Method for designing personalized cancer vaccine using b cell-reactive neoantigen

Abstract

The present disclosure relates to a method for designing a personalized cancer vaccine, using a B cell-reactive neoantigen, the method comprising the steps of: receiving a peptide sequence extracted from cancer tissue and allelic sequence data of B cell receptors; calculating the preferential binding strength between amino acids in the peptide sequence and allelic sequences of B cell receptors; and calculating B cell epitope prediction scores by inputting the preferential binding strength between amino acids to a trained neural network model.

Description

Design of a personalized cancer vaccine using B-cell-reactive neoantigens

[0001] The present disclosure relates to a method for designing a personalized cancer vaccine using a B cell reactive neoantigen and a computer program stored in a recording medium.

[0002] Cancer is one of the leading causes of death worldwide, and various treatments, including resection and chemotherapy, are being implemented clinically to treat various types of cancer. Among these, immunotherapy, which suppresses cancer by reactivating immune cells in the body, has proven its efficacy in various clinical trials. Specifically, immunotherapy using immune checkpoint inhibitors treats inflammatory cancers infiltrated by primed, tumor-responsive T cells.

[0003] Cancer vaccines treat cancer by introducing novel antigens that target cancer-causing viruses or cancer cells, generating primed cancer-responsive T cells. They are emerging as a core technology for immunotherapy in combination with checkpoint blockade. A crucial element in the development of cancer vaccines is determining the antigen to be used. This is achieved when a cancer-causing mutation is identified as a neoantigen through genome analysis of an individual cancer patient. If the neoantigen is found in common across multiple cancer patients, it can be developed into a universal cancer vaccine. If it is found only in a specific patient or cancer cell, it can be developed into a personalized cancer vaccine.

[0004] Recently, machine learning-based algorithms have been used to discover suitable neoantigens with immunogenicity, and attempts are being made to increase accuracy by including steps such as binding affinity with MHC class I and II, T cell receptor (TCR) binding affinity, and antigen processing in the prediction model.

[0005] While anti-tumor immunity has primarily focused on T cells, the recent discovery of tertiary lymphoid structures (TLS) has highlighted the crucial and diverse roles of B cells. Within these structures, B cells function to activate T cells through antigen recognition and interaction. Recently, it has been demonstrated that B-cell-specific checkpoint molecules regulate tumor-specific effector T cell activity through antigen presentation and co-stimulation. Furthermore, in murine cancer models, it has been demonstrated that specific neoantigens can activate tumor-specific B cells and CD4+ T cells, generating follicular helper T cells that produce IL-21. However, it remains unclear whether B cells, in general, contribute to neoantigen-induced immune responses in various types of human cancer. This is because reliable methods for identifying CD4+ T cell- and B cell-reactive neoepitopes have not been developed.

[0006] The purpose of the present invention is to provide a method for designing a customized cancer vaccine using a B cell reactive neoantigen and a computer program stored in a recording medium.

[0007] However, the problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned can be clearly understood by a person having ordinary skill in the relevant technical field from the description below.

[0008] A method for designing a customized cancer vaccine according to one embodiment of the present invention may include the steps of: receiving a peptide sequence extracted from cancer tissue and allelic sequence data of a B cell receptor; calculating a selective binding strength between amino acids of the peptide sequence and the allelic sequence of the B cell receptor; and inputting the selective binding strength between amino acids into a learned neural network model to calculate a B cell epitope prediction score.

[0009] According to one embodiment of the present invention, the step of calculating the selective binding strength between amino acids of the peptide sequence and the allele sequence of the B cell receptor may include the step of generating a matrix including Cα-Cα distance information between amino acids of the peptide sequence and the B cell receptor allele sequence and atomic contact information between amino acid residues.

[0010] According to one embodiment of the present invention, the matrix may include somatic hypermutation (SHM) frequency data per residue of the B cell receptor allele superfamily.

[0011] According to one embodiment of the present invention, the cell values ​​constituting the matrix are defined by the following equation 1:

[0012] [Formula 1]

[0013]

[0014] Here, i and j represent the positions of the immunoglobulin heavy chain variant (IGHV) and peptide sequence of the B cell receptor, and Ⅰ ij is IGHV i Wow peptide j It can exhibit selective binding interactions between them.

[0015] According to one embodiment of the present invention, the neural network model is a convolutional neural network (CNN) model, and the CNN model may include a convolutional layer that detects spatial interactions between the immunoglobulin heavy chain variant (IGHV) of the B cell receptor and a peptide sequence.

[0016] The above convolution layer outputs a feature map through a convolution operation on the input image. The filter that performs the convolution operation is also called a kernel. The size of the filter is called the filter size or kernel size. The operation parameters that constitute the kernel are called kernel parameters, filter parameters, or weights.

[0017] According to one embodiment of the present invention, the CNN model may be composed of a plurality of convolutional layers, a fully connected layer, and a sigmoid layer.

[0018] According to one embodiment of the present invention, the kernel of the CNN model may include the entire amino acid sequence before the CDR3 (complementarity determining region 3) of the immunoglobulin heavy chain variant (IGHV) of the B cell receptor.

[0019] According to one embodiment of the present invention, the convolutional layer can extract features of contact and distal interactions using a pooling-free ReLU (Rectified Linear Unit) function.

[0020] According to one embodiment of the present invention, the neural network model may be an ensemble model that aggregates multiple CNN models.

[0021] According to one embodiment of the present invention, the step of calculating the B cell epitope prediction score may include a step of applying linear regression to the output value of the neural network model.

[0022] According to one embodiment of the present invention, a neoantigen whose calculated B cell epitope prediction score exceeds a predetermined threshold may be determined to have a high immune response-inducing ability. Here, the predetermined threshold may be, but is not limited to, the average B cell epitope prediction score.

[0023] According to another embodiment of the present invention, a computer program stored in a computer-readable recording medium is provided to execute a method for designing a customized cancer vaccine utilizing the B cell reactive neoantigen in combination with hardware on a computer.

[0024] According to another embodiment of the present invention, a computer-readable recording medium is provided, which records a program for executing a method for designing a customized cancer vaccine using the B cell reactive neoantigen on a computer.

[0025] A method according to one embodiment of the present invention can be used to design a customized cancer vaccine by predicting a neoantigen that induces a strong immune response with high accuracy by calculating the binding strength of a B cell receptor and a B cell epitope.

[0026] A method according to one embodiment of the present invention can predict a neoantigen that strongly induces a T cell immune response of a cancer vaccine, and thus can design a cancer vaccine based on a neoantigen with better performance.

[0027] However, the effects of the present invention are not limited to the above-described effects, and should be understood to include all effects that can be inferred from the composition of the invention described in the detailed description or claims of the present invention.

[0028] Figure 1 shows the construction and evaluation of a B-cell epitope prediction model, which shows (A) a schematic diagram of a B-cell epitope prediction model that identifies B-cell epitopes based on a CNN architecture, (B) comparison of the area under the receiver operating characteristic curve (ROC-AUC) between individual allele models and an ensemble model, (C) comparison of the area under the precision-recall curve (PR-AUC) between individual allele models and an ensemble model (statistics are calculated using the Welch two sample t test), (D) comparison of the ROC-AUC values ​​of the B-cell epitope prediction model (DeepNeo-BCR) and existing linear B-cell epitope prediction tools, and (E) comparison of the ROC-AUC values ​​of the B-cell epitope prediction model (DeepNeo-BCR) and existing linear B-cell epitope prediction tools using a shared test dataset (IEDB items that were not used for model construction were used).

[0029] Figure 2. Validation of SARS-CoV-2 vaccine data: (A) Schematic of clustering of bulk BCR sequencing data; (B) Distribution of predicted scores of B cell epitope prediction models for large clones (>mean read fraction) and small clones (<mean read fraction) after Pfizer SARS-CoV-2 vaccination in sample (ID: CME93) (D = 0.32, P = 0.049 in Kolmogorov-Smirnov test); (C) Odds ratio of high-scoring BCR clones belonging to large clonal groups per patient; (D) Schematic of clustering of single-cell BCR sequencing data (pre- and post-vaccination BCR sequences of each patient sample are clustered by sequence similarity); (E) Odds ratio of high-scoring BCR clones that underwent clonal expansion determined by differences in cell fraction before and after vaccination (odds ratio P values ​​were calculated using Fisher exact test); (F) Predicted affinity calculated by B cell epitope prediction models. (G) Odds ratio of high-scoring BCR clones belonging to the large (> mean cell fraction) clone group using post-vaccination data (odds ratio P values ​​calculated using Fisher's exact test), (H) Sequences of the three expanded clones with the highest B-cell epitope prediction scores, (I) Cell annotation for 76,843 cells from 24 samples post-vaccination (J) Difference in mean B-cell epitope prediction scores between mature and naïve B cells (calculated using Welch Two Sample t-test), (K) Odds ratio of high-scoring BCR clones belonging to the mature B-cell population (odds ratio P values ​​calculated using Fisher's exact test).

[0030] Figure 3 shows the impact of neoantigen vaccination on B cell reactivity in mice, showing (A) differences in IFNγ signaling between TCR(+)BCR(high) and TCR(+)BCR(low) vaccines (peptides n = 44 and ELISpot assays n = 264) (assigned to BCR-high and -low groups based on the maximum difference between mutant and wild-type B cell epitope prediction scores for each potential neoepitope); (B) differences in IFNγ signaling between TCR(+)BCR(high) and TCR(+)BCR(low) vaccines for EMT6 breast cancer, LLC1 lung cancer, and B16F10 melanoma models; (C) comparison of individual ELISpot results between TCR(+)BCR(high) and TCR(+)BCR(low) vaccines (each dot represents an individual mouse tested with the vaccine).

[0031] Figure 4 shows the impact of B cell reactivity in humanized neoantigen vaccination, (A) a schematic diagram showing vaccines with somatic mutations, potential neoepitopes, and predicted TCR / BCR reactivity, with multiple 15-mer peptides searched in the vaccine sequence and tested for T cell and B cell reactivity using DeepNeo-TCR and B cell epitope prediction models, respectively, (B) the percentage of neoepitopes predicted to stimulate CD4+ or CD8+ T cell responses normalized by vaccine length compared between IFNγ(+) and IFNγ(-) vaccines, confirming the effect of TCR(+) epitopes in inducing immune responses, (C) the percentage of neoepitopes predicted to stimulate both T and B cells normalized by vaccine length compared between IFNγ(+) and IFNγ(-) vaccines, confirming the effect of TCR(+)BCR(+) neoepitopes in stimulating immune responses. (D) Proportion (Odds) of IFNγ(+) vaccines with high TCR(+)BCR(+) ratios (left: > 20%, right: > 30%) (odds ratio P-values ​​were calculated via Fisher's exact test); (E) Correlation between B-cell epitope prediction scores and responses to vaccines at the patient level (highest B-cell epitope prediction score per patient across all neoepitopes and patient-level immune response (% of IFNγ(+) vaccine) was performed using Pearson's correlation test); (F) Examples of patient cases with high (top) and low (bottom) immune response rates (B-cell epitope prediction scores for potential neoepitopes derived from each vaccine are shown, with the red palette representing IFNγ(+) vaccines and the blue palette representing IFNγ(-) vaccines); (G) High proportion of B-cell reactive neoepitopes (> 20% BCR(+)) Odds (odds ratio P values ​​are calculated using Fisher's exact test) that patients receiving a specific epitope will develop a strong immune response (> 70% of the delivered vaccine).

[0032] Figure 5 shows the correlation between B cell neoepitope load and tumor immune repertoire features, showing (A) the correlation between B cell neoepitope load and BCR features (clonality and entropy) averaged across cancer types in TCGA pan-cancer samples (n = 4,435), and (B) the correlation between B cell or T cell neoepitope load and BCR or TCR features (clonality and entropy) across cancer types (TCGA samples with MHC genotype information were used (n = 3,392)).

[0033] Figure 6 illustrates the impact of mouse neoantigen vaccination on B cell reactivity, repeating the same analysis as in Figure 2 except that the BCR-high and -low groups were classified by their mutant B cell epitope prediction scores, rather than their wild-type B cell epitope prediction scores. The maximum score for all potential neoepitopes for each vaccine was retrieved and used for classification.

[0034] Figure 7 shows the results of the immune response analysis at the patient level, showing (A) the correlation between DeepNeo-TCR scores at the patient level and responses to vaccines using DeepNeo-TCR for both CD8+ and CD4+ epitopes (the highest DeepNeo-TCR score among all neoepitopes per patient and the patient-level immune response (% of IFNγ(+) vaccine) was subjected to Pearson correlation test), (B) the correlation between BCR(+) epitope abundance at the patient level and the response rate to vaccines (the x-axis is the percentage of predicted B cell binding epitopes among all putative peptides derived from all vaccines delivered to each patient).

[0035] Figure 8 illustrates patient-level immune response analysis using different epitope prediction tools, calculating the correlation between patient-level B-cell epitope scores for all epitopes that could be derived from the vaccine sequence using LBtope, LBEEP, IgPred, and BepiPred3.0. The patient's maximum B-cell epitope score was used as the patient-level B-cell epitope score.

[0036] Figure 9 shows additional patient-level immune response analysis using different epitope prediction tools. LBtope, LBEEP, IgPred, and BepiPred3.0 were used to calculate the probability of patients receiving a high proportion of predicted neoepitopes (>20% BCR(+) epitopes) that elicited a strong immune response (>70% of delivered vaccine). LBtope, IgPred, and BepiPred3.0 had no expected effect, while LBEEP had marginal power with an OR=2.19 (P=0.13) (odds ratio P values ​​were calculated via Fisher's exact test).

[0037] Figure 10 shows sample-wise correlations between B cell neoepitope load and immune repertoire conditions, showing (A) the correlation between BCR(+) epitope load and BCR clonality for samples with matched mutation and immune repertoire data (n = 4,214), and (B) the correlation between BCR(+) epitope load and BCR entropy using individual samples (correlation coefficients were calculated using the Pearson correlation test).

[0038] Hereinafter, embodiments are described in detail with reference to the attached drawings. However, the embodiments may be modified in various ways, and the scope of the patent application is not limited or restricted by these embodiments. It should be understood that all modifications, equivalents, or alternatives to the embodiments are included within the scope of the patent application.

[0039] The terms used in the examples are for illustrative purposes only and should not be construed as limiting. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "have" are intended to indicate the presence of a feature, number, step, operation, component, or combination thereof described in the specification, but should be understood to not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, or combinations thereof.

[0040] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the embodiments pertain. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and shall not be interpreted in an idealized or overly formal sense unless explicitly defined herein.

[0041] As used herein, the term "neoantigen" refers to a completely new protein formed in cancer cells when a specific mutation occurs in tumor DNA, and can appear as an epitope on cancer cells. It is mainly caused by non-synonymous mutations, and unlike tumor-associated antigens (TAAs) or cancer-germline antigens (CGAs), which are also expressed to some extent in normal cells, it is characterized by being expressed only in cancer cells. The terms "epitope," "antigenic peptide," "immunogenic peptide," and "peptide" may be used interchangeably.

[0042] In this specification, the term 'neoepitope' refers to an epitope found in diseased cells, such as cancer cells.

[0043] In this specification, the term 'immunogenicity' refers to the ability to induce an immune response in the body, and in the context of cancer treatment, refers to the ability to induce an immune response associated with a treatment regimen.

[0044] In this specification, the term 'peptide' means an amino acid polymer, and may refer to an amino acid polymer or amino acid sequence expressed on the surface of cancer cells.

[0045] In this specification, the term 'exome' refers to a subset of the genome that encodes proteins. An exome may refer to a collection of exons present in a cell, a group of cells, or an individual.

[0046] In this specification, 'DeepNeo-BCR' refers to a cancer vaccine personalized design model utilizing B cell epitope prediction that performs a method according to one embodiment of the present invention.

[0047] When describing an embodiment, if it is determined that a detailed description of a related known technology may unnecessarily obscure the gist of the embodiment, the detailed description is omitted.

[0048] To facilitate understanding of the present invention, preferred examples are presented below. However, the following examples are provided solely to facilitate understanding of the present invention, and the scope of the present invention is not limited by the examples.

[0049]

[0050] Experimental Example 1: Data Construction

[0051] A list of linear epitopes dated before January 17, 2023, was collected from the immune epitope database (IEDB) through in vitro binding tests. Several filters were applied to the collected data: (1) the peptides had to be of the desired length (12, 15, or 16-mer), (2) the peptide binding experiments with B cells had to be tested in humans or mice, (3) the peptides had to be linear, not structural, and (4) the peptides had to be tested in IgG. After applying the filters, a total of 437,396 data sets were entered, in which peptide binding to B cells was measured by qualitative measurements, cytokine secretion, direct measurements, or X-ray crystallography-based binding assays. Because experimental results may vary depending on experimental conditions, if conflicting results were collected for the same peptide, the peptide was marked as a binding epitope. Because the peptides varied in length, padding was applied to the longest peptide (16-mer). More specifically, padding was applied to the first and last two residues for 12-mers, and to the last residue for 15-mers. The collected data were randomly split into training / validation and test sets in a 7:3 ratio to accurately evaluate model performance. Because B cell-epitope binding was measured as a general reactivity rather than a specific BCR sequence, the model was designed to comprehensively predict peptide binding to multiple BCR sequences. More specifically, 48 human and 203 mouse immunoglobulin heavy chain variant (IGHV) alleles were collected from the IMGT database, and the human alleles were used as is.

[0052] For mouse alleles, one representative allele per superfamily was used to reduce computational burden. As a result, a total of 63 representative alleles and their corresponding base sequences were collected, including 48 human and 15 mouse alleles.

[0053] Predicting B-cell epitopes is more challenging than predicting T-cell epitopes, partly due to somatic hypermutation (SHM) occurring in IGHV sequences. To account for SHM, we obtained the SHM frequency per residue of the previously calculated IGHV allele superfamily. This rate was calculated as the average SHM rate per nucleotide position from 249,972 clones. To convert to protein-level data, the SHM rate per amino acid position was calculated as the average of three nucleotide sequences. If the superfamily of the query allele was provided, it was used as is; otherwise, the SHM rate of all alleles was used.

[0054]

[0055] Experimental Example 2: Building a B-cell Epitope Prediction Model

[0056] To capture the unique local features of amino acid interactions between B-cell receptors (BCRs) based on a convolutional neural network (CNN) framework, we developed a B-cell epitope prediction model (DeepNeo-BCR) leveraging the existing DeepNeo model architecture (Jeong Yeon Kim and others, DeepNeo: a webserver for predicting immunogenic neoantigens, Nucleic Acids Research, Volume 51, Issue W1, 5 July 2023, Pages W134-W140). To address the variability of the BCR repertoire, we leveraged sequence information contained in >370 million clones retrieved from 2,152 Ig sequencing samples (Yang, X. et al. Large-scale analysis of 2,152 Ig-seq datasets reveals key features of B cell biology and the antibody repertoire. Cell Rep. 35, 109110 (2021)).

[0057] The CNN architecture aimed to capture the selective binding of BCRs to epitopes at the amino acid level. A two-dimensional matrix containing the interaction strengths between amino acids of BCRs and epitopes was generated based on previously calculated Cα-Cα distances (Jha, AN, Vishveshwara, S. & Banavar, J.R. Amino acid interaction preferences in proteins. Protein Sci. 19, 603-616 (2010)). In summary, the selective binding matrix for each protein was generated by considering the Cα-Cα distances between amino acids, using a cutoff distance of 6.5 Å, excluding the nearest neighbors in the sequence. In addition, atomic contacts between amino acid residues were also considered, and residues i and j were considered to be in contact if an atom of residue i and residue j is within a distance of 4.5 Å from an atom of residue j.

[0058] We used the selective binding strength matrix between these amino acids to generate CNN model input. Furthermore, we implemented the SHM ratio as an additional weight per residue to account for IGHV diversity. The SHM ratios calculated in previous studies were first scaled from 0 to 1 for each superfamily.

[0059] Normalized. Then, the SHM frequency per residue was reduced by a factor of 10, using the SHM ratio as an auxiliary weight for amino acid preferential binding. The input value per matrix cell is defined as follows: Equation 1.

[0060] [Formula 1]

[0061]

[0062] Here, i and j represent the positions of IGHV and peptide sequences.Ⅰ ij is IGHV i Wow peptide jIt indicates selective binding interaction between IGHV and SHM frequency. iis given additional weights according to the value of . Since IEDB contains only information on B-cell binding and not the full sequence or genotype of IGHV alleles, we created an input matrix of preferential amino acid binding between 63 representative IGHV alleles and peptides. For each allele, a CNN model was generated, resulting in 63 independent models predicting the binding of that allele to the peptide. The final prediction score was calculated by applying linear regression to the individual CNN models. Because predicting all 63 models per peptide is computationally difficult, linear regression models were built for each purpose, such as using the full set (63 alleles), the human set (48 alleles), the human subset (25 alleles), and the mouse set (15 alleles). The subset of human alleles was defined as the representative allele with a sequence similarity less than 80% as calculated by CD-HIT. The CNN model for individual alleles consisted of two convolutional layers, a fully connected layer, and a sigmoid layer. The convolutional layers were designed to capture spatial interactions between IGHV and peptide sequences. To account for the non-contiguous nature of contact residues in BCR due to protein folding, the first kernel encompassed the entire IGHV sequence. The convolutional layers used a rectified linear activation function (ReLU) without pooling to extract features of both contact and distal interactions. Two hundred epochs were used to optimize the model, and various hyperparameter sets were used to ensure optimal performance per CNN model. Specifically, various combinations of parameters were tested, including learning rate (0.001, 0.01, 0.1), momentum (0.1, 0.5, 0.9), number of kernels in layers 1 and 2 (10, 30, 50), and regularization parameter (0.0001, 0.001, 0.01), to identify the optimal prediction accuracy.The model was tested with Keras on a TITAN XP GPU, and statistical analysis was performed with the Python scikit-learn package.

[0063] For model training, we collected IgG binding peptides and non-binding peptides of fixed length (12-, 15-, and 16-mer) tested in humans or mice from IEDB, and calculated the ratio of binding peptides per peptide length (Table 1).

[0064] Peptide LengthTotal Peptide Binding Peptides (%)12mer107,64077574 (72.1%)15mer242,31729477 (12.2%)16mer87,43839908 (45.6%)Total437,395146959 (33.6%)

[0065] Because most of this dataset was based on testing non-specific pools of the BCR repertoire, we designed an ensemble model that aggregated 63 CNN models, each representing a human or mouse immunoglobulin heavy chain variant (IGHV) allele. Additionally, somatic hypermutation (SHM) frequencies per IGHV residue were incorporated into the model ( Figure 1A ).

[0066]

[0067] Experimental Example 3: Prediction of Linear B Cell Epitopes

[0068] As described in Experimental Example 1 above, each peptide of the epitope was tested against all individual IGHV alleles, and linear regression was applied to the obtained prediction values ​​to produce the final prediction values. B cell epitope prediction models were constructed using four ensemble models: all allele sets, all human alleles, all mouse alleles, or representative human alleles. As a result, the ensemble B cell epitope prediction model showed better performance than the model based on individual alleles (Fig. 1B-C).

[0069] Furthermore, the performance of the ensemble B-cell epitope prediction model was compared with existing linear epitope prediction tools that analyze fixed-length peptides, such as LBtope (Singh, H., Ansari, H.R. & Raghava, GPS Improved method for linear B-cell epitope prediction using antigen's primary sequence.PLoS One8, e62216 (2013)), LBEEP (Saravanan, V. & Gautham, N. Harnessing Computational Biology for Exact Linear B-Cell Epitope Prediction: A Novel Amino Acid Composition-Based Feature Descriptor.OMICS19, 648-658 (2015).), and IgPred (Gupta, S., Ansari, H.R., Gautam, A., Raghava, GPS & Consortium, OSDD Identification of B-cell epitopes in an antigen for inducing specific class of antibodies.Biol. Direct8, 27 (2013).) and significantly outperformed the BepiBred-3.0 model (Clifford, JN et al. BepiPred-3.0: Improved B-cell epitope prediction using protein language models. Protein Sci. 31, e4497 (2022).) which works for both linear and conformational epitopes (Figure 1D). For further benchmarking, we collected 116 new IEDB data entries that were not included in model training or testing. On that data set as well, the B-cell epitope prediction model outperformed all other existing methods (Figure 1E).

[0070]

[0071] Example 1: Validation using SARS-CoV-2 vaccination data

[0072] B cells harboring BCR sequences that recognize immunogenic epitopes undergo clonal expansion, and B cell clonal expansion in response to SARS-CoV-2 vaccination has been reported in several studies. Therefore, we examined the BCR sequences of vaccinated individuals to test whether BCRs that underwent clonal expansion during COVID-19 vaccination had higher B cell epitope prediction scores for SARS-CoV-2 epitopes. First, we collected large-scale BCR sequencing data from nine subjects who had received the Pfizer SARS-CoV-2 mRNA vaccine. IGHV sequences were extracted and clustered into clones, which were divided into large (> average read rate) and small (< average read rate) groups (Figure 2A). Each IGHV sequence was affinity-evaluated using a B cell epitope prediction model against all translated vaccine sequences. As a result, the average B cell epitope prediction scores of large clones were higher than those of small clones (Figure 2B). Additionally, enrichment of high-affinity clones for large groups (> average B cell epitope prediction score) was calculated per sample, resulting in an average odds ratio of 1.5 (Figure 2C), suggesting that IGHV sequences with high predicted affinity for vaccine sequences are prone to clonal expansion.

[0073] To more accurately identify BCR clones and directly measure clonal expansion, we generated our own single-cell BCR sequencing data for 24 blood samples before and after Pfizer SARS CoV-2 vaccination (Table 2).

[0074] Characteristics N=33% Gender Male 1042 Female 1458 Age Median 36.5 Range 29-44.3 Vaccine Pfizer 24100

[0075] Individual cells from each sample were clustered into clones based on BCR sequence similarity, and clonal expansion was determined by comparing the proportion of matching BCR clones before and after vaccination (Fig. 2D). Epitope binding prediction was performed on the BCR sequences within the identified clones. As a result, clones with higher affinity for the vaccine sequence tended to undergo clonal expansion during vaccination more than clones with lower affinity (Fig. 2E-F). Furthermore, analysis of post-vaccination clones was repeated on a large BCR dataset, and, consistent with previous findings, larger post-vaccination clones (> average read rate) were more likely to have higher affinity for the vaccine sequence predicted by the B cell epitope prediction model (Fig. 2G). A high degree of concordance was also found between clonally expanded BCRs (Fig. 2H). Furthermore, we examined the matched single-cell transcriptome data, which collected a total of 76,843 cells, of which 4,146 were mature B cells and 6,185 were naïve B cells (Fig. 2I). Mature B cells had higher BCR scores than naïve B cells (Fig. 2J), leading to an overrepresentation of high-affinity BCRs in mature B cells compared to naïve B cells (Fig. 2K).

[0076] In summary, the consistency of results obtained across two different datasets validates the validity of the B-cell epitope prediction model. BCR clones with higher B-cell epitope prediction scores, which exhibit stronger predicted affinity for SARS-CoV-2 vaccine sequences, were shown to have an increased propensity for clonal expansion during COVID-19 vaccination.

[0077]

[0078] Example 2: Mouse vaccination data analysis

[0079] DeepNeo-TCR, a previously developed algorithm for evaluating T cell responses to a given peptide-MHC complex (Kim, JY et al. MHC II immunogenicity shapes the neoepitope landscape in human tumors. Nat. Genet. 55, 221-231 (2023)), was used to identify TCR(+) epitopes in three murine cancer models (EMT6 breast cancer, LLC1 lung cancer, and B16F10 melanoma). Based on the predictions, 46 synthetic long peptides were manufactured and administered to mice three times, and spleen cells were harvested and measured for IFNγ secretion using an ELISpot assay. To ensure reproducibility, each peptide was tested in three mice, and each mouse was analyzed twice, resulting in a total of six measurements per peptide. Consequently, a total of 276 measurements were made for the 46 candidate neoantigens. For most neoantigens, IFNγ signals were stronger than in the negative control group.

[0080] Here, we reanalyzed this dataset by running a B-cell epitope prediction model on all potential neoepitopes within the 44 synthetic long peptides after filtering out two outliers. Since wild-type sequences were available for all neoantigens, we subtracted the wild-type score from the mutation score to obtain a differential B-cell epitope prediction score for each neoepitope. We obtained the maximum differential B-cell epitope prediction score across all potential neoepitopes within the vaccine and divided the 44 vaccine constructs into two groups based on this score: TCR(+)BCR(high) indicates vaccines with above-average B-cell epitope prediction scores, and TCR(+)BCR(low) indicates vaccines with below-average B-cell epitope prediction scores (Table 3).

[0081] Cancer cell line mutant peptide sequence IFN analysis B cell epitope prediction model group B6F10Anln-p.E580GGELDVEKSQGEMDQVGAENSEEQE1,1,24,2,7,7low B6F10Cpsf6-p.D519GSRDRHGDYYRERSRERERHRDRDR11,19,1,2,3,6low B6F10Dnajb12-p.P54TLNQKTQSTGDHPQPTDTTHTTTKK0,0,23,7,9,9high B6F10Hjurp-p.Q402PPRMLPSPVEKWYREIKIKFDKLHQ0,0,22,23,8 ,9lowB6F10Hnrnpk-p.R326GPPPPRGGDLMAYDRGGRPGDRYDG0,10,17,1,7,7highB6F10Macroh2a1-p.P107LTIASGGVLLNIHPELLAKKRGSKG106 ,19,36,6,6,72lowB6F10Map2k2-p.A29GTINPTIAEGPSPTSEGGSEANLVD16,19,1,1,48,61lowB6F10Nacc1-p.A434GRKPLDSRVLHAVKYYCQNFGP NFK154,160,3,4,58,60highB6F10Oaz1-p.R24GREKEGDKGSATLHASRTMPLLSQH10,1,21,2,2,2lowB6F10Polr2b-p.I903SRRYTKRDCSTFLRTS ETGSVDQVM17,19,1,3,51,6lowB6F10Rbbp6-p.P1232HRETGKKIGNAENASTTKEHSEKLE0,10,13,1,6,7lowB6F10Rpl12-p.I82NSALINKALKEPPR DRKKQKNIKHS17,3,40,4,62,66highB6F10Sbno1-p.K720NTREAKKARNVGGLTGSSSSDDSGSE22,2,3,3,3,8lowB6F10Spen-p.E1113DAGLGELTHG SVDTQETQPKKAIPSK142,19,2,3,47,93lowEMT6Aatf-p.T442IDFYHQLLRELIERKISSLDPNDQVAMGRQ26,32,42,56,57,78highEMT6Atp6v1a-p.Q441EDKKLAERKHFPSVNWLISYSKYMR10,13,14,32,34,9lowEMT6Atp6v1f-p.N99DSKEHPYDAAKDSILRRAKGMFTAE12,12,13,4,9,9lowEMT6Cdc42-p.D63EPYTLGLFDTAGQEEYDRLRPLSYPQTDV0,0,1,1,1,3highEMT6Cltc-p.Y417FPGQTSPLLQFFGILLDQGQLNKYE40,53,63,66,77,81highEMT6Dnmt1-p.D576GFVVSQVESYDEAKDGDETPIFLSPCMRAL1,2,2,2,7,8highEMT6Dync1h1-p.N2529STVPLPTAPSVPIIDYEVSISGEWS13,16,16,20,22,29lowEMT6Hjurp-p.R157GREASGDPRQRQPAVPGNTLETDLR10,2,2,3,4,7highEMT6Iars-p.V178GKQLYDKGLVYRGVKGMPFSTACGTPLSNF11,16,16,7,7,8highEMT6Med29-p.G152DPTATKPDAVQPDSLPYPQYLAVIK10,10,12,4,6,7highEMT6Msn-p.F84LAQDVRKESPLLFKFRAKLYPEDVS0,1,1,1,3,4lowEMT6Nxf1-p.S598YDNNWDYTRYAQAFTLLKAKGEIPE24,24,28,48,57,61lowEMT6Pdhb-p.H138QSAGLQPVPIVFRGPNGASAGVAAQ0,1,2,3,3,4highEMT6Phip-p.N671DRRSGEAGVSNASRVDRGSVSSTSEVHS0,1,1,4,4,6lowEMT6Rasl2-9-p.F162LDISARSNYNFEKPFLWLARKLIGDPNLEF0,10,1,2,3,6highEMT6Sf3b1-p.H886RMETIEKIMGNLGAADIDRKLEEQL11,13,14,17,18,24lowEMT6Ssrp1-p.K623EKAMKEYEGGRGDSSERDKSKKKKKVKAKM0,1,3,3,6,7lowEMT6Stt3b-p.G804AKQKYLSKKTTKRKRAYVKNKLVFKKGKKT0,0,0,2,2,2lowEMT6Yae1d1-p.E151KDMDLCHVVPSDQAKGDSNAEINTDSDS10,13,17,3,9,9lowLLC1Chd8-p.K445NPASCAPHTAGNTGMEENRRLEHQKK0,19,1,1,1,4lowLLC1Ddb1-p.I959FRDFNPNWMSAVEFLDDDNFLGAEN12,1,20,3,7,9highLLC1Eif5b-p.S815TKDPRTFVTLVPTSAHTGDGMGSLI108,136,19,43,51,9lowLLC1Ewsr1-p.T97AYSQPVQGYGTGAYDSTTATVTTTQ11,1,2,36,3,8highLLC1Mllt3-p.M352TSETSEKKKSTLPPFDDIVDPNDSD1,21,27,2,4,8highLLC1Myh9-p.W26CDKYLYVDKNFINNPLAQADCAAKK14,16,56,61,72,78highLLC1Prpf8-p.K1955NKVILNPDKTTVTEPHHIWPTLTDE1,1,3,7,7,8highLLC1Psmb6-p.A60GDSRTTTGSYIGNRVTDKLTPIHDH0,1,2,3,3,7lowLLC1Rlim-p.I312VSDTGSNSESSGSGQRPPTVVLDLQ10,11,1,3,3,7lowLLC1Rrs1-p.G314WSQKGKRKWGRQGPSGKRKGGPPGQ0,2,4,4,6,6highLLC1Tns3-p.L1051MVNSHGTSPAPGTPLMTTGAADNGFLPH1,2,2,3,4,8high.

[0082] As a result, the TCR(+)BCR(high) vaccine elicited a stronger IFNγ signal than the TCR(+)BCR(low) vaccine (Fig. 3), suggesting that B cell reactivity to T cell-immunogenic neoantigens can induce a more robust immune response. A similar pattern was observed when the same analysis was repeated considering only mutation scores (Fig. 6). These results indicate that neoantigens that stimulate not only B cells but also T cells can induce a stronger immune response than neoantigens that do not bind to B cells.

[0083]

[0084] Example 3: Analysis of human clinical trial data

[0085] To determine the impact of neoantigen vaccines on B cell reactivity in a clinical setting, we collected data from 12 clinical trials in which 125 patients were administered 1,766 vaccines in the form of mRNA constructs, synthetic long peptides, or peptide-loaded DCs (Table 4).

[0086] 산산발하연도전프 수타하니다 수안 타자백신 타자HLA 강형형PMIDRojas202316232PDACmRNAo37165196Awad20229176NSCLCPeptidex36027916Sahin201713124SKCMmRNAx28678784 Ott_20172017698SKCMPeptideo28678778Keskin20198103GBMPeptidex30568305Chen2021674PDACPeptideo34484187Fang2 02011163SKCM,COAD,NSCLC,PDAC,OV,BTCPeptideo32439700Cai2021660HCCPeptidex34903219Ott_2020202034553SKCM,NS CLC,BLCAPeptidex33064988Hu2021227SKCMPeptidex33479501Cafri2020476GImRNAx33016924Peng20221080HCCDendritic cello35476700Total 1251766

[0087] Each patient was treated with 5–30 vaccines producing 14–52-mer peptides. All potential neoepitopes from each vaccine were analyzed using the DeepNeo-TCR and B-cell epitope prediction model (DeepNeo-BCR) to identify TCR(+) and BCR(+) epitopes (Figure 4A). When categorized by T-cell response measured by IFNγ secretion, IFNγ(+) vaccines harbored a significantly higher proportion of TCR(+) epitopes (Figure 4B). This suggests that considering only MHC binding in neoantigen design may not guarantee a robust T-cell response. Crucially, vaccines containing more TCR(+)BCR(+) epitopes were more likely to induce IFNγ release than vaccines containing fewer TCR(+)BCR(+) epitopes (Fig. 4C), suggesting that B cell-reactive neoepitopes also play an important role in promoting T cell responses. For example, vaccines containing ≥20% TCR(+)BCR(+) epitopes were 1.75 times more likely to stimulate IFNγ than vaccines containing fewer of these neoepitopes (P=3.96x10 -3 ) (Fig. 4D, left). This odds ratio was further increased in vaccines with more than 30% TCR(+)BCR(+) epitopes (OR = 2.41, P = 7.34x10 -5 ) (Fig. 4D, right). Activated B cells can potentially interact with CD4+ T cells that recognize other antigens or other epitopes of the same antigen. Therefore, B-cell epitopes could enhance not only the allogeneic vaccine containing that epitope, but also other vaccines simultaneously introduced into the same patient with different epitope compositions. To test this possibility, we assessed each patient's overall immune response to vaccination, measured as the proportion of IFNγ(+) vaccines relative to overall B-cell reactivity to all delivered neoantigens.

[0088] First, we determined the B-cell reactivity of the vaccine at the patient level using the maximum B-cell epitope prediction score among all neoepitopes per patient. This index showed a strong positive correlation with the immune response at the patient level (R=0.32; P=1.9x10 -4 ) (Fig. 4E). For example, the patient classified as 'Sahin P04' was vaccinated with 'vac1', a vaccine with a high B-cell epitope prediction score, and showed IFNγ(+) responses to 8 out of 10 vaccines (Fig. 4F, top). 'Chen P04' responded to 3 out of 13 vaccines that lacked BCR(+) epitopes in all vaccines (Fig. 4F, bottom). In contrast, the DeepNeo-TCR score did not show this correlation with patient-level immune responses (Fig. 7A). This pattern was also not found by other B-cell epitope prediction tools, LBtope, LBEEP, IgPred, and BepiPred3.0 (Fig. 8).

[0089] Next, the percentage of BCR(+) epitopes for all vaccines per patient was used as a measure of total B cell reactivity to the vaccine. This measure also showed a positive correlation with the proportion of IFNγ(+) vaccines (R = 0.25; P = 9.6x10 -3 )(Fig. 7B). For example, when an IFNγ(+) vaccine of 70% or more is considered a sufficient immune response, including 20% ​​or more BCR(+) epitopes increases the probability of inducing a patient-level response to vaccination by 4.72 times (P =4.11x10 -3 ; Figure 4G). When the same analysis was repeated with other B cell epitope prediction tools, no contribution of B cell neoepitopes to the patient's vaccine response was observed (Figure 9).

[0090] In summary, these results suggest that patients vaccinated with a vaccine formulation favorable for B cell reactivity are more likely to exhibit robust T cell responses, and that the B cell epitope prediction model can reliably predict B cell reactivity when considering the overall effects of all vaccines administered, not just individual neoantigens.

[0091]

[0092] Example 4: Tumor Tissue Data Analysis

[0093] Next, we investigated whether B-cell neoepitopes could influence the BCR repertoire within tumors. To this end, we analyzed the exomes and matched transcriptomes of 4,435 TCGA samples. We measured the B-cell neoepitope load by applying a B-cell epitope prediction model to the somatic mutations in each sample. The BCR repertoire characteristics were calculated in terms of clonality and entropy. For comparison, we used DeepNeo-TCR to calculate the T-cell neoepitope load and used transcriptome data to determine the clonality and entropy of the T-cell receptor (TCR) repertoire.

[0094] First, it was observed that individual tumors with higher BCR(+) epitope load exhibited higher BCR clonality and lower BCR entropy (Fig. 10). Then, each sample was classified into its corresponding tumor type, and the average value for each tumor type was derived. As a result, the positive correlation between BCR(+) epitope load and BCR clonality and the negative correlation between BCR(+) epitope load and BCR entropy were more pronounced (Fig. 5A). Furthermore, these correlations were most pronounced between BCR(+) epitope load and BCR properties. BCR clonality and entropy were less correlated with TCR(+) epitope load than with BCR(+) epitope load (Fig. 5B). Similarly, TCR clonality was more weakly correlated with BCR(+) epitope load compared to TCR(+) epitope load (Fig. 5B). However, unlike BCR entropy, TCR entropy did not show a negative correlation with TCR(+) epitope load. These results demonstrate that the B-cell epitope prediction model can identify properties of B-cell neoepitopes even within the local tumor environment.

[0095] The above-described method may be provided as a computer program stored on a computer-readable recording medium for execution on a computer. The medium may be one that continuously stores a computer-executable program or one that temporarily stores it for execution or download. In addition, the medium may be various recording means or storage means in the form of a single or multiple hardware combinations, and is not limited to a medium directly connected to a computer system, but may also be distributed over a network. Examples of the medium may include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and those configured to store program instructions, including ROM, RAM, and flash memory. In addition, examples of other media may include recording or storage media managed by app stores that distribute applications, sites that supply or distribute various software, servers, etc.

[0096] The methods, operations, or techniques described above may be implemented using various means. For example, these techniques may be implemented using hardware, firmware, software, or a combination thereof. Whether these methods are implemented as hardware or software depends on the specific application and the design requirements imposed on the overall system. Those skilled in the art may implement the described functions in various ways for each specific application, but such implementations should not be construed as deviating from the scope of the present disclosure.

[0097] In a hardware implementation, the processing units used to perform the techniques may be implemented within one or more ASICs, DSPs, digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, electronic devices, other electronic units designed to perform the functions described herein, a computer, or a combination thereof.

[0098] When implemented in software, the methods described above may be stored on or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media, including any medium that facilitates transfer of a computer program from one place to another. Storage media may be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium.

[0099] Although the embodiments described above have been described with limited drawings, those skilled in the art will appreciate that various technical modifications and variations can be applied based on the above. For example, appropriate results can still be achieved even if the described techniques are performed in a different order than described, and / or the described components are combined or combined in a different manner than described, or are replaced or substituted with other components or equivalents.

[0100] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims described below.

Claims

1. A step of receiving peptide sequences extracted from cancer tissue and allelic sequence data of a B cell receptor; A step of calculating the selective binding strength between amino acids of the above peptide sequence and the allelic sequence of the B cell receptor; and A step of inputting the selective binding strength between the above amino acids into a learned neural network model to calculate a B cell epitope prediction score; A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

2. In paragraph 1, The step of calculating the selective binding strength between amino acids of the above peptide sequence and the allelic sequence of the B cell receptor is as follows: A step of generating a matrix including Cα-Cα distance information between amino acids of the peptide sequence and the B cell receptor allele sequence and atomic contact information between amino acid residues, A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

3. In paragraph 2, The above matrix is, Contains somatic hypermutation (SHM) frequency data per residue of the B cell receptor allele superfamily. A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

4. In paragraph 3, The cell values ​​constituting the above matrix are defined by the following equation 1, [Formula 1] Here, i and j represent the positions of the immunoglobulin heavy chain variant (IGHV) and peptide sequence of the B cell receptor, and Ⅰ ij is IGHV i Wow peptide j Indicating selective binding interactions between, A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

5. In paragraph 1, The above neural network model is a CNN (Convolutional Neural Network) model, The CNN model includes a convolutional layer that detects spatial interactions between the immunoglobulin heavy chain variant (IGHV) of the B cell receptor and the peptide sequence. A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

6. In paragraph 5, The above CNN model consists of multiple convolutional layers, fully connected layers, and sigmoid layers. A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

7. In paragraph 5, The kernel of the above CNN model includes the entire amino acid sequence of the immunoglobulin heavy chain variant (IGHV) of the B cell receptor. A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

8. In paragraph 5, The above convolutional layer extracts features of contact and distal interactions using the Rectified Linear Unit (ReLU) function without pooling. A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

9. In paragraph 5, The above neural network model is an ensemble model that aggregates multiple CNN models. A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

10. In paragraph 1, The step of calculating the above B cell epitope prediction score is: A method for designing a customized cancer vaccine using a B cell reactive neoantigen, comprising a step of applying linear regression to the output value of the above neural network model.

11. In paragraph 1, A neoantigen whose calculated B cell epitope prediction score is higher than the average B cell epitope prediction score is judged to have a high ability to induce an immune response. A method for designing personalized cancer vaccines using B cell-reactive neoantigens.

12. A computer program stored on a computer-readable recording medium for executing the method of any one of claims 1 to 10 in combination with hardware.

13. A computer-readable recording medium having recorded thereon a program for executing the method of any one of claims 1 to 10 on a computer.

Images