Construction method and application of antibody two-dimensional code

Abstract

The invention belongs to the technical field of biology, and particularly relates to a construction method and application of an antibody two-dimensional code. The invention creatively develops a method for generating an antibody two-dimensional code (Ab-QRcode) only based on an antibody sequence without antigen information. The antibody two-dimensional code generated by the method can describe species specificity, variety and targeting specificity information of the antibody, and each antibody has a unique two-dimensional code which is unique and different from other antibodies. Based on the two-dimensional code, the target specificity of the antibody can be distinguished, and the antibody can be clustered according to the difference between the target specificity of the antibody, so that the clustering of the antibody for identifying the same or similar antigen can be realized.

Description

Technical Field

[0001] This invention belongs to the field of biotechnology, specifically relating to a method for constructing antibody QR codes and their applications. Background Technology

[0002] Natural antibodies are immunoglobulins produced by plasma cells and play a crucial role in the immune defense of vertebrates. They activate immune responses by recognizing and specifically binding to specific epitopes of antigens (BUSS NA, HENDERSON SJ, MCFARLANE M, et al. Monoclonal antibody therapeutics: History and future[J]. Curr Opin Pharmacol, 2012, 12(5):615-622.). Antibodies exhibit high specificity and affinity for their target antigens, and their targeting specificity is determined by the intermolecular interactions at the antibody-antigen interface. This highly specific type of interaction in antibody-antigen binding is usually different from general protein-protein interactions. The epitopes targeted by antibodies are crucial for understanding B cell-mediated immune responses and are essential information for developing therapeutic antibodies. Compared with the disadvantages of small molecule drugs, such as poor specificity, high toxicity, and short half-life, antibodies have the advantages of high specificity, high efficacy, high safety, and long half-life. This makes them show unparalleled advantages over other types of drugs in the diagnosis and treatment of various diseases (Li Wenzhen, Zhou Yuwei, Liu Wenwen, et al. Research progress on humanization evaluation and humanization calculation methods of therapeutic antibodies [J]. Journal of University of Electronic Science and Technology of China, 2024, 53(04):629-634.).

[0003] In the development of therapeutic antibodies, labor-intensive experimental methods can be used to obtain antibody target specificity information. However, with the advancement of antibody technology platforms, next-generation sequencing can easily obtain tens of thousands of antibody clones, which undoubtedly poses a significant challenge to existing experimental methods. Therefore, it is necessary to develop computational methods that are high-throughput, fast, and low-cost. Antibodies have a conserved structural framework composed of six "hypervariate loops" called complementarity-determining regions (CDRs). Generally, antibodies with similar targeting characteristics may interact with similar epitopes on the surface of antigens. However, this target specificity cannot be distinguished solely by the antibody CDR sequence. For example, both nexituzumab and cetuximab bind to the epidermal growth factor receptor (EGFR) and have very similar target epitopes, but the length and composition of their CDRs are completely different.

[0004] Computational antibody targeting specificity exploration has been a long-standing goal, but due to the vast diversity of antibodies and the complexity of antigen-antigen combinations, there are currently few computational tools available in this field. Experimental methods are costly, and the limited availability of computational methods makes them unsuitable for large volumes of antibody data. Therefore, it is essential to develop a computational tool that is applicable to large numbers of antibody samples, easy to operate, and cost-effective. Summary of the Invention

[0005] This invention innovatively develops a method for generating antibody QR codes (Ab-QR codes) based solely on antibody sequences without requiring antigen information. The antibody QR codes generated by this method can characterize the species specificity, type, and target specificity of the antibody, with each antibody possessing a unique QR code distinct from other antibodies. Based on these QR codes, the target specificity of antibodies can be distinguished, and antibodies can be clustered according to the differences in their target specificities, thereby enabling the clustering of antibodies that recognize the same or similar antigens.

[0006] In a first aspect, the present invention provides an antibody QR code (Ab-QRcode), which is generated by mapping the antibody primary sequence onto an ordered two-dimensional map, and can characterize the species specificity, type and target specificity information of the antibody.

[0007] Secondly, the present invention provides a method for constructing the antibody QR code described in the first aspect, the method specifically comprising the following steps:

[0008] S1. Obtain a non-redundant antibody sequence dataset for generating antibody QR code frameworks;

[0009] S2. Split the antibody sequences in the dataset obtained in step S1;

[0010] S3. Select an amino acid feature set to describe the split peptide set and generate a peptide-specific feature matrix;

[0011] S4. Calculate the correlation between every two columns in the peptide-specific feature matrix obtained in step S3, and generate a feature distance matrix;

[0012] S5. Project the feature distance matrix obtained in step S4 onto a two-dimensional space to obtain the antibody QR code framework;

[0013] S6. The antibody sequence to be generated into a QR code is split in the manner described in step S2. The peptide segments are described using the feature set selected in step S3 to generate a one-dimensional feature value matrix. Each feature value in the feature value matrix is ​​mapped to the frame obtained in step S6 according to the feature point position. The color saturation of the point is determined according to the feature value magnitude, thereby generating the antibody QR code.

[0014] Furthermore, in step S1, the antibody sequence can be collected from public antibody databases, publicly published literature, sequencing results from biological samples, or private databases developed by companies or research institutions.

[0015] Furthermore, the collected antibody sequences must contain complete or at most 10 missing amino acid variable region sequences.

[0016] Furthermore, the collected antibody sequences are complete sequences with paired light and heavy chains.

[0017] Furthermore, the collected antibody sequences contain at least 90 amino acids.

[0018] Furthermore, the collected antibody sequences number at least 50.

[0019] Preferably, the collected antibody sequences are at least 100 or more.

[0020] Furthermore, in step S1, the non-redundant antibody sequence dataset can be constructed in the following way: using the filtering formula Fun_Filterdata={seq i |seq i =True,if length(seq i )>90∩HL_paired(seq i )∩Similarity(seq i The collected sequences are processed to obtain the dataset Ant_data = {Ant < 100%} for pre-learning. i |i=1~N}; where N represents the total number of antibody sequences collected, Aant i This represents the protein sequence of the i-th antibody, length(seq) i )>90 indicates that the sequence length is greater than 90 amino acids, Similarity(seq i A value less than 100% indicates that the sequence similarity is less than 100%.

[0021] Further, in step S2, the method of splitting the antibody sequence is as follows: the frame region sequence in each antibody sequence is sequentially split into peptides of length n1 amino acids, the CDR1 and CDR2 sequences in each antibody sequence are sequentially split into peptides of length n2 amino acids, and the CDR3 sequence is sequentially split into peptides of length n3 amino acids, wherein n1∈[1,20] and is an integer, n2∈[1,10] and is an integer, and n3∈[1,5] and is an integer.

[0022] Furthermore, the split antibody sequence can be determined according to the formula Fun_SplitV={P|P=Σi＝i+M The antibody V region sequence is split into peptides of different lengths, where P represents the split peptide, N represents the peptide length, and n1, n2, n3 represent the number of amino acids contained in the peptide.

[0023] Furthermore, when site i is in the frame region, N takes the value n1, where n1 ∈ [1, 20] and is an integer; when site i is in CDR1 and CDR2, N takes the value n2, where n2 ∈ [1, 10] and is an integer; when site i is in CDR3, N takes the value n3, where n3 ∈ [1, 5] and is an integer, indicating that the sequence is split into peptides of N amino acids.

[0024] Furthermore, the frame region refers to a site in the antibody sequence that does not belong to CDR1, CDR2, or CDR3.

[0025] Furthermore, in step S3, the amino acid feature set is selected from one or more of the following: amino acid composition, protein secondary conformation, amino acid physicochemical properties, evolutionary conservation, microenvironment amino acid distribution and / or microenvironment amino acid composition.

[0026] Furthermore, the amino acid composition feature set is composed of 20 common amino acid types.

[0027] Furthermore, the amino acid composition feature set can be based on the formula Fun_com={Cscore} i |seq i =ai1ai2ai3…ai j ai j =Cscore i Cscore i The amino acid composition feature set AA_com = {Cscore} is defined and constructed using {20 amino acid types} and {j ∈ {variable region sequence sites}}. i Cscore i ∈{20 amino acid types}}, where ai j This indicates the amino acid composition of the sequence.

[0028] Furthermore, the protein secondary conformation feature set can be constructed using general protein secondary conformation calculation tools.

[0029] Furthermore, the protein secondary conformation calculation tool can be the GORIV tool.

[0030] Furthermore, the protein secondary conformation feature set can be calculated according to the formula Fun_sen={Sscore} i |seq i =ai1ai2ai3…aij ai j =GOR(Sscore) i Sscore i ∈{protein secondary conformation}, j∈{variable region sequence site}} Define the secondary conformation feature set AA_sec={Sscore i |Sscore i ∈{protein secondary conformation}}, where GOR(Sscore i () represents the prediction score using the protein secondary conformation prediction tool GOR4, ai j The score representing the secondary conformation prediction score of the amino acid at site j.

[0031] Furthermore, the aforementioned amino acid physicochemical property feature set can be constructed based on public or self-built amino acid physicochemical property databases.

[0032] Furthermore, the aforementioned public database of amino acid physicochemical properties could be AAindexDatabase.

[0033] Furthermore, the aforementioned amino acid physicochemical property feature set can be obtained through the formula Download_feature={Ofea i |i=1~N f Collect physicochemical characteristics from the physicochemical property database, where N f For the total number of physicochemical characteristics collected, Ofea i Representing the i-th physicochemical feature, and then through Fun_Filterfeature={aafea i |aafea i =True,if NotNA(aafea i )∩(Similarity(aafea i The collected physicochemical features are screened to obtain a set of physicochemical features describing the peptide composition: AA_index={Iscore}. i Iscore i ∈{physicochemical properties}}, where aafea i Each feature is represented by aafea i =True indicates that the feature will be preserved, NotNA(aafea) i ) indicates that the feature values ​​are not missing, Similarity(aafea i A Pearson correlation coefficient of less than 0.8 indicates that the similarity between features is less than 0.8.

[0034] Furthermore, the evolutionary conservation feature set can be constructed by calculating the conservation score of each amino acid in the antibody variable region sequence using a general site-specific sequence alignment tool.

[0035] Furthermore, the site-specific sequence alignment tool is PSI-BLAST.

[0036] Furthermore, the aforementioned evolutionary conservation feature set can be calculated using the formula Fun_psi={Pscore} i |seq i =ai1ai2ai3…ai j ai j =PSIblast(Pscore) i Pscore i Define and construct the evolutionary conservation feature set AA_psi={Pscore}, where j∈{V region sequence sites}}. i Pscore i ∈{conservatism}}; where, PSIblast(Pscore i The score is calculated by the site-specific alignment tool.

[0037] Furthermore, the aforementioned microenvironment amino acid distribution feature set can be obtained by describing the distribution characteristics of each amino acid and its microenvironment through a three-dimensional spatial structure.

[0038] Furthermore, the microenvironment amino acid distribution characteristics can be described as follows: Draw a circle with each amino acid as the center point and a radius of 10 angstroms. Divide the circle into spherical shells every two angstroms, resulting in a total of 5 spherical shells. Add up the number of amino acids within each spherical shell to describe the amino acid distribution in the microenvironment surrounding each amino acid, forming an amino acid distribution characteristic set AA_dist={Dscore i |Dscore i ∈{number of amino acids}}.

[0039] Furthermore, the aforementioned microenvironment amino acid composition feature set can be obtained by revealing the combination features of amino acids and their neighboring amino acids through three-dimensional spatial structure.

[0040] Furthermore, the microenvironment amino acid composition characteristics can be described as follows: Taking each amino acid as the center, traverse all amino acids within a 4 angstrom range, count the amino acid triangle types, assign a unique integer to each triangle type, and finally generate an amino acid triangle composition feature set AA_tria={Tscore i |Tscore i ∈{amino acid triangle type}}.

[0041] Furthermore, in step S3, the peptide-specific feature matrix is ​​determined according to the formula... Determined, where m represents the total number of antibody sequences in the aforementioned non-redundant antibody sequence dataset, n represents the number of peptides split into for each antibody, p represents the number of types of the selected feature set, and the feature value X ij (1≤i≤m, 1≤j≤n) by the formula FeaScore (a) is calculated by summing the characteristic values ​​of amino acids in each peptide segment. i () represents the predicted or calculated score in the feature set corresponding to each amino acid. This indicates that the score is accumulated, and Np is the number of amino acids contained in the peptide segment.

[0042] Further, in step S4, the correlation is calculated according to the formula Fun_corr=1-xy / ||x||||y||, respectively, to calculate the distance Y between every two columns in Pep_Mat (the matrix has n columns in total, each column representing one peptide segment, for a total of n peptide segments). ij (1≤i≤q,1≤j≤q), ultimately generating a feature distance matrix representing peptide correlation. Where x and y represent the feature values ​​in Pep_Mat, q = n × p, n represents that each antibody is split into n peptides, and p represents the number of feature sets used.

[0043] Furthermore, in step S5, the antibody QR code framework is based on the intrinsic correlation between associated antibody sequences. The feature distance matrix Fea_mat is projected into a two-dimensional feature space using high-dimensional data visualization and dimensionality reduction tools, and then the formula Fun_map=||x embed -y grid || 2 Feature points are generated by linearly assigning them to a two-dimensional grid, and then coloring the point according to the feature set category or subset category to which it belongs, where x embed Let y represent the unprocessed feature distance matrix. grid Let ||x| represent the feature distance matrix after linear assignment. embed -y grid || 2 This indicates the minimum loss value.

[0044] Furthermore, the high-dimensional data visualization and dimensionality reduction tools include, but are not limited to, UMAP, tSNE, and hypertools.

[0045] Furthermore, the linear allocation algorithm includes, but is not limited to, Jonker–Volgenant and kuhn-munkres.

[0046] Furthermore, the coloring can be applied to the corresponding feature points using 4-20 different colors, depending on the different classification methods of the feature set.

[0047] Furthermore, the coloring can assign four different colors to the corresponding feature points based on a set of features such as amino acid composition, protein secondary conformation, amino acid physicochemical properties, and evolutionary conservation.

[0048] Furthermore, the coloring can assign six different colors to corresponding feature points based on a set of features such as amino acid composition, protein secondary conformation, amino acid physicochemical properties, evolutionary conservation, microenvironment amino acid distribution, and microenvironment amino acid composition.

[0049] Furthermore, the coloring can be based on the amino acid composition, protein secondary conformation, evolutionary conservation, and nine subsets of the amino acid physicochemical property feature set (amino acid composition subset, hydrophobicity subset 1, hydrophobicity subset 2, α and rotation tendency subset 1, α and rotation tendency subset 2, β tendency subset, physicochemical property subset, binding feature subset, and other property subsets), assigning 12 different colors to the corresponding feature points.

[0050] Furthermore, in step S6, the one-dimensional eigenvalue matrix can be represented by single_pep = (fea1, ..., fea) n ) n It means that fea n The value represents the characteristic value of the peptide, and n indicates that the antibody can be separated into n peptides.

[0051] Thirdly, the present invention provides a system for constructing antibody QR codes, the system comprising a data input module, a data processing module, and a data output module:

[0052] The data input module is for inputting antibody sequence data;

[0053] The data processing module includes: obtaining a non-redundant antibody sequence dataset, splitting antibody sequences, selecting a feature set, constructing a peptide-specific feature matrix, generating a feature distance matrix, obtaining an antibody QR code framework, and generating an antibody QR code.

[0054] The data output module displays the antibody QR code.

[0055] Furthermore, the input antibody sequences can be collected from public antibody databases, publicly published literature, sequencing results from biological samples, or private databases developed by companies or research institutions.

[0056] Furthermore, the collected antibody sequences are complete sequences with paired light and heavy chains.

[0057] Furthermore, the collected antibody sequences contain at least 90 amino acids.

[0058] Furthermore, the collected antibody sequences number at least 50.

[0059] Preferably, the collected antibody sequences are at least 100 or more.

[0060] Furthermore, the non-redundant antibody sequence dataset can be constructed in the following way: using the filtering formula Fun_Filterdata={seq i |seq i =True,if length(seq i )>90∩HL_paired(seq i )∩Similarity(seq i The collected sequences are processed to obtain the dataset Ant_data = {Ant < 100%} for pre-learning. i |i=1~N}; where N represents the total number of antibody sequences collected, Aant i This represents the protein sequence of the i-th antibody, length(seq) i )>90 indicates that the sequence length is greater than 90 amino acids, Similarity(seq i A value less than 100% indicates that the sequence similarity is less than 100%.

[0061] Furthermore, the method for splitting the antibody sequence is as follows: the frame region sequence in each antibody sequence is sequentially split into peptides of length n1 amino acids, the CDR1 and CDR2 sequences in each antibody sequence are sequentially split into peptides of length n2 amino acids, and the CDR3 sequence is sequentially split into peptides of length n3 amino acids, wherein n1∈[1,20] and is an integer, n2∈[1,10] and is an integer, and n3∈[1,5] and is an integer.

[0062] Furthermore, the split antibody sequence can be determined according to the formula Fun_SplitV={P|P=∑ i＝i+M The antibody V region sequence is split into peptides of different lengths, where P represents the split peptide, N represents the peptide length, and n1, n2, n3 represent the number of amino acids contained in the peptide.

[0063] Furthermore, when site i is in the frame region, N takes the value n1, where n1 ∈ [1, 20] and is an integer; when site i is in CDR1 and CDR2, N takes the value n2, where n2 ∈ [1, 10] and is an integer; when site i is in CDR3, N takes the value n3, where n3 ∈ [1, 5] and is an integer, indicating that the sequence is split into peptides of N amino acids.

[0064] Furthermore, the frame region refers to a site in the antibody sequence that does not belong to CDR1, CDR2, or CDR3.

[0065] Furthermore, the feature set includes: amino acid composition, protein secondary conformation, amino acid physicochemical properties, evolutionary conservation, microenvironment amino acid distribution and / or microenvironment amino acid composition.

[0066] Furthermore, the amino acid composition feature set is composed of 20 common amino acid types.

[0067] Furthermore, the amino acid composition feature set can be determined according to the formula Fun_com={Cscore} i |seq i =ai1ai2ai3…ai j ai j =Cscore i Cscore i The amino acid composition feature set AA_com = {Cscore} is defined and constructed using {20 amino acid types} and {j ∈ {variable region sequence sites}}. i Cscore i ∈{20 amino acid types}}, where ai j This indicates the amino acid composition of the sequence.

[0068] Furthermore, the protein secondary conformation feature set can be constructed using general protein secondary conformation calculation tools.

[0069] Furthermore, the protein secondary conformation calculation tool can be the GOR IV tool.

[0070] Furthermore, the protein secondary conformation feature set can be calculated according to the formula Fun_sen={Sscore} i |seq i =ai1ai2ai3…ai j ai j =GOR(Sscore) i Sscore i ∈{protein secondary conformation}, j∈{variable region sequence site}} Define the secondary conformation feature set AA_sec={Sscore i |Sscore i ∈{protein secondary conformation}}, where GOR(Sscore i () represents the prediction score using the protein secondary conformation prediction tool GOR4, ai j The score representing the secondary conformation prediction score of the amino acid at site j.

[0071] Furthermore, the aforementioned amino acid physicochemical property feature set can be constructed based on public or self-built amino acid physicochemical property databases.

[0072] Furthermore, the aforementioned public database of amino acid physicochemical properties could be the AAindex Database.

[0073] Furthermore, the aforementioned amino acid physicochemical property feature set can be obtained through the formula Download_feature={Ofea i |i=1~N f Collect physicochemical characteristics from the physicochemical property database, where N f For the total number of physicochemical characteristics collected, Ofea i Representing the i-th physicochemical feature, and then through Fun_Filterfeature={aafea i |aafea i =True,if NotNA(aafea i )∩(Similarity(aafea i The collected physicochemical features are screened to obtain a set of physicochemical features describing the peptide composition: AA_index={Iscore}. i Iscore i ∈{physicochemical properties}}, where aafea i Each feature is represented by aafea i =True indicates that the feature will be preserved, NotNA(aafea) i ) indicates that the feature values ​​are not missing, Similarity(aafea i A Pearson correlation coefficient of less than 0.8 indicates that the similarity between features is less than 0.8.

[0074] Furthermore, the evolutionary conservation feature set can be constructed by calculating the conservation score of each amino acid in the antibody variable region sequence using a general site-specific sequence alignment tool.

[0075] Furthermore, the site-specific sequence alignment tool is PSI-BLAST.

[0076] Furthermore, the aforementioned evolutionary conservation feature set can be calculated using the formula Fun_psi={Pscore} i |seq i =ai1ai2ai3…ai j ai j =PSIblast(Pscore) i Pscore iDefine and construct the evolutionary conservation feature set AA_psi={Pscore}, where j∈{V region sequence sites}}. i Pscore i ∈{conservatism}}; where, PSIblast(Pscore i The score is calculated by the site-specific alignment tool.

[0077] Furthermore, the aforementioned microenvironment amino acid distribution feature set can be obtained by describing the distribution characteristics of each amino acid and its microenvironment through a three-dimensional spatial structure.

[0078] Furthermore, the microenvironment amino acid distribution characteristics can be described as follows: Draw a circle with each amino acid as the center point and a radius of 10 angstroms. Divide the circle into spherical shells every two angstroms, resulting in a total of 5 spherical shells. Add up the number of amino acids within each spherical shell to describe the amino acid distribution in the microenvironment surrounding each amino acid, forming an amino acid distribution characteristic set AA_dist={Dscore i |Dscore i ∈{number of amino acids}}.

[0079] Furthermore, the aforementioned microenvironment amino acid composition feature set can be obtained by revealing the combination features of amino acids and their neighboring amino acids through three-dimensional spatial structure.

[0080] Furthermore, the microenvironment amino acid composition characteristics can be described as follows: Taking each amino acid as the center, traverse all amino acids within a 4 angstrom range, count the amino acid triangle types, assign a unique integer to each triangle type, and finally generate an amino acid triangle composition feature set AA_tria={Tscore i |Tscore i ∈{amino acid triangle type}}.

[0081] Furthermore, the peptide-specific feature matrix is ​​based on the formula... Determined, where m represents the total number of antibody sequences in the aforementioned non-redundant antibody sequence dataset, n represents the number of peptides split into for each antibody, p represents the number of types of the selected feature set, and the feature value X ij (1≤i≤m, 1≤j≤n) by the formula FeaScore (a) is calculated by summing the characteristic values ​​of amino acids in each peptide segment. i () represents the predicted or calculated score in the feature set corresponding to each amino acid. This indicates that the score is accumulated, and Np is the number of amino acids contained in the peptide segment.

[0082] Furthermore, the feature distance matrix is ​​generated by calculating the correlation between every two columns in the peptide-specific feature matrix.

[0083] Furthermore, the correlation is calculated using the formula Fun_corr = 1 - xy / ||x||||y||, respectively, to determine the distance Y between every two columns in Pep_Mat (this matrix has n columns, each representing one peptide segment, for a total of n peptide segments). ij (1≤i≤q,1≤j≤q), ultimately generating a feature distance matrix representing peptide correlation. Where x and y represent the feature values ​​in Pep_Mat, q = n × p, n represents that each antibody is split into n peptides, and p represents the number of feature sets used.

[0084] Furthermore, the antibody QR code framework is based on the intrinsic correlation between associated antibody sequences. It projects the feature distance matrix Fea_mat into a two-dimensional feature space using high-dimensional data visualization and dimensionality reduction tools, and then uses the formula Fun_map=||x embed -y grid || 2 Feature points are generated by linearly assigning them to a two-dimensional grid, and then coloring the point according to the feature set category or subset category to which it belongs, where x embed Let y represent the unprocessed feature distance matrix. grid Let ||x| represent the feature distance matrix after linear assignment. embed -y grid || 2 This indicates the minimum loss value.

[0085] Furthermore, the high-dimensional data visualization and dimensionality reduction tools include, but are not limited to, UMAP, tSNE, and hypertools.

[0086] Furthermore, the linear allocation algorithm includes, but is not limited to, Jonker–Volgenant and kuhn-munkres.

[0087] Furthermore, the coloring can be applied to the corresponding feature points using 4-20 different colors, depending on the different classification methods of the feature set.

[0088] Furthermore, the coloring can assign four different colors to the corresponding feature points based on a set of features such as amino acid composition, protein secondary conformation, amino acid physicochemical properties, and evolutionary conservation.

[0089] Furthermore, the coloring can assign six different colors to corresponding feature points based on a set of features such as amino acid composition, protein secondary conformation, amino acid physicochemical properties, evolutionary conservation, microenvironment amino acid distribution, and microenvironment amino acid composition.

[0090] Furthermore, the coloring can be based on the amino acid composition, protein secondary conformation, evolutionary conservation, and nine subsets of the amino acid physicochemical property feature set (amino acid composition subset, hydrophobicity subset 1, hydrophobicity subset 2, α and rotation tendency subset 1, α and rotation tendency subset 2, β tendency subset, physicochemical property subset, binding feature subset, and other property subsets), assigning 12 different colors to the corresponding feature points.

[0091] Furthermore, the antibody QR code is generated by splitting the antibody sequence to be generated into a QR code, selecting a feature set to describe the peptide segments, generating a one-dimensional feature value matrix, mapping each feature value in the feature value matrix to the QR code frame according to the feature point position, and determining the color saturation of the point according to the feature value magnitude.

[0092] Furthermore, the one-dimensional eigenvalue matrix can be represented as single_pep = (fea1, ..., fea) n ) n It means that fea n The value represents the characteristic value of the peptide, and n indicates that the antibody can be separated into n peptides.

[0093] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps described in the second or third aspect of the present invention.

[0094] Fifthly, the present invention provides the use of the antibody QR code described in the first aspect in antibody-targeted recognition specific cluster analysis, including one or more of the following application directions:

[0095] A) Differentiate antibodies produced by the human body under physiological and / or various pathological conditions;

[0096] B) Antibodies that distinguish between different antigens or epitopes.

[0097] Furthermore, the antibody QR code undergoes cluster analysis through the following steps:

[0098] 1) Image recognition;

[0099] 2) Similarity calculation: Calculate the distance between features using algorithms;

[0100] 3) Dimensionality Reduction Clustering: High-dimensional data visualization and dimensionality reduction tools are used to reduce the dimensionality of the feature distance matrix and visualize the antibody clusters.

[0101] Furthermore, in step 1), the image recognition refers to using an image deep learning model to extract the antibody-targeting specific features contained in the two-dimensional map and generate a feature representation matrix.

[0102] Furthermore, the two-dimensional map refers to the antibody QR code generated according to the method described in the second aspect of the present invention.

[0103] Furthermore, the feature representation matrix is ​​as follows: Among them, X ij The feature values ​​represent the target specificity, i and j represent the dimensions of the extracted matrix, i∈[1-7], j∈[1-2048], and i and j are both integers.

[0104] Furthermore, the image deep learning models include, but are not limited to, ResNet50, Sobel, Canny, Hough, SIFT, and SURF.

[0105] Further, in step 2), the similarity calculation is performed by calculating the distance between each pair of feature representation matrices according to the formula Fun_Feacorr=1-xy / ||x||||y||, ultimately generating a v×v dimensional distance matrix representing the differences between antibody-specific two-dimensional maps. Where x and y represent the eigenvalues ​​in Sepc_Fea, respectively, X vv The distance between any two two-dimensional graphs is represented by v, where v represents the total number of two-dimensional graphs.

[0106] Furthermore, in step 2), the algorithm includes, but is not limited to, the Peirce correlation algorithm, the Euclidean distance algorithm, the Hamming distance algorithm, and the cosine similarity algorithm.

[0107] Furthermore, in step 3), the high-dimensional data visualization and dimensionality reduction tools include, but are not limited to, UMAP, tSNE, and hypertools.

[0108] Furthermore, the application of antibody QR codes in distinguishing antibodies produced by the human body under physiological and / or various pathological conditions refers to the ability to differentiate antibodies from healthy organisms or patients with specific diseases by using cluster analysis after QR codes of antibody sequences from different individuals.

[0109] Furthermore, the application of antibody QR codes in distinguishing antibodies against different antigens or epitopes refers to the ability to differentiate antibodies targeting different antigens or antibodies targeting the same antigen but with different binding epitopes through cluster analysis, using the antibody targeting specificity information contained in the antibody QR code.

[0110] Beneficial effects

[0111] 1. The antibody QR code generation technology of the present invention can effectively distinguish antibody sequences with different targeting properties.

[0112] 2. Ab-QRcode can distinguish antibodies with completely different target specificities, accurately dividing antibodies from human and mouse sources into two distinct clusters, and has the ability to identify subtle differences among antibodies of diverse origins; Ab-QRcode can also effectively distinguish antibodies induced by the same viral antigen.

[0113] 3. Ab-QRcode is significantly better than Ab-Ligity and BLAST methods in distinguishing antibodies with different target specificities; compared with Ab-Ligity, the prediction performance has increased by up to 159.46%, and compared with BLAST, the prediction performance has increased by up to 296.69%.

[0114] 4. Ab-QRcode can distinguish between antibodies induced by the same virus but targeting different viral antigens or even different structural domains, especially showing strong distinguishing ability for these two types of specific antibodies targeting the RBD and S2 domains.

[0115] 5. Ab-QRcode has excellent recognition ability for antibodies with large sequence differences but completely consistent target specificity.

[0116] 6. Ab-QRcode can identify antibodies that target the same or adjacent epitopes but have completely different CDRH3 sequences. Attached Figure Description

[0117] Figure 1 This is a QR code displaying SARS-CoV-2 antibodies.

[0118] Figure 2 This image displays some QR codes for SARS-CoV-2 and HIV antibodies. The first row shows HIV antibodies, and the second row shows SARS-CoV-2 antibodies (from left to right, the first row contains QR codes for HIV antibodies numbered 1U8L, 6OT1, and 5U3N; the second row contains QR codes for SARS-CoV-2 antibodies numbered XGv-289, OC220302-SN0739, and BD45-70).

[0119] Figure 3 This is an antibody clustering diagram. In the diagram: a, clustering of SARS-CoV-2 and HIV-1 specific antibodies based on Ab-QRcode generation technology; b, clustering of SARS-CoV-2 and HIV-1 specific antibodies based on sequence similarity (BLAST). Figure 4The figures show the performance of Ab-Qrcode, Ab-Ligity, and BLAST on different species datasets. In the figures: ac, based on Ab-Qrcode, dimensionality reduction clusters of specific antibodies targeting SARS-CoV-2 spike protein, lysosomes, H1N1 HA, and H3N2 HA proteins using UMAP; df, based on Ab-Ligity, dimensionality reduction clusters of specific antibodies targeting SARS-CoV-2 spike protein, lysosomes, H1N1 HA, and H3N2 HA proteins using UMAP; gi, based on Ab-Ligity, dimensionality reduction clusters of specific antibodies targeting SARS-CoV-2 spike protein, lysosomes, H1N1 HA, and H3N2 HA proteins using UMAP; j, cluster scores for each dataset based on Ab-Qrcode, Ab-Ligity, and BLAST; k, percentage increase between Ab-Qrcode and Ab-Ligity, and between Ab-Qrcode and BLAST.

[0120] Figure 5 The figures show the performance of Ab-QRcode, Ab-Ligity, and BLAST on datasets of different antigen proteins / domains within the same species. In the figures: ab, dimensionality reduction clustering of specific antibodies targeting H1N1 HA and NA proteins and SARS-CoV-2 spike protein RBD and S2 domains using UMAP based on Ab-QRcode; c, clustering scores for each dataset based on Ab-QRcode, Ab-Ligity, and BLAST; de, dimensionality reduction clustering of specific antibodies targeting H1N1 HA and NA proteins and SARS-CoV-2 spike protein RBD and S2 domains using UMAP based on Ab-Ligity; fg, dimensionality reduction clustering of specific antibodies targeting H1N1 HA and NA proteins and SARS-CoV-2 spike protein RBD and S2 domains using UMAP based on BLAST; h, percentage increase between Ab-QRcode and Ab-Ligity, and between Ab-QRcode and BLAST.

[0121] Figure 6The performance of Ab-QRcode on different epitope datasets is shown. In the figure: a, mapping the epitopes targeted by antibodies H2897 and CH65 to the HA protein structure (A / Solomon Islands / 3 / 2006, PDB id:5UGY, chain:A; PDB id:5UG0, chain:A); b, V region sequence similarity between antibodies H2897 and CH65; c, dimensionality reduction clustering of specific antibodies targeting the SARS-CoV-2 spike protein using UMAP based on Ab-QRcode; d, distance heatmap after dimensionality reduction clustering of five specific antibodies (PW5-4, PW5-5, PW5-534, PW5-535, PW5-570) targeting the SARS-CoV-2 spike protein using UMAP based on Ab-QRcode; e, heatmap of ELISA competitive binding assays among the five antibodies; f, CDR sequence similarity heatmap among the five antibodies calculated based on BLAST.

[0122] Figure 7 This image displays QR codes for antibodies from healthy donors and systemic lupus erythematosus (SLE) patients. The first row shows antibodies from healthy donors, and the second row shows antibodies from SLE donors.

[0123] Figure 8 To demonstrate the clustering of antibodies from systemic lupus erythematosus (SLE) donors and healthy donors (HD) using antibody QR codes, it was found that antibodies in pathological and physiological states could be completely separated, exhibiting a very high degree of discrimination. Detailed Implementation

[0124] The specific embodiments of the present invention will be further described below. It should be noted that the present invention is not limited to the specific details of the following embodiments. Within the scope of the technical concept of the present invention, various modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0125] It should also be noted that the various specific technical features and steps described in the following specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

[0126] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

[0127] Terminology Explanation

[0128] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0129] Antibody variable region: Antibodies exist as one or more Y-shaped monomers. Each Y-shaped monomer consists of four polypeptide chains, including two identical heavy chains and two identical light chains. The top regions of the heavy and light chains are called variable regions. Variable regions are the main regions where antibodies bind to antigens and determine the antibody's specific targeting recognition of antigens.

[0130] Complementarity-determining regions (CDRs): Within the variable domains of both the heavy and light chains of immunoglobulins, there are three specific segments whose amino acid composition and sequence are highly variable, called hypervariable regions (HVRs), denoted as HVR1, HVR2, and HVR3, respectively. These three hypervariable regions within the VH and VL chains together constitute the antigen-binding site of Ig. This site can bind complementary to the corresponding antigenic epitopes; therefore, HVRs are also called complementarity-determining regions, denoted as CDR1, CDR2, and CDR3, respectively.

[0131] CDR Numbering: Different antibodies have different amino acid compositions and sequences in their CDRs, which determine the specificity of the antibody's binding to the corresponding antigenic epitope. The CDR is responsible for recognizing and binding to the antigen, thereby exerting an immune effect. Well-known and commonly used CDR numbering and definition systems include the Kabat, IGMT, Cothia, and Martin schemes, which are reviewed in the article "Understanding the Significance and Implications of Antibody Numbering and Antigen-Binding Surface / Residue Definition," Front Immunol. 2018; 9:2278. Many online calculation tools also exist that can select different schemes to label the CDR of antibody sequences, such as https: / / www.novopro.cn / tools / cdr.html.

[0132] Framework Region (FR): The remaining amino acid residues in the antibody variable region, excluding the complementarity-determining region, constitute the framework region. The framework region is relatively stable and conserved, with few changes, and occupies about 75% of the variable region. It is mainly used to support the complementarity-determining region and also to maintain the three-dimensional conformation of the variable region.

[0133] Unsupervised learning is a machine learning method that uses machine learning algorithms to analyze unlabeled datasets and perform clustering or dimensionality reduction. Unsupervised learning can discover the inherent structure and relationships in data without human intervention.

[0134] Clustering is a data mining technique that groups unlabeled data based on similarity or difference. It's used to process raw, unclassified data objects into distinct groups, representing these groups through the structure or pattern of information. Clustering algorithms can be categorized into several types, including exclusive, overlapping, hierarchical, and probabilistic.

[0135] Dimensionality reduction is an unsupervised learning method that projects high-dimensional data into a low-dimensional space while preserving the main features of the data. Common dimensionality reduction algorithms include Principal Component Analysis (PCA) and t-SNE.

[0136] Antibody QR code framework: After the antibody sequence is split into peptides, the peptide features are described using amino acid feature sets, and their correlations are calculated. Guided by these correlations, the feature values ​​describing the peptide features are linearly assigned to a two-dimensional plane. Each feature value corresponds to a pixel on the two-dimensional plane, called a feature point. All feature points form the QR code framework. The QR code framework reflects the relative position of the feature values ​​corresponding to each feature set on the plane, and different colors are assigned to feature points from different feature set types for differentiation.

[0137] Example 1: Generation of Antibody-Targeted Specific QR Codes (Ab-QRcode)

[0138] 1. Collect antibody sequence set Download_data from the antibody public database, which contains 12429 antibody sequences. Filter the sequence using the formula Fun_Filterdata = {seq i |seq i =True,if length(seq i )>90∩HL_paired(seq i )∩Similarity(seq i If )<100%}, obtain the Ant_data dataset for pre-learning, which contains a total of 3219 sequences.

[0139] 2. Split the antibody variable region (V) sequence into peptides of different lengths: according to the formula Fun_SplitV={P|P=∑ i＝i+NThe antibody V region sequence is split into peptides of different lengths using the formula i+N, where i ∈ {V region sequence sites} and N ∈ {n1, n2, n3}. When site i is in the frame region, N is n1, n1 = 3; when site i is in CDR1 and CDR2, N is n2, n2 = 2; when site i is in CDR3, N is n3, n3 = 1 (frame region sites: 1-26, 39-55, 66-104, 118-128; CDR1 site: 27-38; CDR2 site: 56-65; CDR3 site: 105-117). A total of 110 peptides are obtained, each containing 1-3 amino acids.

[0140] 3. According to the formula Fun_com={Cscore i |seq i =ai1ai2ai3…ai j ai j =Cscore i Cscore i ∈{20 amino acid types}, j∈{V region sequence sites}} defines the amino acid composition feature set AA_com={Cscore i Cscore i =1,2,…20}; filtered by the formula Fun_Filterfeature={aafea i |aafea i =True,if NotNA(aafea i )∩Similarity(aafea i If the value is less than 0.8, obtain the physicochemical characteristic set AA_index={Iscore} describing the peptide composition. i Iscore i ∈{0.4938,0.2716,…0.5309}}; according to the formula Fun_sen={Sscore i |seq i =ai1ai2ai3…ai j ai j =GOR(Sscore) i Sscore i ∈{protein secondary conformation}, j∈{V region sequence site}} Define the secondary structure feature set AA_sec={Sscore i |Sscore i ∈{111,222,333}} (Sscore when the second-order conformation is an α-helix) i =111, which is the Sscore when β-folded. i =222, which is the Sscore for random curl. i=333); According to the formula Fun_psi={Pscore i |seq i =ai1ai2ai3…ai j ai j =PSIblast(Pscore) i ), Pscore i ∈{conservatism}, j∈{V region sequence sites}} Define the conservatism feature set AA_psi={Pscore i Pscore i ∈{0-1}}(Pscore i When P = 1, the site exhibits the strongest conservation. i (The site is least conserved when the value is 0).

[0141] 4. Use the feature set selected in step 3 to describe the peptide segments split in step 2, and construct a peptide-specific feature set: according to the formula Generate a 3219×110 dimensional peptide-specific matrix (There are a total of 3219 antibody sequences, and each antibody is divided into 110 peptide segments).

[0142] 5. Capture the intrinsic correlation of antibody sequence features and calculate the pairwise correlations between peptide-specific matrices Pep_Mat: Calculate the distance between each pair of columns in Pep_Mat using the formula Fun_corr = 1 - xy / ||x||||y||, ultimately generating a 110×110 dimensional feature distance matrix representing peptide correlations. (Each antibody is divided into 110 peptide segments). Using high-dimensional data visualization and dimensionality reduction tools such as UMAP, the feature distance matrix Fea_mat is projected into a two-dimensional feature space, and then linear allocation is performed using the Jonker-Volgenant linear allocation algorithm. Finally, by associating the intrinsic relationships between antibody sequences, a fixed position is found for each feature point. The amino acid physicochemical property feature set is further subdivided into 9 subsets: amino acid component subset, hydrophobicity subset 1, hydrophobicity subset 2, α and rotational tendency subset 1, α and rotational tendency subset 2, β tendency subset, physicochemical property subset, binding feature subset, and other property subset. Each subset is assigned a different color, along with the other three feature sets, assigning different colors to each type of feature point, for a total of 12 colors.

[0143] 6. For specific SARS-CoV-2 antibodies (heavy chain sequence: QVQLVQSGAEVKKPGASVKVSCKASGYTFTSYAMHWVRQAPGQRLEWMGWINAGNG NTRYSQKFQGRVTITRDTSASTAYMELSSLRSEDTAVYYCASRREQWLGDLGYYYYGM DVWGQGTTVTVSS; light chain sequence: DVVMTQSPLSLPVTLGQPASISCRSSQSLVYSDGNTYLNWFQQRPGQSPRRLIYKVSNRD SGVPDRFSGSGSGTDFTLKISRVEAEDVGVYYCMQGTHSPWTFGQGTKVEIK), the variable region (V) sequence of the antibody is split into peptides of different lengths according to step 2. Each peptide contains 1-3 amino acids (the peptides in the framework region contain 3 amino acids, the peptides in CDR1 and 2 contain 2 amino acids, and the peptides in CDR3 contain 1 amino acid). The target specificity of peptides is described using the feature set selected in step 3, generating a peptide specificity matrix single_pep = (1.45, ..., 11.57). 1×110 (Each antibody is divided into 110 peptide segments). Then, the feature values ​​in `single_pep` are mapped to the QR code frame according to the arrangement pattern in step 5, generating a unique QR code for each antibody, such as... Figure 1 As shown.

[0144] Example 2: Specific antibody clusters targeting SARS-CoV-2 and HIV-1

[0145] Based on this unique antibody QR code generation technology, clustered display of specific antibody datasets targeting SARS-CoV-2 and HIV-1 is performed.

[0146] 1. Antibody sequence sets for SARS-CoV-2 and HIV-1 were collected from public antibody databases, including 51 antibodies targeting the SARS-CoV-2 Spike protein (Table 1, data from https: / / doi.org / 10.1038 / s41586-022-04980-y) and 52 antibodies targeting the HIV GP-41 protein (Table 2, data from PDB database).

[0147] Table 1. Antibody numbers for SARS-CoV-2 Spike protein

[0148]

[0149]

[0150] Table 2. Antibody numbers for HIV GP-41 protein.

[0151] 6VKN 4KHT 5U3O 6NC2 1U8L 4KHX 5UM8 6NC3 6V6W 4NRX 5UTF 6NF2 3LEX 4XC1 5WDF 6NQD 6VLR 4YDV 6CDI 6O42 3LEY 5I8H 6CUE 6OLP 6VN0 5IQ7 6CUF 6OPA 3MNW 5IQ9 6CUF 6OT1 7KMD 5JSA 6CUF 6OT1 3P30 5U3J 6MPH 6VPX 7LG6 5U3L 6MTT 6X9R 4G6F 5U3M 6N1V 7EKB 7N65 5U3N 6N1W 7EKK

[0152] 2. Following the calculation steps in Implementation Case 1, generate a unique and distinctive QR code for each antibody, such as... Figure 2 As shown.

[0153] 3. The features of SARS-CoV-2 and HIV antibody QR codes were extracted using the ResNet50 image deep learning model. The distance between the feature representation matrices was calculated according to the formula Fun_Feacorr=1-xy / ||x||||y||. Finally, the distance matrix QR_mat representing the differences between antibody-specific two-dimensional maps was generated. Finally, the high-dimensional data visualization and dimensionality reduction tool UMAP was used for dimensionality reduction visualization to display antibody clusters.

[0154] Clustering results showed that although both SARS-CoV-2 and HIV-1 specific antibodies are induced by the virus, these two types of antibodies cannot be distinguished solely by sequence similarity (BLAST: an online sequence similarity comparison tool, https: / / blast.ncbi.nlm.nih.gov / Blast.cgi). Antibody QR codes generated using this method can differentiate between antibodies targeting SARS-CoV-2 and HIV-1, such as... Figure 3 As shown, this demonstrates that the antibody QR code generation technology of the present invention can effectively distinguish antibody sequences with different targeting properties.

[0155] Example 3: Comparison of the clustering analysis performance of different evaluation tools for antibody target specificity

[0156] Using antibody datasets targeting different species as an example, clustering was performed according to the methods described in Examples 1 and 2, and the performance was compared with Ab-Ligity, the only currently available tool for evaluating antibody targeting specificity, and BLAST, a tool for evaluating sequence similarity.

[0157] The results showed that antibodies targeting the SARS-CoV-2 spike protein and lysosomes, two types of antibodies with completely different target specificities, could be distinguished. Figure 4 a. Data are shown in Tables 1 and 3). Furthermore, Ab-QRcode can accurately separate antibodies from human and murine origins into two distinct clusters, demonstrating its ability to discern subtle differences among antibodies of diverse origins. While Ab-Ligity can differentiate antibodies from different sources (mutine and human), it incorrectly clusters some lysosome-specific antibodies with antibodies specifically targeting the SARS-CoV-2 spike protein, such as... Figure 4 As shown in d.

[0158] Table 3 Lysosomal Antibody Numbers

[0159] 1BVK 1NDG 2DQG 1FDL 1NDM 2DQH 1G7J 1P2C 2EIZ 1IC4 1UA6 2EKS 1IC5 1XGP 2YSS 1IC7 1XGQ 4TSA 1J1X 1YQV 4TSB 1KIP 2DQC 4TSC 1KIQ 2DQD 4TTD 1MLC 2DQE 6P4A 1NBY 2DQF 6P4B

[0160] Ab-QR codes can also effectively distinguish between antibodies with similar maturation environments induced by the same virus, such as SARS-CoV-2 and HIV-1 specific antibodies. Figure 4 b). However, Ab-Ligity incorrectly classified 18 of the 51 SARS-CoV-2 spike protein-specific antibodies as HIV-1 GP-41-specific antibodies, such as... Figure 4 As shown in e.

[0161] Distinguishing antibodies against multiple subtypes of viruses, such as influenza viruses H1N1 and H3N2, is a challenging but meaningful task. Since many antibodies can target both viruses simultaneously, this invention uses enzyme-linked immunosorbent assay (ELISA) to filter out cross-protective antibodies between H1N1 and H3N2, resulting in 10 antibodies specifically targeting H1N1 HA (Table 4, data from http: / / dx.doi.org / 10.1016 / j.cell.2016.06.043) and 3 antibodies specifically targeting H3N2 HA (Table 5, http: / / dx.doi.org / 10.1016 / j.cell.2016.06.043). Hierarchical clustering based on Ab-QR codes shows a clear clustering tendency between these two types of antibodies. Figure 4 c). Although Ab-Ligity also shows a clear tendency to differentiate between these two types of antibodies, it cannot recognize antibodies H2897 and CH65 (PDB id: 5ug0, 5ugy) that target the same epitope. Ab-QRcode, however, can cluster these two antibodies with completely different sequences but the same targeting at the nearest intragroup distance. Figure 4 cf)

[0162] Table 4. Antibody numbers for H1N1 HA

[0163] H1N1.294_16_IgG_1D04 H1N1.294_16_IgG_1D04 H1N1.300_16_IgG_2E02 H1N1.300_16_IgG_2E02 H1N1.301_16_IgA_1G04 H1N1.301_16_IgA_1G04 H1N1.294_16_IgA_1E04 H1N1.294_16_IgA_1E04 H1N1.294_16_IgG_1G04 H1N1.294_16_IgG_1G04

[0164] Table 5. Antibody numbers for H3N2 HA

[0165] H3N2.229_14_1C01 H3N2.229_14_2E06 H3N2.228_14_2E05

[0166] BLAST, which clusters based solely on sequence similarity rather than antibody target specificity, is therefore completely incapable of addressing this specific problem. Figure 4 gi).

[0167] To quantify the clustering effect of the models, clustering scores are introduced to measure the predictive performance of the three methods. Generally, the more distant the phylogenetic relationship between antigens, the easier it becomes to classify specific antibodies targeting that antigen, and therefore the higher the clustering score between specific antibodies will be. A higher score indicates better differentiation of antibodies targeting different specificities.

[0168] The results show that Ab-QRcode outperforms other methods, achieving a maximum performance improvement of 159.46% compared to Ab-Ligity and a maximum performance improvement of 296.69% compared to BLAST. Figure 4 As shown in jk.

[0169] Example 4: Comparison of cluster analysis performance of different evaluation tools for antibodies targeting different antigen proteins / domains of the same species

[0170] Using antibody datasets targeting different antigen proteins / domains of the same species as an example, clustering was performed using the methods described in Examples 1 and 2.

[0171] To evaluate whether Ab-QRcode can be used to distinguish antibodies induced by the same virus but targeting different viral antigens or even different domains, this invention collected antibodies targeting the H1N1 HA and NA proteins (Table 4) and the SARS-CoV-2 spike protein RBD and S2 domains (Table 6).

[0172] Table 6. Antibody IDs for the S2 domain

[0173]

[0174]

[0175] The results showed that Ab-QRcode exhibited good prediction performance on both datasets, especially demonstrating strong discriminative ability for the two specific antibody classes targeting the RBD and S2 domains, with a clustering score as high as 4.14. Figure 5 While Ab-Ligity performs well in distinguishing specific antibodies targeting different species, it performs poorly in distinguishing specific antibodies targeting different antigen proteins / domains of the same species. Figure 5 This also illustrates the significant challenge of identifying specific antibodies targeting different antigenic proteins / domains of the same species. In contrast, BLAST performs slightly better than other methods in distinguishing specific antibodies targeting HA and NA proteins, due to the higher sequence similarity between antibodies with the same targeting. However, its predictive performance drops significantly on datasets of antibodies with the same targeting specificity but large sequence differences; BLAST is completely unable to distinguish specific antibodies targeting the RBD and S2 domains. Figure 5In summary, Ab-QRcode's prediction performance far surpasses Ab-Ligity, with a maximum improvement of 187.50% (fg). Figure 5 h).

[0176] Example 5: Comparison of cluster analysis performance of different evaluation tools for antibodies targeting different epitopes

[0177] Using antibody datasets targeting different epitopes as an example, clustering was performed using the methods described in Examples 1 and 2.

[0178] Typically, the target specificity of some antibodies can be inferred through sequence similarity, for example, by assuming that antibodies from the same clone or lineage will exhibit similar binding patterns. However, there are also cases where CDR sequences and genotypes are completely different, yet the target specificity remains consistent. This invention introduces two antibodies (H2897 and CH65) with highly consistent binding epitope regions into the validation dataset, such as... Figure 6 As shown in a.

[0179] Sequence alignment results showed that the sequences of antibodies H2897 and CH65 were completely different, with only 51.97% sequence similarity between the heavy chains and only 47.66% sequence similarity between the light chains. Figure 6 As shown in b. Ab-QRcode can effectively group these two antibodies together with the shortest distance, such as... Figure 4 As shown in c. However, Ab-Ligity and BLAST cannot ( Figure 6 (f, i). This indicates that Ab-QRcode has excellent recognition ability for antibodies with large sequence differences but completely consistent target specificity.

[0180] To further explore the predictive ability of Ab-QRcode in distinguishing antibodies targeting similar epitopes, this invention collected data on 634 antibodies targeting the RBD domain of the SARS-CoV-2 spike protein (data from https: / / doi.org / 10.1038 / s41421-024-00648-1). In this antibody dataset, experimental data on competitive binding by ELISA for five specific antibodies (PW5-4, PW5-5, PW5-534, PW5-535, and PW5-570) were reported to assess whether these antibodies bind to the same or highly adjacent epitope regions.

[0181] The results of cluster analysis of 634 antibody clusters based on Ab-QR codes showed that PW5-4, PW5-5, PW5-534, and PW5-570 were relatively close together, while PW5-535 appeared in a more distant cluster, such as... Figure 6As shown in c. It is worth noting that the distance heatmap of Ab-QRcode shows that PW5-4 and PW5-5 have the highest scores and exhibit the most similar combination patterns, such as... Figure 6 As shown in d, PW5-534 and PW5-570 competitively bind to PW5-4 and PW5-5, respectively, indicating similar targeting properties. Furthermore, PW5-535 exhibits a completely different targeting specificity compared to the other four antibodies, which is reflected in the Ab-QRcode similarity score (clustered individually). In summary, as... Figure 6 As shown in the figure, the results inferred based on Ab-QRcode are in high agreement with the results of the competitive binding assay of the five antibodies with ELISA. However, for this very challenging task, BLAST performed poorly and failed to achieve the goal of clustering antibodies according to their target specificity. These results indicate that Ab-QRcode can recognize antibodies that target the same or adjacent epitopes but have completely different CDRH3 sequences.

[0182] Example 6: Cluster analysis of antibodies from healthy and diseased individuals using Ab-QRcode.

[0183] Bone marrow fluid samples were collected from 3 patients with systemic lupus erythematosus (SLE) and 5 healthy individuals (HD). Single-cell suspensions were obtained, and cells were encapsulated in gel beads with cell-tagged sequences within individual droplets using a 10X Genomics microfluidic chip. mRNA released after cell lysis was linked to the cell-tagged sequences of the gel beads and reverse transcribed into cDNA within the droplets. 5' single-cell transcriptome sequencing and BCR / TCRV(D)J sequencing were performed using an Illumina sequencing platform to obtain high-quality single-cell sequencing data, from which BCR antibody sequences were annotated. Antibodies from SLE and HD samples were merged based on antibody expression levels. After redundancy removal, 12 high-expression antibodies for HD and 15 high-expression antibodies for SLE were obtained. Ab-QR codes were used to generate QR code maps for each antibody, such as... Figure 7 As shown.

[0184] Using a method similar to that in Examples 1 and 2, the ResNet50 image deep learning model was used to extract the QR code map feature matrices of HD donor antibodies and SLE donor antibodies. The distance between each pair of antibody map feature matrices was calculated to obtain a distance matrix QR_mat that describes the differences between antibody maps. The distance matrix QR_mat was then subjected to dimensionality reduction and clustering using the high-dimensional data visualization and dimensionality reduction tool UMAP. The final result is as follows: Figure 8 As shown.

Claims

1. An antibody QR code (Ab-QRcode), wherein the antibody QR code is generated by mapping the antibody primary sequence onto an ordered two-dimensional map, and can characterize the species specificity, type and target specificity information of the antibody.

2. A method for generating an antibody QR code as described in claim 1, the method specifically comprising the following steps: S1. Obtain a non-redundant antibody sequence dataset for generating antibody QR code frameworks; S2. The antibody sequences in the dataset obtained in step S1 are split to form a set of split peptide segments; S3. Select an amino acid feature set to describe the split peptide set and generate a peptide-specific feature matrix; S4. Calculate the correlation between every two columns in the peptide-specific feature matrix obtained in step S3, and generate a feature distance matrix; S5. Project the feature distance matrix obtained in step S4 onto a two-dimensional space to obtain the antibody QR code framework; S6. The antibody sequence to be generated into a QR code is split in the manner described in step S2. The peptide segments are described using the feature set selected in step S3 to generate a one-dimensional feature value matrix. Each feature value in the feature value matrix is ​​mapped to the frame obtained in step S5 according to the feature point position. The color saturation of the point is determined according to the feature value, thereby generating the antibody QR code.

3. The method as described in claim 2, wherein in step S2, the antibody sequence splitting method is as follows: the framework region sequence in each antibody sequence is sequentially split into peptides of length n1 amino acids, the CDR1 and CDR2 sequences in each antibody sequence are sequentially split into peptides of length n2 amino acids, and the CDR3 sequence is sequentially split into peptides of length n3 amino acids, wherein, n1∈[1,20] and is an integer, n2∈[1,10] and is an integer, n3∈[1,5] and is an integer.

4. The method according to claim 2, in step S3, the amino acid feature set is selected from one or more of the following: amino acid composition, protein secondary conformation, amino acid physicochemical properties, evolutionary conservation, microenvironment amino acid distribution and / or microenvironment amino acid composition.

5. The method as described in claim 2, in step S3, the peptide-specific feature matrix is ​​determined according to the formula... Determined, where m represents the total number of antibody sequences in the non-redundant antibody sequence dataset, n represents the number of peptides split into for each antibody, p represents the number of types of the selected feature set, and the feature value X. ij (1≤i≤m, 1≤j≤n) by the formula FeaScore (a) is calculated by summing the characteristic values ​​of amino acids in each peptide segment. i () represents the predicted or calculated score in the feature set corresponding to each amino acid. This indicates that the score is accumulated, and Np is the number of amino acids contained in the peptide segment.

6. The method as described in claim 2, in step S4, the correlation is calculated according to the formula Fun_corr=1-xy / ||x||·||y||, respectively, to determine the distance Y between every two columns in the same feature set of Pep_Mat. ij (1≤i≤q,1≤j≤q), ultimately generating a feature distance matrix representing peptide correlation. in, x and y represent the feature values ​​in Pep_Mat, respectively, q = n × p, where n represents the number of peptides split into each antibody, and p represents the number of feature sets used.

7. The method as described in claim 2, in step S5, the antibody QR code framework is based on the intrinsic correlation between associated antibody sequences, by projecting the feature distance matrix Fea_mat onto a two-dimensional feature space using high-dimensional data visualization and dimensionality reduction tools, and then using the formula Fun_map=||x embed -y grid || 2 Feature points are generated by linearly assigning them to a two-dimensional grid, and then coloring the point according to the feature set category or subset category to which it belongs. x embed Let y represent the unprocessed feature distance matrix. grid Let ||x| represent the feature distance matrix after linear assignment. embed -y grid || 2 This indicates the minimum loss value.

8. A system for constructing antibody QR codes, the system comprising a data input module, a data processing module, and a data output module: The data input module collects antibody data according to the method described in any one of claims 2 to 7 and inputs the antibody data. The data processing module obtains a non-redundant antibody sequence dataset, splits the antibody sequence, selects a feature set, constructs a peptide-specific feature matrix, generates a feature distance matrix, obtains an antibody QR code framework, and generates an antibody QR code according to the method described in any one of claims 2 to 7. The data output module displays the antibody QR code.

9. A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, performs the steps of any one of claims 2 to 7.

10. The use of the antibody QR code as described in claim 1 in antibody target recognition specific cluster analysis includes one or more of the following application directions: A) Differentiate antibodies produced by the human body under physiological and / or various pathological conditions; B) Antibodies that distinguish between different antigens or epitopes.

Images