain multiple target antigen epitopes, and generating first antigen descriptors corresponding to the respective target antigen epitopes;

[0067] Then, the spatial distribution results of the antigen epitopes can be further checked, and the 8 antigen epitopes obtained in the above steps can be screened again. Specifically, some outlier antigen sites (i.e., amino acids) are eliminated, and epitopes that have not mutated in the experimental data set (i.e., the strain pair data used) are excluded, and finally 6 target antigen epitopes for modeling are determined.

[0068] like Figure 2 As shown in the figure, each ball represents an amino acid, and an epitope is a collection of multiple amino acids. For example, epitope A is composed of amino acids 59, 60, 630, 631, 632, and 208. The gray ones in the figure are other amino acids, and the colored ones are antigen epitope amino acids.

[0069] From the definition of epitope, epitope should be amino acids clustered together. Figure 2 The purple amino acids shown in the red box are not clustered together with most of the other purple amino acids. It can be seen that the possibility of these outlier purple amino acids belonging to the purple epitope is very small. In order to make the definition of the antigen epitope more accurate, they are removed.

[0070] In the embodiment of the present invention, after screening, a total of 5 amino acids were eliminated (combining the actual experimental process, the amino acids 40, 744, 841, 855, and 566 were specifically deleted in this step), and these 5 sites were distributed on 2 epitopes.

[0071] Since the embodiment of the present invention predicts antigenic epitopes by prediction tools, most antigenic sites can be predicted in theory. In actual situations, even if the antigenic epitopes that have not mutated are retained, the model cannot learn their influence on antigenicity. On the contrary, it may affect the performance of the model. Therefore, the deletion operation is also performed for the antigenic epitopes that have not mutated. In the embodiment of the present invention, 2 antigenic epitopes that have not mutated are deleted, and finally 6 target antigenic epitopes for modeling can be obtained.

[0072] In order to enable those skilled in the art to better understand the relevant technical solutions of the present invention, Figure 4 A schematic diagram of an antigen epitope screening process provided by an embodiment of the present invention is shown.

[0073] First, step S11 is performed to obtain the intersection of the ScanNet and FreeSASA screening results, and 164 amino acids can be obtained, thereby determining that multiple sets of 164 amino acids may be used as antigen epitopes. Then, based on the K-means clustering algorithm, clustering is performed according to the site coordinates, and the above 164 amino acids are divided into 8 clusters (such as Figure 4 The colored balls of different colors in the sample are the 8 epitopes. Then, the spatial distribution of the epitopes was further checked and it was found that 5 amino acids could not be clustered together in space (outlier sites), so they were deleted. Based on the remaining 159 amino acids, 8 epitopes were defined.

[0074] Since 2 of the antigenic epitopes are too "conservative", no mutation occurs in the experimental data, so these 2 antigenic epitopes are deleted, and finally 6 target antigenic epitopes are obtained. Based on these 6 target antigenic epitopes, six antigenic descriptors can be generated. In order to distinguish the antigenic descriptors generated here from the antigenic descriptors generated by subsequent different processes, the antigenic descriptors generated based on each target antigenic epitope in this step are used as the first antigenic descriptors, and these 6 first antigenic descriptors are used together as the six antigenic descriptors corresponding to the entire antigenic epitope screening process.

[0075] Step S14: Screening out several categories of physicochemical properties from the physicochemical property database, determining the physicochemical property items corresponding to each category of physicochemical properties based on the strain pair sequence and combined with random forest classification prediction, and generating second antigen descriptors corresponding to each physicochemical property item respectively;

[0076] In terms of screening of physicochemical property items, the embodiments of the present invention refer to five major categories of physicochemical properties commonly used in influenza virus antigen analysis, namely: hydrophobicity, charge, polarity, volume, and accessible surface area.

[0077] First, we can refer to the PREDAC-CNN (predicting antigenic clusters of seasonal influenza A viruses with convolutional neural network) method to select entries that can characterize these five types of physical and chemical properties from the physical and chemical properties database (i.e., the AAindex database).

[0078] Among them, the AAindex database is a database used to store physicochemical property entries and their quantitative representations. The AAindex database contains quantitative representations of more than 500 physicochemical property entries. In biochemistry and molecular biology, in order to facilitate the understanding and analysis of the relationship between protein structure and function, as well as to predict protein behavior and interactions, for the 20 amino acids involved in the experiment, the physicochemical properties of each amino acid (hydrophobicity, charge, polarity, volume, etc. as mentioned above) can be represented by one or a group of specific numerical values. This value represents the situation of this amino acid in this physicochemical property. However, for the same physicochemical property, such as hydrophobicity, there may actually be multiple entries that can reflect it. In other words, each type of physicochemical property corresponds to at least one candidate entry.

[0079] For physical and chemical properties with more than 1 candidate items, a default random forest classification model can be established based on each type of feature; then the physical and chemical properties with multiple candidate items are input into the corresponding random forest classification model respectively, the characteristics of this category are calculated and classification predictions are performed; for each type of physical and chemical property, the candidate item with the largest feature contribution is selected as the representation of this type of physical and chemical property.

[0080] The calculation method of this type of feature is: first calculate the absolute value of the difference in the specific properties of the different amino acids in a pair of sequences, and then take the average. If the number of different amino acid sites is greater than 3, only the top 3 with the largest difference are considered in the calculation. When a site on one sequence is missing and the corresponding site on the other is not missing, the specific property of the site with the largest numerical difference among the remaining 19 amino acids is used as the property characterization of the missing site.

[0081] Feature contribution measures the importance and influence of each feature in the model prediction process. This indicator is usually obtained by calculating the importance score of each feature in all decision trees in the forest and then taking the average. The larger the feature contribution, the more important the feature is in the classification process.

[0082] It is understandable that this type of features tends to describe the inherent properties of a type of sample, while feature contribution focuses on analyzing the role and influence of features in the prediction model. In the actual prediction process, identifying this type of features helps data preprocessing and feature selection, while analyzing feature contribution helps model interpretation and optimization.

[0083] After the physicochemical property items corresponding to each type of physicochemical property are determined by the classification prediction of the random forest, the second antigen descriptors corresponding to each physicochemical property item can be generated respectively. In the embodiment of the present invention, a total of five major types of physicochemical properties are considered, and the second antigen descriptors corresponding to the five physicochemical property items can be generated, and these five second antigen descriptors are used as the five-bit antigen descriptors corresponding to the entire physicochemical property item screening process.

[0084] In combination with the above-mentioned physical and chemical property related processing steps, illustratively, Figure 5 A schematic diagram of a physical and chemical property item screening process provided by an embodiment of the present invention is shown. Figure 5 It can be seen that after screening, the antigen descriptors corresponding to each physicochemical property can be obtained. For example, the antigen descriptor corresponding to the charge physicochemical property is KLEP840101, and the antigen descriptor corresponding to the accessible surface area physicochemical property is RADA880106, etc. These five antigen descriptors are collectively regarded as five-bit antigen descriptors.

[0085] Step S15: performing distance difference analysis from the mutation site to the receptor binding region on the strain sequence, determining the distance difference feature, and generating a third antigen descriptor corresponding to the distance difference feature;

[0086] In terms of calculating the difference in distance from the mutation site to the receptor binding region, for each pair of strains, the embodiment of the present invention first obtains the spatial position information (i.e., three-dimensional coordinates) of each amino acid in the pair of strains from 7DZW.pdb. Then, from related studies, the amino acids that bind to the S protein and the ACE2 receptor are determined. Studies have shown that the contact surface can be divided into three clusters, the first of which is "Gln498, Thr500, and Asn501", the second is "Lys417, Tyr453", and the third is "Gln474 and Phe486". The three letters in the above symbols represent the types of amino acids, and the numbers represent the amino acid numbers on the strains.

[0087] For each pair of strains, the distances from all differential amino acids to the above three clusters (which can be understood as sites in this case) are first calculated, and the average is taken as the distance difference feature of the pair of strains. If there are more than three different amino acids in the pair of strains, only the three closest values ​​are retained, and the average is taken based on these three values.

[0088] After the distance difference feature is determined, an antigen descriptor corresponding to the distance difference feature may be generated as a third antigen descriptor corresponding to the distance difference analysis process.

[0089] Step S16: The third antigen descriptor, each second antigen descriptor and each first antigen descriptor are taken together as a multi-bit antigen descriptor corresponding to the strain pair sequence.

[0090] In an embodiment of the present invention, through the above-mentioned screening steps, one one-digit antigen descriptor (corresponding to the distance difference feature), one five-digit antigen descriptor (corresponding to the physical and chemical property item feature) and one six-digit antigen descriptor (corresponding to the antigen epitope feature) can be obtained. For this pair of strain sequences, the corresponding one is a twelve-digit antigen descriptor.

[0091] Since each group of strain pairs uses the same antigen descriptor generation steps, each group of strain pairs can generate their own corresponding 12-bit antigen descriptors through the above antigen descriptor generation steps. Since the relevant processing steps are the same, they will not be repeated here.

[0092] Step 102, based on the plurality of multi-bit antigen descriptors, converting the plurality of groups of strain pair sequences into antigen description data sets, and obtaining an antigen relationship prediction model through model training according to the antigen description data sets;

[0093] For the modeling process, all data are first represented as twelve-bit antigen descriptors, that is, converted into antigen description data sets. Then they are divided into training sets and test sets in a ratio of 8:2. Then, ten-fold cross validation and grid search are used to select the optimal parameters on the training set, and then verified on the test set. In this step, four classification prediction models, XGBoost (eXtreme GradientBoosting, an optimized distributed gradient boosting (GradientBoosting) framework), RF (RandomForest, Random Forest), MLP (MultilayerPerceptron) and LR (LogisticRegression, Logistic Regression), are mainly considered as model frameworks for training and verification of antigen relationship prediction, and the best performing model is selected for subsequent analysis.

[0094] In addition, according to the actual experimental test requirements, in addition to the aforementioned prediction model framework, it is also possible to simultaneously consider using machine learning algorithms such as KNN (k-Nearest Neighbor, K nearest neighbor classification algorithm) and SVM (Support Vector Machine, Support Vector Machine) for training verification. It is understandable that the present invention is not limited to this.

[0095] Specifically, the process of obtaining the antigen relationship prediction model through model training according to the antigen description data set can be achieved by executing the following sub-steps S21 to S23:

[0096] Step S21: dividing the antigen description data set into a training set and a test set according to a preset ratio, and constructing several different initial prediction models based on different types of classification model frameworks;

[0097] Step S22: for each initial prediction model, use the training set to perform antigen relationship prediction training on the initial prediction model, and perform parameter optimization during the model training process;

[0098] Step S23: using the test set to verify each of the trained initial prediction models, and taking the trained initial prediction model with the best verification effect as the antigen relationship prediction model.

[0099] In the embodiment of the present invention, based on the training data used, a prediction model based on the random forest classification framework is finally selected for subsequent antigenic relationship prediction. Thus, based on the S protein sequence feature information, a network model for predicting the antigenic relationship of the new coronavirus is constructed, which can accurately distinguish antigenically similar virus pairs and antigenically variant virus pairs, with an accuracy rate of up to 0.81.

[0100] Step 103, obtaining strain sequence information, and performing hierarchical sampling on the strain sequence information to obtain sequence information to be tested;

[0101] First, the strain sequence information is screened from the virus database GISAID (Global Initiative on Sharing All Influenza Data, an international scientific initiative and data sharing platform), and the strain sequence information is hierarchically sampled to obtain the sequence information to be tested.

[0102] Among them, GISAID is a database for studying viruses, which provides tens of millions of genomes and S protein sequences of the new coronavirus. In the embodiment of the present invention, only the S protein sequence information of the strains with accurate country and location information in the sequence name is selected and retained.

[0103] The strain sequence information may include multiple S protein sequences containing location information (such as country or region information) and time information (such as year, month, etc.). The process of hierarchical sampling of the strain sequence information to obtain the sequence information to be tested can be achieved by executing the following sub-steps S31 to S32:

[0104] Step S31: for each region corresponding to the location information, based on the time information, the strain sequence information is sampled in the order of time nodes from far to near and corresponding sampling numbers from small to large, to obtain multiple S protein sequences of the target strain;

[0105] Specifically, information sampling can be performed in a hierarchical incremental manner based on the time information and location information, based on the country and month levels, or based on each specific region and month (or specific to the date) level.

[0106] Taking the hierarchical information sampling method based on the country and month level as an example, for example, in the period from December 2019 to February 2021, set 1 sample per country per month; in the period from March 2021 to February 2022, set 2 samples per country per month; in the period from March 2022 to February 2023, set 3 samples per country per month; in the period from March 2023 to March 2024, set 4 samples per country per month. In this way, multiple S protein sequences of the target strain can be further screened from the acquired strain sequence information for use in the subsequent strain antigen relationship prediction process.

[0107] For the sampling of data to be tested, a common method for infectious diseases is to sample according to the scale of the outbreak. In the evolution of the novel coronavirus from far to near, if this sampling method is chosen, due to the decline in monitoring efforts, there will be differences in quantity and quality between the early case data and the later case data. From the actual situation, although there are more data recorded in the early stage, which is closer to the actual evolution, the data recorded in the later stage is very small and cannot truly reflect the overall situation.

[0108] Another commonly used data sampling method is balanced sampling. That is, the same number of samples is taken in all time periods. However, in actual situations, the recorded sequences are classified. Some categories appear later. If balanced sampling is used, the number of sequences in the later categories will be sampled less, which will lead to an imbalance in the final sampling amount of each category, resulting in poor clustering effect and affecting the prediction effect.

[0109] Therefore, in order to reflect the overall effect through a small number of sequences, different from sampling according to the scale of the outbreak and directly using the strain sequence information to predict the antigen relationship, or using a balanced sampling method to process and then predict the antigen relationship, the embodiment of the present invention adopts a hierarchical incremental data sampling method for the obtained strain sequence information. That is, hierarchical sampling is performed by using a method with a small sampling amount in the front time period and a large sampling amount in the back time period. In this way, the number of sequences of each category of strains sampled can reach a certain scale, thereby solving the problem of unbalanced sampling.

[0110] In summary, based on the actual situation, by taking into account the sampling methods of countries (regions) and time nodes at the same time, the S protein sequence information of strains in various time periods around the world can be better captured.

[0111] Step S32: Perform multiple sequence alignment on the S protein sequences of multiple target strains, delete the sequences with sequencing errors, and obtain the sequence information to be tested.

[0112] In practical applications, the data obtained through sampling may contain duplication or sequencing errors. In this case, in order to avoid prediction errors in the subsequent prediction process, it is necessary to perform redundancy operations on the data. Specifically, MAFFT (a multiple sequence alignment software) can be used to perform multiple sequence alignments on the S protein sequences of multiple target strains. If a gap (empty position) on the reference strain is detected in 99% of the S protein sequences of the target strains, it is considered that the site is a sequencing error and should not be an insertion mutation. At this time, the S protein sequence of the target strain where the site is not a gap can be deleted.

[0113] Through the sequence information sampling and comparison screening process introduced in the previous steps, 4,000 S protein sequences of the target strain were finally retained.

[0114] Step 104, using the antigen relationship prediction model to predict antigen similarity of the sequence information to be tested, and obtaining prediction results of antigen similarity relationships between multiple strains in the sequence information to be tested.

[0115] In this step, based on the constructed antigen relationship prediction model, antigen similarity prediction is performed on the sequence information to be tested obtained in step 103 (i.e., the S protein sequences of 4000 target strains), and multiple groups of different antigen relationship prediction results can be obtained. Each group of prediction results is characterized by the antigenic similarity of the strains in this group, or the antigenic dissimilarity.

[0116] The embodiment of the present invention uses a machine learning model based on antigen-related features to predict antigen relationships, which can accurately and efficiently predict the antigen relationships between any strains. Compared with the traditional experimental determination and antigen mapping methods, it is lower in cost, faster, and more suitable for large-scale antigen determination.

[0117] Furthermore, based on multiple groups of different antigen relationship prediction results, an antigen network (used to reflect the antigen similarity relationship between each group of strain pairs) can be constructed to better analyze the evolution process of the new coronavirus. At the same time, vaccine strains can be recommended based on the constructed antigen network.

[0118] Specifically, the antigen network construction process can be implemented by executing the following sub-steps S41 to S43:

[0119] Step S41: Take each strain in the sequence information to be tested as a node, and take the antigenic similarity between the strains as the node edge, and construct the initial antigen network of the new coronavirus based on each node and each node edge;

[0120] In the embodiment of the present invention, Cytoscape (a software for graphically displaying networks and performing analysis and editing) is used to construct an antigen network. The new coronavirus strains are regarded as nodes, and the antigenic similarity relationship between two new coronavirus strains is used as the edge between the two nodes, thereby obtaining an original antigen network.

[0121] Step S42: performing multiple Markov clustering on the prediction results based on multiple pre-set expansion coefficients, adjusting parameters according to the calculated silhouette coefficients, and obtaining the best clustering result, the clustering result including antigen cluster information;

[0122] Based on step S41, only the antigenic relationship of all sequences can be obtained, but it is not known how many antigenic clusters the entire antigenic network can be divided into, and which antigenic cluster each sequence belongs to. Therefore, it is necessary to perform clustering calculation on the sequences in the antigenic network to obtain antigenic cluster information.

[0123] Specifically, based on the predicted results of multiple groups of strain antigen relationships, a parameter (expansion coefficient) is selected to perform Markov clustering (Markov Clustering Algorithm, MCL) on the prediction results, and the clustering results corresponding to the expansion coefficient can be output. Among them, Markov clustering will classify each sequence according to the size of the expansion coefficient. If the expansion coefficient is different, the classification results will be different, and the silhouette coefficients corresponding to the clustering results will also be different. Among them, the clustering result is how many antigen clusters the antigen network constructed based on the prediction results can be divided into.

[0124] In the embodiment of the present invention, multiple clustering results with expansion coefficients ranging from 1.0 to 1.5 and intervals of 0.01 are calculated. And by calculating the average class size under each expansion coefficient, a line graph is drawn. For example, the schematic diagram of Markov clustering results under different expansion coefficients is shown in Figure 6 As shown. Among them, Figure 6 The horizontal axis represents the expansion coefficient, and the vertical axis represents the average cluster size. Then the expansion coefficient at the first plateau is taken as the final expansion coefficient, such as Figure 6 Then, according to the clustering result corresponding to the expansion coefficient of 1.01, the antigen cluster information (i.e., how many clusters can be divided into in total and which cluster each sequence belongs to) can be obtained.

[0125] Step S43: adding the antigen cluster information to the initial antigen network to divide the initial antigen network into multiple antigen clusters to obtain a target antigen network, which is used to reflect the prevalence and transition of multiple antigen clusters in various regions.

[0126] Then, based on the clustering results, antigen cluster information can be added to the originally constructed antigen network to obtain a target antigen network with antigen cluster information.

[0127] Based on the clustering results and the S protein sequences of 4,000 target strains, the prevalence and transformation of these antigen clusters around the world and in each continent can be viewed in real time.

[0128] Therefore, in the embodiment of the present invention, based on the constructed antigen relationship model and the sequence information obtained from GISAID screening, an antigen similarity network of the new coronavirus is constructed, which can more accurately reflect the antigenic changes and evolutionary trends of the new coronavirus.

[0129] As an optional embodiment, the constructed target antigen network can assist in the recommendation of vaccine strains for the new coronavirus.

[0130] It is known that the antigenic conversion rate of the new coronavirus is very fast, and the antigenic class conversion is also very fast. Therefore, it is necessary to monitor the strains in real time and add new strains to the constructed antigen network. On this basis, the embodiment of the present invention considers recommending vaccine strains based on the antigenic class of the sequences in the past month.

[0131] Specifically, the process of recommending vaccine strains can be implemented by executing the following sub-steps S51 to S58:

[0132] Step S51: setting a recommended time point every preset time period;

[0133] For example, if a whole year is used as the major interval for the recommendation time period, and if the recommendation method is adopted with equal time periods, the vaccine strains of the new coronavirus can be recommended in January, April, July, and October of each year (or other suitable months can be selected according to actual needs). If the method of using unequal time periods for recommendation is chosen, it can be set in increasing or decreasing time periods. Taking increasing time periods as an example, the vaccine strains of the new coronavirus can be recommended in January, March, June, and October of each year (the time period increases in February, March, and April).

[0134] Regarding the setting of the interval time period and the recommended time point of the vaccine strain, those skilled in the art can set it according to the actual situation. It is understandable that the present invention does not limit this.

[0135] Step S52: Analyze the antigenic classes of all historical virus strain sequences that appeared before the current recommended time point, and predict the antigenic similarity between all historical virus strain sequences based on the antigenic relationship prediction model;

[0136] Step S53: Perform Markov clustering calculation based on the antigen similarity prediction results, combine the antigen class analysis results, determine all the epidemic classes currently appearing in the new coronavirus, and generate the current antigen network;

[0137] The generation of the antigen network can be carried out by referring to the generation steps of the target antigen network in the above-mentioned embodiment, which will not be described in detail here.

[0138] Step S54: Obtain the historical antigen network generated corresponding to the last recommended time point, compare the current antigen network with the historical antigen network, and determine the newly added strains and corresponding strain types from the last recommended time point to the current recommended time point;

[0139] If the previous recommended time point was January and the current recommended time point is April, it is necessary to determine the new strains that have been added within the three months from January to April and which strain category each new strain belongs to.

[0140] Step S55: Calculate the intra-class degree value of each newly added strain and the same type of strains respectively, where the intra-class degree value indicates the number of strains in the current antigen network that are similar in antigenicity to the newly added strain and belong to the same type;

[0141] Step S56: If there is a new antigen class in the current antigen network whose prevalence exceeds the preset prevalence threshold, then the strain with the largest intraclass degree value is selected from the new antigen class as the recommended vaccine strain;

[0142] When a new antigen class is found in the clustering results and its prevalence exceeds 15% (this value can also be set according to actual application requirements or virus analysis experience), the strain with the largest intra-class degree value that appeared between the last recommended time point and the current recommended time point (such as within the nearly 3 months from January to April) is selected from the new antigen class as the recommended vaccine strain.

[0143] Step S57: If there are two or more new antigen classes in the current antigen network whose prevalence exceeds the preset prevalence threshold, the strain with the largest intraclass value is selected from the new antigen class with the highest prevalence as the recommended vaccine strain;

[0144] Although the possibility is small, assuming that there are two or more new antigen classes in the current antigen network whose prevalence exceeds the preset prevalence threshold (15%), the strain with the largest intraclass degree value is selected from the new antigen class with the highest prevalence as the recommended vaccine strain. For example, assuming that compared with the statistical analysis in January, a new antigen class A with a prevalence of 17% and a new antigen class B with a prevalence of 19% appeared after statistical analysis in April, then the strain with the largest intraclass degree value is selected from the new antigen class B as the recommended vaccine strain.

[0145] Step S58: If there is no new antigen class in the current antigen network, or there is no new antigen class whose prevalence exceeds the preset epidemic threshold, the strain with the largest intra-class degree value is selected from the epidemic class with the highest prevalence as the recommended vaccine strain.

[0146] When no new antigenic class has emerged between the last recommended time point and the current recommended time point, or the prevalence of the newly emerged antigenic class is low (such as less than 15%), the strain with the largest intraclass value is selected from the epidemic class with the highest prevalence as the recommended vaccine strain.

[0147] More specifically, when no new antigen class appears between the last recommended time point and the current recommended time point, the strain with the largest intra-class value can be selected as the recommended vaccine strain from the epidemic class with the highest prevalence rate counted so far, or in the last 3 or 6 months.

[0148] When a new antigen class appears between the last recommended time point and the current recommended time point, but the prevalence is low, the strain with the largest intra-class value can be selected as the recommended vaccine strain from the epidemic class with the highest prevalence counted so far, or in the last 3 or 6 months.

[0149] Although the possibility is small, when two or more new antigenic classes appear between the last recommended time point and the current recommended time point, but the prevalence rates are low (for example, all below 15%), two methods can be used to recommend vaccine strains:

[0150] One way is to directly select the antigen class with the highest prevalence from the newly emerged antigen classes, and select the strain with the largest intraclass value as the recommended vaccine strain. For example, if there is a new antigen class C with a prevalence of 14% and a new antigen class D with a prevalence of 12%, the strain with the largest intraclass value can be selected from the new antigen class C as the recommended vaccine strain.

[0151] Another way is to consider different time spans and select the strain with the largest intraclass value as the recommended vaccine strain from the epidemic category with the highest prevalence rate in the statistics so far or in the recent 3 months or 6 months. For example, if the epidemic category with the highest prevalence rate in the recent 3 months is antigen category E, with a prevalence rate of 17%, then the strain with the largest intraclass value in antigen category E can be selected as the recommended vaccine strain.

[0152] Therefore, based on different actual situations, the epidemic trends of the new coronavirus can be fully considered and analyzed, and a more flexible and complete vaccine strain recommendation strategy can be given.

[0153] According to continuous monitoring data, the representative strains determined each year show strong antigenic similarity with the epidemic virus. It can be shown that these strains have a wide range of protective potential as ideal candidate vaccines. Before the new antigen clusters become widespread, the new antigen clusters can be discovered and tracked in time through the antigen relationship prediction model, and suitable candidate vaccine strains can be screened for immunization, thereby timely controlling the spread of the new antigen clusters and reducing the risk of pandemic. In an embodiment of the present invention, based on the antigen relationship prediction method of strain pairs, the vaccine strain recommendation strategy is improved by further real-time monitoring, recommendation by class, and recommendation for the use of new vaccine strains when new classes appear and reach a certain epidemic scale, to achieve more intelligent and effective vaccine strain recommendations, and to provide more reference opinions for the recommendation of vaccine strains of different new coronavirus epidemic classes, and to provide an effective solution to the problem of vaccine strain mismatch.

[0154] It should be noted that, in order to enable those skilled in the art to better distinguish data of the same type but with different actual meanings, some technical features are distinguished by the first, second, and third in the embodiments of the present invention. The first, second, and third are only used for data distinction and have no other special meanings. It can be understood that the present invention is not limited to this.

[0155] In an embodiment of the present invention, a method for constructing a novel coronavirus antigen prediction model and its application are provided. First, by simultaneously considering the distance differences of antigen epitopes, physicochemical properties, and receptor binding regions, a more reference-based multi-bit antigen descriptor is generated, so that when the model is trained using a data set for prediction, it can quickly capture more critical and effective feature information based on multi-bit antigen descriptors, thereby improving the prediction accuracy, timeliness, and effectiveness of model prediction. Secondly, by hierarchical sampling of the sequence information to be predicted, a sequence to be tested that can better reflect the overall sampling amount balance effect can be obtained, further making the prediction results obtained based on model prediction more accurate. Again, by adopting a prediction model constructed based on antigen-related features, the antigenic relationship between any two novel coronaviruses can be accurately and efficiently predicted. In practical applications, only two sequences need to be input, and the antigen similarity prediction results are quickly given by model prediction. In addition, since the entire prediction process is automatically implemented by a machine learning algorithm, there is no need to perform a large number of neutralization experiments and additional mapping, which can also greatly reduce the time and economic costs. Furthermore, based on a large amount of sequence data, an antigen similarity network is constructed, and new antigen clusters are quickly identified through clustering methods. Based on the antigen clusters, the constructed antigen similarity network is described, and the dynamics of antigen clusters are tracked in real time. On this basis, by improving the vaccine strain recommendation strategy, efficient and rapid vaccine strain recommendation can be achieved.

[0156] For better explanation, refer to Figure 7, showing a schematic diagram of the overall process of a method for rapid antigenicity identification of a novel coronavirus strain provided by an embodiment of the present invention. It should be noted that this embodiment only briefly describes the general process of rapid antigenicity identification of a novel coronavirus strain, and the specific implementation process of each step can be understood by referring to the relevant content in the aforementioned embodiment, which will not be described here. It can be understood that the present invention is not limited to this.

[0157] (1) Obtaining in vitro human sera that are initially infected with different lineages of the novel coronavirus, conducting neutralization experiments based on the in vitro human sera, and defining the antigenic similarity between strain pairs based on the results of the neutralization experiments, while obtaining multiple sets of strain pair sequences;

[0158] (2) Considering the differences in antigenic epitopes, physicochemical properties, and distances between receptor binding regions, a multi-bit antigenic descriptor corresponding to each group of strains is generated;

[0159] (3) based on each multi-bit antigen descriptor, multiple sets of virus pair sequences are converted into antigen description datasets, and antigen relationship prediction models are constructed and trained based on the antigen description datasets;

[0160] (4) Screening and obtaining a large amount of strain sequence information from the GISAID database, and hierarchically sampling the strain sequence information in a random selection manner according to time and location, to obtain the sequence information to be tested containing multiple S protein sequences of the target strain;

[0161] (5) using the trained antigen relationship prediction model to predict the antigen relationship of the sequence information to be tested, and constructing an initial antigen network based on the prediction results;

[0162] (6) performing cluster calculation on the prediction results based on Markov clustering to obtain antigen cluster information, and adding the antigen cluster information to the initial antigen network to describe the network as antigen clusters to obtain the target antigen network;

[0163] (7) Vaccine strain recommendations are made based on recent sequence antigen analysis and target antigen networks.

[0164] Reference Figure 8 , shows a structural block diagram of a device for rapid antigenicity identification of a novel coronavirus strain provided by an embodiment of the present invention, which may specifically include:

[0165] The data acquisition module 801 is used to obtain multiple sets of strain pair sequences of the novel coronavirus, and multiple antigen descriptors corresponding to each set of strain pair sequences;

[0166] A prediction model training module 802 is used to convert the multiple sets of strain pair sequences into antigen description data sets based on the multiple multi-bit antigen descriptors, and obtain an antigen relationship prediction model through model training according to the antigen description data sets;

[0167] The hierarchical sampling module 803 is used to obtain the strain sequence information and perform hierarchical sampling on the strain sequence information to obtain the sequence information to be tested;

[0168] The antigen similarity prediction module 804 is used to use the antigen relationship prediction model to perform antigen similarity prediction on the sequence information to be tested, and obtain the prediction results of the antigen similarity relationship between multiple strains in the sequence information to be tested.

[0169] In an optional embodiment, the data acquisition module 801 includes:

[0170] A neutralization experiment module is used to obtain in vitro human serum that is initially infected with different lineages of the novel coronavirus, and to conduct a neutralization experiment based on the in vitro human serum to obtain titer values ​​corresponding to multiple groups of original strains;

[0171] An antigenic relationship determination module, for determining the antigenic relationship of the original strain pair for each group of the original strain pair according to a preset similarity threshold and the titer value, wherein the antigenic relationship indicates that the original strain pair is antigenically similar, or antigenically dissimilar;

[0172] The strain pair data processing module is used to pre-process the multiple groups of original strain pairs to obtain multiple groups of strain pair sequences of the new coronavirus.

[0173] In an optional embodiment, the device further includes:

[0174] An initial antigenic site screening module is used to predict and screen the antigenic sites of the S protein of the strain sequence to obtain multiple initial antigenic sites;

[0175] An antigen site clustering calculation module, used to perform clustering calculation based on the multiple initial antigen sites to determine multiple antigen epitopes that meet preset site screening conditions;

[0176] A target antigen epitope determination module is used to perform secondary screening on the multiple antigen epitopes to obtain multiple target antigen epitopes and respectively generate a first antigen descriptor corresponding to each of the target antigen epitopes;

[0177] A physicochemical property item determination module is used to screen out several types of physicochemical properties from the physicochemical property database, determine the physicochemical property items corresponding to each type of the physicochemical property based on the strain pair sequence and combined with random forest classification prediction, and generate a second antigen descriptor corresponding to each of the physicochemical property items;

[0178] A distance difference feature determination module is used to perform distance difference analysis from the mutation site to the receptor binding region of the strain sequence, determine the distance difference feature, and generate a third antigen descriptor corresponding to the distance difference feature;

[0179] The multi-bit antigen descriptor determination module is used to use the third antigen descriptor, each of the second antigen descriptors and each of the first antigen descriptors as the multi-bit antigen descriptors corresponding to the strain pair sequence.

[0180] In an optional embodiment, the prediction model training module 802 includes:

[0181] An initial prediction model building module is used to divide the antigen description data set into a training set and a test set according to a preset ratio, and to build several different initial prediction models based on different types of classification model frameworks;

[0182] An antigen relationship prediction training module is used to perform antigen relationship prediction training on each of the initial prediction models using the training set, and to optimize parameters during the model training process;

[0183] The model verification module is used to use the test set to verify each trained initial prediction model, and use the trained initial prediction model with the best verification effect as the antigen relationship prediction model.

[0184] In an optional embodiment, the strain sequence information includes multiple S protein sequences of the strain containing location information and time information, and the hierarchical sampling module 803 includes:

[0185] The target strain S protein sequence sampling module is used to sample the strain sequence information for each region corresponding to the location information based on the time information, in the order of time nodes from far to near and corresponding sampling quantities from small to large, to obtain multiple target strain S protein sequences;

[0186] The multiple sequence alignment module is used to perform multiple sequence alignment on the S protein sequences of the multiple target strains, delete the sequences with sequencing errors, and obtain the sequence information to be tested.

[0187] In an optional embodiment, the device further includes:

[0188] An initial antigen network construction module is used to take each strain in the sequence information to be tested as a node, and take the antigenic similarity relationship between strains as node edges, and construct an initial antigen network of the new coronavirus based on each of the nodes and each of the node edges;

[0189] A Markov clustering module, used for performing multiple Markov clustering on the prediction results based on multiple pre-set expansion coefficients, adjusting parameters according to the calculated silhouette coefficients, and obtaining the best clustering result, wherein the clustering result includes antigen cluster information;

[0190] The target antigen network improvement module is used to add the antigen cluster information to the initial antigen network to divide the initial antigen network into multiple antigen clusters to obtain a target antigen network, which is used to reflect the prevalence and transformation of the multiple antigen clusters in various regions.

[0191] In an optional embodiment, the device further includes:

[0192] A recommended time point setting module is used to set a recommended time point every preset time period;

[0193] A historical strain sequence analysis module is used to analyze the antigenic classes of all historical strain sequences that appeared before the current recommended time point, and predict the antigenic similarity between all historical strain sequences based on the antigenic relationship prediction model;

[0194] The current antigen network generation module is used to perform Markov clustering calculations based on the antigen similarity prediction results, combine the antigen class analysis results, determine all the epidemic classes currently appearing in the new coronavirus, and generate the current antigen network;

[0195] A new strain determination module is used to obtain the historical antigen network generated corresponding to the last recommended time point, compare the current antigen network with the historical antigen network, and determine the new strains and corresponding strain types from the last recommended time point to the current recommended time point;

[0196] An intra-class degree calculation module, used to calculate the intra-class degree values ​​of each of the newly added strains and strains of the same type, respectively, wherein the intra-class degree value represents the number of strains of the same type in the current antigen network that have antigenicity similar to that of the newly added strain;

[0197] A first vaccine strain recommendation module is used to select a strain with the largest intraclass value as a recommended vaccine strain from the new antigen class when there is a new antigen class with a prevalence rate exceeding a preset prevalence threshold in the current antigen network;

[0198] The second vaccine strain recommendation module is used to select the strain with the largest intraclass value from the new antigen class with the highest prevalence rate as the recommended vaccine strain when there are two or more new antigen classes with prevalence rates exceeding a preset prevalence threshold in the current antigen network;

[0199] The third vaccine strain recommendation module is used to select the strain with the largest intra-class value as the recommended vaccine strain from the epidemic class with the highest prevalence rate when there is no new antigen class in the current antigen network, or there is no new antigen class whose prevalence rate exceeds a preset epidemic threshold.

[0200] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the aforementioned method embodiment.

[0201] An embodiment of the present invention further provides an electronic device, the device comprising a processor and a memory:

[0202] The memory is used to store the program code and transmit the program code to the processor;

[0203] The processor is used to execute the method for rapid antigenicity identification of the new coronavirus strain of any embodiment of the present invention according to the instructions in the program code.

[0204] An embodiment of the present invention also provides a computer-readable storage medium, which is used to store program code, and the program code is used to execute the method for rapid antigenicity identification of a new coronavirus strain according to any embodiment of the present invention.

[0205] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0206] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0207] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0208] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0209] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0210] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for rapid antigenicity identification of a novel coronavirus strain, characterized in that: include: Obtain multiple sets of strain pair sequences of the novel coronavirus, and multiple antigen descriptors corresponding to each set of strain pair sequences; Based on the multiple multi-bit antigen descriptors, the multiple groups of strain pair sequences are converted into antigen description data sets, and according to the antigen description data sets, an antigen relationship prediction model is obtained through model training; Obtaining strain sequence information, and performing hierarchical sampling on the strain sequence information to obtain sequence information to be tested; Using the antigen relationship prediction model to predict antigen similarity of the sequence information to be tested, and obtaining prediction results of antigen similarity relationships between multiple strains in the sequence information to be tested; The method of obtaining multiple sets of novel coronavirus strain pair sequences includes: Obtain in vitro human serum that is initially infected with different lineages of novel coronavirus, conduct neutralization experiments based on the in vitro human serum, and obtain titer values ​​corresponding to multiple groups of original strains; For each group of the original strain pairs, the antigenic relationship of the original strain pairs is determined according to a preset similarity threshold and the titer value, wherein the antigenic relationship indicates that the original strain pairs are antigenically similar or antigenically dissimilar; Preprocessing the multiple groups of original strain pairs to obtain multiple groups of strain pair sequences of the new coronavirus; The step of generating the multi-bit antigen descriptor comprises: Predicting and screening the antigenic sites of the S protein of the strain sequence to obtain multiple initial antigenic sites; Performing cluster calculation based on the multiple initial antigen sites to determine multiple antigen epitopes that meet the preset site screening conditions; Performing secondary screening on the multiple antigen epitopes to obtain multiple target antigen epitopes, and generating first antigen descriptors corresponding to each of the target antigen epitopes; Screening out several categories of physicochemical properties from the physicochemical property database, determining the physicochemical property items corresponding to each category of the physicochemical property based on the strain pair sequence and combined with random forest classification prediction, and generating second antigen descriptors corresponding to each physicochemical property item respectively; Performing distance difference analysis from the mutation site to the receptor binding region on the strain sequence, determining the distance difference feature, and generating a third antigen descriptor corresponding to the distance difference feature; The third antigen descriptor, each of the second antigen descriptors and each of the first antigen descriptors are collectively used as a multi-bit antigen descriptor corresponding to the strain pair sequence.

2. The method for rapid antigenicity identification of a novel coronavirus strain according to claim 1, characterized in that: The step of obtaining an antigen relationship prediction model through model training according to the antigen description dataset comprises: The antigen description data set is divided into a training set and a test set according to a preset ratio, and a number of different initial prediction models are constructed based on different types of classification model frameworks; For each of the initial prediction models, the training set is used to perform antigen relationship prediction training on the initial prediction model, and parameter optimization is performed during the model training process; The test set is used to verify each of the trained initial prediction models, and the trained initial prediction model with the best verification effect is used as the antigen relationship prediction model.

3. The method for rapid antigenicity identification of a novel coronavirus strain according to claim 1, characterized in that: The strain sequence information includes multiple S protein sequences of the strain containing location information and time information, and the hierarchical sampling of the strain sequence information to obtain the sequence information to be tested includes: For each region corresponding to the location information, based on the time information, the strain sequence information is sampled in the order of time nodes from far to near and corresponding sampling quantities from small to large, to obtain multiple S protein sequences of the target strain; A multiple sequence alignment is performed on the S protein sequences of the multiple target strains, sequences with sequencing errors are deleted, and sequence information to be tested is obtained.

4. The method for rapid antigenicity identification of a novel coronavirus strain according to any one of claims 1 to 3, characterized in that: Also includes: Each strain in the sequence information to be tested is taken as a node, and the antigenic similarity relationship between strains is taken as a node edge, and the initial antigen network of the new coronavirus is constructed based on each of the nodes and each of the node edges; Performing multiple Markov clustering on the prediction results based on multiple pre-set expansion coefficients, adjusting parameters according to the calculated silhouette coefficients, and obtaining the best clustering result, wherein the clustering result includes antigen cluster information; The antigen cluster information is added to the initial antigen network to divide the initial antigen network into multiple antigen clusters to obtain a target antigen network, which is used to reflect the prevalence and transformation of the multiple antigen clusters in various regions.

5. The method for rapid antigenicity identification of a novel coronavirus strain according to claim 4, characterized in that: Also includes: Set a recommended time point every preset time period; Analyze the antigenic classes of all historical virus strain sequences that appeared before the current recommended time point, and predict the antigenic similarity between all historical virus strain sequences based on the antigenic relationship prediction model; Based on the results of antigen similarity prediction, Markov clustering calculation is performed, combined with the results of antigen class analysis, to determine all the epidemic classes currently appearing in the new coronavirus and generate the current antigen network; Obtain the historical antigen network generated corresponding to the last recommended time point, compare the current antigen network with the historical antigen network, and determine the newly added strains and corresponding strain types from the last recommended time point to the current recommended time point; Calculating the intra-class degree value of each of the newly added strains and strains of the same type respectively, the intra-class degree value represents the number of strains of the same type in the current antigen network that are antigenically similar to the newly added strains; If there is a new antigen class in the current antigen network whose prevalence exceeds a preset prevalence threshold, then the strain with the largest intraclass value is selected from the new antigen class as the recommended vaccine strain; If there are two or more new antigen classes in the current antigen network whose prevalence exceeds the preset prevalence threshold, the strain with the largest intraclass value is selected from the new antigen class with the highest prevalence as the recommended vaccine strain; If there is no new antigen class in the current antigen network, or there is no new antigen class whose prevalence exceeds a preset epidemic threshold, the strain with the largest intra-class degree value is selected from the epidemic class with the highest prevalence as the recommended vaccine strain.

6. A device for rapid identification of antigenicity of a novel coronavirus strain, characterized in that: include: A data acquisition module, used to obtain multiple sets of strain pair sequences of the novel coronavirus, and multiple antigen descriptors corresponding to each set of strain pair sequences; A prediction model training module, used for converting the plurality of groups of strain pair sequences into antigen description data sets based on the plurality of multi-bit antigen descriptors, and obtaining an antigen relationship prediction model through model training according to the antigen description data sets; A hierarchical sampling module is used to obtain strain sequence information and perform hierarchical sampling on the strain sequence information to obtain sequence information to be tested; An antigen similarity prediction module is used to use the antigen relationship prediction model to perform antigen similarity prediction on the sequence information to be tested, and obtain the prediction results of the antigen similarity relationship between multiple strains in the sequence information to be tested; The data acquisition module comprises: A neutralization experiment module is used to obtain in vitro human serum that is initially infected with different lineages of the novel coronavirus, and to conduct a neutralization experiment based on the in vitro human serum to obtain titer values ​​corresponding to multiple groups of original strains; An antigenic relationship determination module, for determining the antigenic relationship of the original strain pair for each group of the original strain pair according to a preset similarity threshold and the titer value, wherein the antigenic relationship indicates that the original strain pair is antigenically similar, or antigenically dissimilar; A strain pair data processing module, used to pre-process the multiple groups of original strain pairs to obtain multiple groups of strain pair sequences of the new coronavirus; The device also includes: An initial antigenic site screening module is used to predict and screen the antigenic sites of the S protein of the strain sequence to obtain multiple initial antigenic sites; An antigen site clustering calculation module, used to perform clustering calculation based on the multiple initial antigen sites to determine multiple antigen epitopes that meet preset site screening conditions; A target antigen epitope determination module is used to perform secondary screening on the multiple antigen epitopes to obtain multiple target antigen epitopes and respectively generate a first antigen descriptor corresponding to each of the target antigen epitopes; A physicochemical property item determination module is used to screen out several types of physicochemical properties from the physicochemical property database, determine the physicochemical property items corresponding to each type of the physicochemical property based on the strain pair sequence and combined with random forest classification prediction, and generate a second antigen descriptor corresponding to each of the physicochemical property items; A distance difference feature determination module is used to perform distance difference analysis from the mutation site to the receptor binding region of the strain sequence, determine the distance difference feature, and generate a third antigen descriptor corresponding to the distance difference feature; The multi-bit antigen descriptor determination module is used to use the third antigen descriptor, each of the second antigen descriptors and each of the first antigen descriptors as the multi-bit antigen descriptors corresponding to the strain pair sequence.

7. An electronic device, characterized in that: The device comprises a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the method for rapid antigenicity identification of the new coronavirus strain according to any one of claims 1-5 according to the instructions in the program code.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program code, and the program code is used to execute the method for rapid antigenicity identification of the new coronavirus strain as described in any one of claims 1-5.

Images