Method and System for Analyzing Receptor Interactions

Abstract

A computational framework for high-throughput mapping, validation, and prediction of receptor sequence interactions is described.

Description

Technology Field

[0001] Cross-reference regarding related applications

[0002] This application claims the benefit of priority of U.S. Provisional Application No. 63 / 013,480, filed April 21, 2020; U.S. Provisional Application No. 63 / 090,498, filed October 12, 2020; and U.S. Provisional Application No. 63 / 111,395, filed November 9, 2020. The contents of these previously filed applications are incorporated herein by reference in their entirety. Background Technology

[0003] T cell antigen specificity, mediated through T cell receptors (TCRs), is a characteristic of cellular immunity. TCRs are heterodimeric proteins found on the surface of T cells and are generally composed of α- and β-chains. The TCR α- and β-chain genes consist of discrete V, D (β-chain alone), and J segments linked by somatic recombination during T cell development. These genetic rearrangements generate a highly diverse TCR repertoire (estimated to range from 10¹⁵ to 10⁶¹¹ receptors in humans) to ensure efficient control of viral infections and other pathogen-induced diseases. TCR diversity is primarily manifested in complementarity determining domain (CDR) loops (CDR1, CDR2, and CDR3), which bind to peptides presented by major histocompatibility complex (MHC) proteins, directly determining the specificity of T cell pMHC binding.

[0004] Although the factors underlying TCR-pMHC recognition are not fully understood, recent studies have shown that T cells binding to specific pMHCs share common TCR sequence features, and that, when selected, it is possible to predict the specific binding probability of unseen TCR sequences based on learned TCR sequence features. However, these studies were limited by the volume and diversity of training data generated by traditional single-multimer classified or antigen re-exposure analyses. Further understanding of TCR-pMHC specific binding requires innovation in both computational and experimental methods. 10x Genomics recently released a dataset generated from a highly multiplexed pooled dextramer binding immune profiling platform that couples feature-barcoded dextramers with single-cell TCR sequencing. This approach enables the generation of high-dimensional pMHC-specific binding data at the single-cell level using paired T cell α- and β-chain sequences, whereas other large-scale pooled multimer approaches only estimate the composition of pMHC-specific binding T cells.

[0005] As with any other high-throughput technology, highly multiplexed dextramer binding data are often associated with low signal-to-noise ratios. This makes it bioinformatically difficult to reliably identify TCR-pMHC binding events using such large binding datasets. We observed unexpectedly high cross-HLA and cross-pMHC associations from binding events provided by 10x Genomics (see Fig. 11a). These low signal-to-noise datasets require more sophisticated computational normalization methods to distinguish true TCR-pMHC binding events from a non-specific background.

[0006] As next-generation screening technologies increase the volume of available TCR-pMHC binding data, state-of-the-art functional classifiers for computationally validating and subsequently predicting TCR-pMHC-specific recognition have become more feasible. While the results of early TCR-pMHC binding classifiers are encouraging, they were trained only when using CDR loop sequences and thus could not learn full complex sequence patterns from full-length TCR sequences, resulting in suboptimal prediction accuracy for a wide variety of pMHC binding TCRs. Leveraging the ability of deep learning algorithms to learn complex patterns, several deep learning frameworks have recently been proposed to identify binding patterns in large, highly complex TCR sequence datasets.

[0007] In this study, a computational framework for mapping, computationally verifying, and predicting TCR-pMHC-specific recognition using highly multiplexed dextramer binding data is described.

[0008] A method is disclosed comprising the steps of: receiving single-cell sequencing data including single-cell sequence data, dextramer sequence data, and single-cell T cell receptor (TCR) sequence data; filtering data associated with low-quality cells from the dextramer sequence data based on the single-cell sequence data; adjusting the dextramer sequence data based on background noise measurements; filtering data based on the presence or absence of α-chains or β-chains from the dextramer sequence data based on the single-cell TCR data; and identifying data remaining in the normalized filtered dextramer sequence data as associated with reliable TCR-pMHC binding events.

[0009] A step of receiving single-cell sequence data, dextramer sequence data, and single-cell T-cell receptor (TCR) sequence data; a step of determining the number of genes for each cell indicated in the dextramer sequence data based on the single-cell sequence data; a step of removing from the dextramer sequence data data data associated with cells having a number of genes outside a gene threshold range; a step of determining the fraction of mitochondrial gene expression for each cell indicated in the dextramer sequence data based on the single-cell sequence data; a step of removing from the dextramer sequence data data data associated with cells having a fraction of mitochondrial gene expression exceeding a gene expression threshold; a step of determining classified dextramer sequence data and unclassified dextramer sequence data based on the dextramer sequence data, wherein the classified dextramer sequence data includes classified test dextramer sequence data and negative control dextramer sequence data, and the unclassified dextramer sequence data includes unclassified test dextramer sequence data; A step of determining the maximum negative control dextramer signal based on the negative control dextramer sequence data for each cell indicated in the dextramer sequence data; a step of determining the maximum classified dextramer signal based on the classified test dextramer sequence data for each cell indicated in the dextramer sequence data; a step of determining the maximum unclassified dextramer signal based on the unclassified test dextramer sequence data for each cell indicated in the dextramer sequence data; a step of estimating dextramer binding background noise based on the maximum negative control dextramer signal; and a step of estimating dextramer classification gate efficiency based on the maximum classified dextramer signal and the maximum unclassified dextramer signal.A step of determining a measurement of background noise based on the dextramer binding background noise and the dextramer classification gate efficiency; a step of subtracting a measurement of background noise from the dextramer signal associated with each cell for each cell indicated in the dextramer sequence data; a step of performing cell-specific normalization on the dextramer signal associated with each cell for each cell indicated in the dextramer sequence data; a step of performing pMHC-specific normalization for each cell indicated in the dextramer sequence data; a step of determining the presence or absence of at least one α-chain and at least one β-chain for each cell indicated in the dextramer sequence data based on the single-cell TCR sequence data; and a step of removing data associated with cells having only an α-chain, only a β-chain, or multiple α- or β-chains from the normalized dextramer sequence data based on the presence or absence of at least one α-chain and at least one β-chain. A method is disclosed comprising the step of identifying data remaining in normalized dextramer sequence data as associated with reliable TCR-pMHC binding events.;

[0010] A method is disclosed comprising: a step of identifying a plurality of TCR-pMHC binding events by performing normalization of TCR-pMHC binding specificity data for dextramer sequence data; a step of determining a training dataset containing a plurality of TCR sequences based on the normalized dextramer sequence data, wherein each TCR sequence is associated with binding affinity; a step of determining a plurality of feature parts for a prediction model based on the plurality of TCR sequences; a step of training a prediction model according to the plurality of feature parts based on a first part of the training dataset; a step of testing a prediction model based on a second part of the training dataset; and a step of outputting the prediction model based on the test.

[0011] A method is disclosed comprising the steps of: presenting an unknown TCR sequence to a trained predictive model, wherein the trained predictive model is trained based on a training data set derived according to the disclosed method; and predicting binding affinity by the trained predictive model.

[0012] A step of receiving single-cell sequence data, dextramer sequence data, and single-cell T-cell receptor (TCR) sequence data; a step of determining the number of genes for each cell indicated in the dextramer sequence data based on the single-cell sequence data; a step of removing data associated with cells having a number of genes outside a gene threshold range from the dextramer sequence data; a step of determining a fraction of mitochondrial gene expression for each cell indicated in the dextramer sequence data based on the single-cell sequence data; a step of removing data associated with cells having a fraction of mitochondrial gene expression exceeding a gene expression threshold from the dextramer sequence data; a step of determining classified dextramer sequence data based on the dextramer sequence data, wherein the classified dextramer sequence data includes classified test dextramer sequence data and negative control dextramer sequence data; a step of determining a maximum negative control dextramer signal for each cell indicated in the dextramer sequence data based on the negative control dextramer sequence data; and a step of determining each indicated in the dextramer sequence data For a cell, a step of determining the maximum classified dextramer signal based on the classified test dextramer sequence data; a step of estimating dextramer binding background noise based on the maximum negative control dextramer signal and the maximum classified dextramer signal; a step of determining the presence or absence of at least one α-chain and at least one β-chain based on the single-cell TCR sequence data for each cell indicated in the dextramer sequence data; a step of removing data associated with cells having only an α-chain, only a β-chain, or multiple α- or β-chains from the dextramer sequence data based on the presence or absence of the at least one α-chain and the at least one β-chain.For each dextramer binding to a given cell indicated in the dextramer sequence data, a step of determining the ratio of the intracellular dextramer signal to the sum of all dextramers binding to the cell (a measure of dextramer binding specificity for the cell); for each dextramer binding to a given TCR clone type indicated in the dextramer sequence data, a step of determining the fraction of T cells within the clone that bind to a specific dextramer (a measure of dextramer binding specificity for the clone type to which the cell belongs); for each dextramer binding to a given cell indicated in the dextramer sequence data, a step of determining the corrected dextramer signal associated with each dextramer binding to the cell based on the measure of dextramer binding specificity for the cell and the measure of dextramer binding specificity for the clone type to which the cell belongs; for each cell indicated in the dextramer sequence data, a step of performing cell-specific normalization on the dextramer signal associated with each cell; for each cell indicated in the dextramer sequence data, pMHC-specific normalization A method is disclosed comprising the steps of performing, and identifying data remaining in normalized dextramer sequence data as associated with reliable TCR-pMHC binding events based on a threshold.

[0013] A device configured to perform any one of the disclosed methods is disclosed.

[0014] A computer-readable medium having an embodiment of a processor-executable instruction configured to enable a device to perform any one of the disclosed methods is disclosed.

[0015] Additional benefits of the disclosed methods and compositions will be presented in part from the following description, understood in part from this description, or learned through the practice of the disclosed methods and compositions. The benefits of the disclosed methods and compositions will be realized and achieved by the elements and combinations specifically mentioned in the appended claims. It should be understood that the following general description and the following detailed description are all illustrative and for illustrative purposes only and are not limiting to the claimed invention. Brief explanation of the drawing

[0016] The accompanying drawings included in and constituting part of this specification illustrate multiple embodiments of the disclosed method and composition and serve to explain the principles of the method and composition disclosed together with this specification. Figure 1 illustrates an exemplary operating environment. Figure 2 illustrates an experimental approach for generating multi-omics high-throughput TCR-pMHC binding data: PBMC T cells derived from healthy human donors were labeled to sort CD8+ cells. The sorted CD8+ T cells were stained with a pool of 50 dCODE dextramer antibodies. Dextramer-positive CD8+ T cells were sorted by flow cytometry and individually captured as inputs for the production of 10x Genomics single-cell sequencing libraries. Three libraries were generated for gene expression, cell surface protein / dCODE expression, and paired TCR sequences for each CD8+ T cell. Figure 3 illustrates an exemplary method. Figure 4 illustrates an exemplary method. Figure 5 illustrates an exemplary method. Figures 6a and 6b are ICON( I ntegrative CO ntext-specific N Illustrate an example of a workflow system (normalization).a. From top left to bottom left: I. Distribution of dCODE dextramer raw expression in UMI (Unique Molecular Identifier). Maximum dCODE dextramer expression within UMI in CD8+ cells derived from Dex_sorted (maximum UMI of test dextramer from dextramer-sorted CD8+ T cells), NC_dex (maximum UMI of negative control dextramer from dextramer-sorted CD8+ T cells), and Dex_unsorted (maximum UMI of test dextramer from dextramer-stained but unsorted CD8+ cells). II. Filtering out low-quality cells based on single-cell RNA-seq. Each dot is a T cell. Red dots are unhealthy cells. III. Dextramer binding background noise based on dCODE dextramer expression data ( ) and dextramer classification gate efficiency Estimating. IIII. Adjust for background noise by subtracting. V. Normalization of background-subtracted dextramer expression by cell and pMHC. VI. Selection of cells with single-pair TCR αβ chains. VII. Distribution of normalized dextramer expression. UMI*: Normalized UMI. Refer to Methods for details. b. TCR-pMHC binding specificity of extended TCR clones. The 50 largest TCR clones from Donor 1 are plotted along with their binding specificity and match. Circles indicate that members of at least one clone type were classified as specific to a particular pMHC. Circle size represents the total clone size within the donor. Circle color represents the proportion of cells within the clone type that bind to the dextramer ('binding match'). Left panel: The 50 largest clones identified by 10x Genomics using the full cutoff. Right panel: The 50 largest clones from the pMHC repertoire containing the 50 largest clones identified by 10x Genomics from Donor 1. Figures 7a to 7e show the pMHC binding patterns of 10x Genomics dextramer binding data. a. Network of identified pMHC-specific binding T cell repertoires. Each intersection represents a pie chart of the pMHC repertoire and the number of unique paired TCRs from each donor binding to that pMHC. Donor 1 is gray, Donor 2 is red, and Donor 4 is yellow. The size of the intersection indicates the total number of T cells binding to the corresponding pMHC. Each edge represents a unique TCR(s) shared by two pMHCs. The thickness of the edge indicates the number of shared unique TCR(s). b. Most of the identified binders interact with 7 pMHCs. c. Venn diagram of unique paired TCRs identified from donor 1, donor 2, and donor 3. d. Composition of uniquely paired TCR αβ chains. According to TCRB standards, 1 to 1 means that one unique TCR β-chain is paired with one unique TCR α-chain; 1 to >= 2 & binding to the same pMHC means that uniquely paired TCRs with shared β-chains recognize the same pMHC with different α-chains; and 1 to >= 2 & binding to >= 2 pMHCs means that uniquely paired TCRs with shared β-chains recognize different pMHCs with different α-chains. According to TCRA standards, 1 to 1 means that one unique TCR β-chain is paired with one unique TCR α-chain; 1 to >= 2 & binding to the same pMHC means that unique paired TCRs with shared α-chains but different β-chains recognize the same pMHC; 1 to >= 2 & binding to >= 2 pMHC means that unique paired TCRs with shared α-chains but different β-chains recognize different pMHCs. e.TCR-pMHC binding specificity and TCR cross-HLA recognition. Left, pie chart of T cells binding to one pMHC or at least two pMHCs. Right, pie chart of T cells: HLA type-matched binding, supertype-matched binding, or cross-matched binding. Figures 8a through 8d illustrate the Synthetic Neural Network (CNN)-based classification of TCR-pMHC-binding TCRs. a. CNN-based TCR sequence classification framework. Left panel, V and J segments (derived from alpha and beta) were converted into embedding vectors. Trainable embeddings for the amino acids constituting the CDR3 alpha or beta sequences were used, and a 1D CNN was applied to the embeddings. Then, all embeddings were concatenated together and fed through the concatenated layers. Subsequently, sequence class probabilities were output using a SoftMax layer. Right panel, a toy example illustrates the input and output from the deep learning sequence classifier. Refer to the Methods section for details. b. ROC curves for the CNN-based classifier by binary mode using 11 selected paired TCR-pMHC binding repertoires. The binders are unique TCRs that bind to a specific pMHC, and the non-binders are unique TCRs that bind to the other 10 pMHCs. Paired α & β TCR sequences were used as input data. c. Comparison of classification power between CNN-based and distance-based binary classifiers with identical definitions of binders and non-binders as described in b. Paired α & β TCR sequences were used as input data (Methods). d. Correlation of pMHC repertoire diversity measured by Shannon entropy and prediction performance between CNN-based classifiers and distance-based classifiers. AUC = CNN-based AUC - Distance-based AUC. Figures 9a-9e illustrate the CNN-based classification of the top 7 pMHC binding repertoires identified from the 10x Genomics dataset. a.ROC curves for a CNN-based classifier in binary mode using 7 pMHC binding repertoires identified from the 10x Genomics high-throughput dataset. The binders are native TCRs that bind to a specific pMHC, and the non-binders are native TCRs that bind to the other 6 pMHCs. Paired α & β TCR sequences were used as input data. b. ROC curve of prediction results from a CNN-based classifier using an independent test dataset from VDJdb: T cells binding to another set of A*02:01_GILGFVFTL_Flu-MP_Influenza, A*02:01_ELAGIGILTV_MART-1_Cancer, A*02:01_GLCTLVAML_BMLF1_EBV and A*11:01_AVFDRKSDAK_EBNA-3B_EBV and MART-1 (REGN_ A*02:01_ELAGIGILTV_MART-1_Cancer) binders from an in-house independent experiment (Methods). The module was trained on a pMHC repertoire identified from 10x Genomics data for prediction. c. Comparison of classification performance using TCRα alone, TCRβ alone, or paired TCRα & β chains as sequence inputs. d . Use of T cell V and J gene segments for T cells binding to these 7 pMHCs. Less than 5% of gene segments were combined and indicated in gray. e. CDR3 motifs of the 10 most predictive paired TCRs from the 7 pMHC repertoires. Figures 10a-10e show the immunophenotype of pMHC-bound CD8+ T cells. a. Classification of pMHC-bound cells. Clusters were visualized using UMAP, and cell types were indicated in different colors. b. Heatmap of gene or protein expression of cell type marker genes for annotating CD8+ T cell subsets. C.pMHC binding environment based on T cell immune subtypes. Bars represent the number of pMHC-bound T cells on a log2 scale. d. Extended clonal types are abundant in the unprocessed section. Each dot represents a unique TCR clone. e. HLA-matched and HLA-mismatched binding ratios in untreated and non-untreated combined T cells. Tpm: Peripheral memory cells; Tcm: Central memory cells; Tem: Effector memory cells; Temra: Terminally differentiated effector memory cells; Others: Marker expression of CD43 lo KLRG1 hi Other memory cells containing CD127. Figures 11a-11b show the TCR-pMHC binding specificity of extended clonal types from binding events of 10x Genomics identified from each donor. The 50 largest clonal types are plotted along with their binding specificity and concordance. a. Circles indicate that members of at least one clonal type were classified as specific to a particular pMHC. Circle size represents the total clonal type size within the donor. Circle color represents the proportion of cells within the clonal type binding to the dextramer ('binding concordance'). b. Scatter plot of cell classification results for the re-evaluation of CD8+ T cell dextramer binding from 10x Genomics donors 3 and 4 (Methods). Figures 12a–12f are examples of estimating the background of 10x Genomics high-throughput data and tuning the dextramer binding signal. Dex_sorted (maximum UMI of test dextramer from dextramer-sorted CD8+ T cells), NC_dex (maximum UMI of negative control dextramer from dextramer-sorted CD8+ T cells), and Dex_unsorted (maximum UMI of test dextramer in dextramer-stained but unsorted control CD8+ cells). a.Scatter plot of the number of detected genes relative to the percentage of mitochondrial gene expression using single-cell RNA data. Each dot represents a cell. Dots marked in red are dead cells or duplexes. b. Distribution of dextramer expression data before and after the ICON process. C & d. Estimation of dextramer classification efficiency. c. Cumulative distribution of dextramer UMI. Each point is a data point of a unique dextramer UMI. d. p-value distribution of the KS test using a single dextramer UMI data point as a sliding window (Dex_sorted vs. Dex_unsorted). The gray dashed line is the threshold value of dextramer classification efficiency. e. Dex_sorted scatter plot before (x-axis) and after (y-axis) background deduction for each donor. f. Density distribution. : Log-rank of each intracellular dextramer signal (Method). The blue dashed line is for the threshold of pMHC-specific binding. Figures 13a–13c show the binding specificity of extended clonal types identified in this study from three donors. The 50 largest T cell clones are plotted along with their binding specificity and concordance. Circle size indicates the T cell clone size. Circle color indicates the proportion of cells within the clone binding to dextramer and binding concordance. Figures 14a and 14b show ROC curves for a distance-based classifier using a selected pMHC binding repertoire. b. Shannon entropy scores for selected pMHC binding repertoires. Figures 15a-15c show the characterization of the top 7 pMHC-binding T cell repertoires. a. Pie chart of the ratio of HLA type-matched, supertype-matched, and mismatched combined T cells. b.Power law distribution of unique T cell clone sizes of the top 7 pMHC binding repertoires. Lowess Smoothing was used for fitting. c. Simpson diversity index and TCRB generation probability of the TCR-pMHC repertoire. The R package Vegan was used to calculate the Simpson diversity index. The generation probability of TCRB CDR3 amino acid sequences for binders specific to each pMHC was calculated using OLGA. Then, as described by Sethna et al., the fraction of the repertoire specific to each pMHC (indicated by red triangles) is obtained as the sum of the generation probabilities for each corresponding CDR3 sequence. The result is defined in the sense that the net fraction of these pMHC-specific TCRs is the inverse of the number of independent TCR recombination events (10 8 ) being big (10 7 to 10 4 It shows the range, which means that any individual is likely to have these combined T cells in their T repertoire. Each point on the TCRB generation probability plot represents a unique T cell clone, and the color bar represents the T cell clone size. Figures 16a-16c show the classification of TCR-pMHC-binding TCRs. a. Distance-based distribution of pMHC binders and non-binders using α-chain alone, β-chain alone, and paired αβ chains. b. ROC curves for a distance-based classifier using the top 7 pMHC binding repertoires identified from the 10x Genomics high-throughput dataset. Paired α & β TCR sequences were used as input data. c. Comparison of classification power of CNN-based and distance-based classifiers. Figures 17a and 17b show the CDR3 motifs of four pMHC binding repertoires from the overlay of VDJdb and the top seven pMHC repertoires identified from 10x Genomics high-throughput data. b. ROC curves for a CNN-based classifier in polynomial mode using 7 pMHC binding repertoires identified from the 10x Genomics high-throughput dataset. Paired α & β TCR sequences were used as input data. Figures 18a and 18b show examples of clusters of pMHC-coupled CD8+ cells using single-cell RNA-seq data. a. Based on the number of clusters. b. Overlaid with donor information. Figure 19 is a table containing information on the T cell donors used in the disclosed study. Figure 20 is a list of dCODE dextramer reagents used in the disclosed study and NetMHC peptide HLA allele binding prediction. Figure 21 is a summary table of pMHC-TCR binding events. Figure 22 shows the TCR-pMHC repertoire diversity and peptide characteristics. Figure 23 shows a summary of 11 pMHC repertoires collected from VDJdb and McPAS. Figure 24 shows the specificity of extended TCR clonal pMHCs in binders identified by 10x Genomics. The 50 largest TCR clones from donors 1 through 4 are plotted along with their binding specificity and match. Circles indicate that members of at least one clonal type were classified as specific to a particular pMHC. Circle size represents the total clonal type size within the donors. Circle color represents the proportion of cells within the clonal type that bind to the dextramer ('binding match'). Figures 25a–g illustrate the identification and characterization of pMHC-binding T cells from high-throughput pMHC binding data. (A) ICON (Integrated Text-Specific Normalization) workflow scheme. RT: Fraction of T cells within the clone binding to a specific dextramer; RC: Ratio of intracellular dextramer signals to the sum of all dextramers binding to the cell. (B) The pMHC binding environment network of ICON identified dextramer binders. Each intersection represents a pMHC repertoire and is displayed as a pie chart of the number of unique paired TCRs for each donor binding to the corresponding pMHC. The size of the intersection indicates the total number of unique TCRs binding to a given pMHC. Each edge represents a unique TCR(s) shared by two pMHCs. The thickness of the edge indicates the number of shared unique TCR(s). (C) Correlation between the results of flow classification for a single dextramer binding and the ICON-estimated relative abundance of pMHC-binding T cells. The number of dextramers for verification is 21. (D) Uniqueness and overlap of identified pMHC-binding TCRs among donors 1, 2, 3, 4, and V. (E) Most of the identified binders interact with 9 pMHCs. (F) Use of V and J gene segments for T cell binding to 9 pMHCs. Less than 5% of gene segments were combined and are shown in gray. (G) HLA type restricted and unrestricted binding. Figures 26a-d illustrate the processing of high-throughput data using ICON. A ) Scatter plot of the number of detected genes relative to the percentage of mitochondrial gene expression using single-cell RNA data. Each dot represents a cell. Red dots are dead cells or diplets. B ) 　 Distribution of dextramer signals in UMI from negative control and test dextramer. Sorted_nc: Negative control dextramer; sorted_dex: Test dextramer. (C ) Scatter plot of RT versus RC. RC is the ratio of intracellular dextramer signal to the sum of all dextramers binding to T cells. RT is the fraction of intraclonal T cells binding to a specific dextramer. D Hierarchical clusters of ) ICONs identified pMHC-binding T cells. Each row is a dextramer, and each column is a T cell. Figure 27 shows dextramer from donor V. + Illustrates pooled dextramer FACS gating for fluorescence activation sorting of T cells (FACS). Figures 28a-b show a single oligo-dextramer classification. ( A ) Representative gating for fluorescence-activated sorting (FACS) of dextramer-positive T cells. T cells were previously enriched from donor V peripheral blood mononuclear cells (PBMCs) and then stained with a single oligo-dextramer. The desired dextramer+ population for sorting was isolated using the following sequential gating strategy. ( B ) Scatter plot of single oligo-dextramer cell sorting results for each of the 21 test dextramers and 2 negative control dextramers. Figure 29 is a table showing a summary of pMHC-TCR binding event icons identified from high-throughput pMHC binding data. Figures 30a-b show the characterization of pMHC-binding T cells identified from a high-throughput dataset. A ) Strategic law distribution of intrinsic T cell clone sizes of the top 9 most abundant pMHC-binding T cell repertoires. B Shannon diversity scores of the top 9 pMHC repertoires. Figures 31a-c show the TCRAI models and performance on the gold-standard dataset. ( A ) 　αSchematic diagram of the TCRAI framework for a model receiving inputs of CDR3 on both the β and V chains, and V and J genes. The trained TCRAI model generates numerical fingerprints and predictions for a given TCR. B ) 　 ROC curves for TCRAI sorting performance using 8 selected open TCR-pMHC binding repertoires. Binders are native TCRs that bind to specific pMHCs, and non-binders are native TCRs that bind to different pMHCs. Paired α & β TCR sequences were used as input data. FPR: False positive rate; TPR: True positive rate. C ) Classification Performance Comparison. TCRAI was compared with the predictive classifiers NetTCR, TCRdist, and DeepTCR. Area under the ROC curve (AUC) scores for NetTCR and TCRdist were generated using the original classifiers with default parameters. AUC scores for DeepTCR (multinomial classifier) ​​were derived from a slightly modified and hyperparameter-optimized version of DeepTCR to compare with these binary classifiers NetTCR and TCRdist (Methods). For comparison, the binary mode of TCRAI was used. Figures 32a-c show the ROC performance of the TCR antigen-specific classifier (a and b). (c) Shows the ROC curve for TCRAI in polynomial mode using nine pMHC binding repertoires identified from a high-throughput dataset. Paired α and β TCR sequences were used as input data. FPR: False positive rate; TPR: True positive rate. Figure 33 is a table showing a comparison of TCR-antigen specificity classifiers. Figures 34a-d show the TCRAI performance on high-throughput datasets. ( AROC curves for TCRAI on the top 9 most abundant pMHC binding repertoires. Binders are native TCRs that bind to specific pMHCs, and non-binders are native TCRs that bind to different pMHCs. Paired α & β TCR sequences were used as input data. FPR: False positive rate; TPR: True positive rate. 　 ( B ) Comparison of classification performance using TCRα alone, TCRβ alone, or paired TCRα & β chains as sequence inputs. C ROC curves from independent tests of four nested pMHC repertoires between the selected public dataset and the high-throughput dataset. TCRAI was trained on pMHC repertoires identified from the high-throughput dataset and tested on the selected public dataset. D UMAP of both training (high-throughput data) and test ("gold standard" data) TCRAI fingerprints extracted from a high-throughput trained model. The left panel shows strong overlap between the A*02:01_ELAGIGILTV_MART-1_Cancer training and test sets, while weak overlap between the A*02:01_NLVPMVATV_pp65_CMV training and test datasets is shown in the right panel. Black circles highlight areas with almost no overlapping fingerprints of the combined TCR. Figure 35 shows the ROC curve for TCRAI in polynomial mode using nine pMHC binding repertoires identified from a high-throughput dataset. Paired α & β TCR sequences were used as input data. FPR: False positive rate; TPR: True positive rate. Figures 36a-b illustrate TCRAI fingerprint comparisons between models trained on different datasets. AAs a comparison of high-throughput and "gold standard" TCR fingerprints generated by a high-throughput data model trained for two cases not shown in Fig. 3d, it demonstrates good overlap of the binder in both cases. B The inference problem was performed in reverse, training a model with "gold-standard" data and computing fingerprints of "gold-standard" and high-throughput TCRs. For A*02:01_NLVPMVATV_pp65 / CMV, which exhibits poor cross-dataset performance, the model trained on "gold-standard" data containing TCRs from many donors separates large groups of combined TCRs. However, high-throughput combined TCRs originate primarily from single donors and contain only combined TCRs from small clusters that do not adequately represent the range of combined TCRs occurring in the broader population of the TCR space. Black circles highlight TCRs unique to high-throughput data. Figures 37a-g show the characterization of the TCR group. A Clustering high-confidence TCRAI fingerprints identified on a high-throughput dataset by a model trained to predict A*02:01_GILGFVFTL_Flu-MP_Influenza binders yields two TCR clusters: Cluster 0 (orange) and Cluster 1 (green). B ) 　 Dextramer signal (UMI) distribution of clusters 0 and 1. C ) Conserved CDR3 motifs and gene usage among these two Flu peptide-binding TCR clusters. For Cluster 0, gene usage for the 30 most common unique gene-using tetramers was shown so that core diversity could be seen in a single plot. D 3D structure of the Flu peptide-bound TCR-pMHC binding complex for Cluster 0 TCR (PDB 2VLJ) and Cluster 1 TCR (PDB 5JHD). Top panel, 4 of the Phe-5 ring. Only the non-peptide residues (pink β-chain, blue α-chain, green MHC) are shown. Bottom panel, comparison of peptide structures from Cluster 0 and Cluster 1 TCR-pMHC binding complexes. E Clustering of TCRAI fingerprints of high-confidence TCRs combined into A*02-01_GLCTLVAML_BMLF1_EBV from a high-throughput dataset. F ) Dextramer signal (UMI) distribution of EBV peptide binding clusters 0 to 2. G Use of the conserved CDR3 motif and gene among these three EBV peptide-binding TCR clusters. Figures 38a-f show the immunophenotypes of pMHC-bound CD8+ T cells. A ) 　 Classification of pMHC-bound cells. Clusters were visualized using UMAP, and cell types were indicated in different colors. 　 ( B ) 　 Heatmap of expression of CD8+ T cell type marker genes and proteins. *: CITE-seq. by T cell immune subtype. 　 ( C ) 　 Protein expression measured by the pMHC binding environment. Bars represent the number of pMHC-bound T cells on a log2 scale. 　 ( D ) 　 Extended clonal types are abundant in the untreated section. Each dot represents a unique TCR clone. E The pie chart describes the subpopulation of pMHC-binding CD8+ T cells. F ) 　 Ratio of HLA-matched and HLA-mismatched pairings in untreated and non-untreated paired T cells. Tpm: Peripheral memory cells; Tcm: Central memory cells; Tem: Effector memory cells; Temra: Terminally differentiated effector memory cells; Others: Marker expression CD43lo KLRG1 hi Other memory cells containing CD127. Figure 39 illustrates the importance of VJ genetic information. When comparing models trained using only the whole input or genetic input, the error in AUC is calculated by propagating the error in AUC for each model (whole or genetic), assuming there is no covariance between the results. The error in AUC for each model is the greater of the difference between the mean AUC for the best hyperparameters during MCCV and the mean AUC for the final models trained with these hyperparameters, or the standard deviation of the AUC during MCCV. △AUC 전체 - AUC 유전자 . Figures 40a-b show the characterization of the TCR group. A Dextramer signal distribution of all 5 TCR clusters identified for A*02-01_GLCTLVAML_BMLF1_EBV as illustrated in the fingerprint space of Fig. 4e. B ) Use of EBV peptide-binding TCR cluster 3 and 4 motifs and genes. FIG. 41 illustrates an exemplary operating environment. Fig. 42 illustrates an exemplary method. Fig. 43 illustrates an exemplary method. Fig. 44 illustrates an exemplary method. Fig. 45 illustrates an exemplary method. Figure 46 illustrates an exemplary method. Specific details for implementing the invention

[0017] The disclosed method and composition may be more easily understood by referring to the detailed description of specific embodiments below, the examples and drawings included therein, and the above and below descriptions thereof.

[0018] A. definition

[0019] It is understood that the disclosed methods and compositions are not limited to the specific methodologies, protocols, and reagents described, as they may vary. Furthermore, it should be understood that the terms used herein are merely for describing specific embodiments and are not intended to limit the scope of the method, which is defined solely by the appended claims.

[0020] It should be noted that, as used herein and in the appended claims, the singular forms (“a,” “an,” and “the”) include plural references unless explicitly otherwise indicated in the context. Thus, for example, a reference to “TCR” includes plural such TCRs, and a reference to “dextramer” is a reference to one or more dextramers and equivalents thereof known to those skilled in the art.

[0021] The terms “subject” or “donor” may refer to animals such as mammalian species (preferably human) or avian species (e.g., birds). More specifically, the subject or donor may be vertebrates, e.g., mice, primates, apes, or mammals such as humans. Animals include farm animals, sports animals, and pets. The subject or donor may be a healthy individual, an individual with symptoms or signs, or suspected of having a disease or predisposition to a disease, or an individual requiring treatment or suspected of requiring treatment. In some embodiments, the subject donor is a human, e.g., a human who has or is suspected of having cancer.

[0022] As used herein, the term “barcode” generally refers to a label that can be attached to a molecule (e.g., dextramer, cell) to convey information about the molecule. For example, a DNA barcode may be a polynucleotide sequence attached to each dextramer, and a common sequencing barcode may be a polynucleotide sequence attached during sequencing. This barcode can then be sequenced. If the same barcode exists on multiple sequences, it may provide information about the origin of the sequences. For example, the barcode may indicate that the sequence originated from a specific dextramer. The barcode may also indicate that the sequence originated from a specific cell / dextramer combination.

[0023] As used herein, the terms “sequencing” or “sequencer” refer to any one of a number of techniques used to determine the sequence of biomolecules, such as nucleic acids, for example, DNA or RNA. Exemplary sequencing methods include, but are not limited to, targeted sequencing, single-molecule real-time sequencing, exon sequencing, electron microscopy-based sequencing, panel sequencing, transistor-mediated sequencing, direct sequencing, random shotgun sequencing, Sanger dideoxy termination sequencing, whole-genome sequencing, sequencing by hybridization, thermal sequencing, divertet sequencing, cycle sequencing, single nucleotide extension sequencing, solid-phase sequencing, high-throughput sequencing, large-scale parallel signature sequencing, emulsion PCR, co-amplification-PCR at low denaturation temperature (COLD-PCR), multiplet PCR, sequencing by reversible dye terminators, paired term sequencing, short-term sequencing, exonuclease sequencing, sequencing by linkage, short-read sequencing, single-molecule sequencing, sequencing by synthesis, real-time sequencing, reverse terminator sequencing, and nanopore Sequencing, 454 sequencing, Solexa genome analyzer sequencing, SOLiD sequencing, MS-PET sequencing, and combinations thereof are included. In some embodiments, sequencing may be performed by a gene analyzer, such as a commercially available gene analyzer from, for example, Illumina or Applied Biosystems.

[0024] "Polynucleotide," "nucleic acid," "nucleic acid molecule," or "oligonucleotide" refers to a linear polymer of nucleosides (including deoxyribonucleosides, ribonucleosides, or analogs thereof) linked by internucleoside bonds. Generally, a polynucleotide contains at least three nucleosides. Oligonucleotides often range in size from a few monomeric units, for example, from three to four to several hundred monomeric units. Whenever a polynucleotide is represented by a sequence of letters such as "ATGCCTG," unless otherwise specified, the nucleotides are in 5'' order from left to right, and "A" is understood to represent adenosine, "C" represents cytosine, "G" represents guanosine, and "T" represents thymidine. The letters A, C, G, and T may be used to refer to the bases themselves, to refer to nucleosides, or to refer to nucleotides containing bases, as is standard in the art.

[0025] The term "DNA (deoxyribonucleic acid)" refers to a chain of nucleotides containing deoxyribonucleosides, each containing one of four nucleobases: adenine (A), thymine (T), cytosine (C), and guanine (G). The term "RNA (ribonucleic acid)" refers to a chain of nucleotides containing four types of ribonucleosides, each containing one of four nucleobases: A, uracil (U), G, and C. Specific pairs of nucleotides bind to each other in a complementary manner (called complementary base pairing). In DNA, adenine (A) pairs with thymine (T) and cytosine (C) pairs with guanine (G). In RNA, adenine (A) pairs with uracil (U) and cytosine (C) pairs with guanine (G). When a first nucleic acid strand binds to a second nucleic acid strand consisting of nucleotides complementary to the nucleotides in the first strand, the two strands combine to form a double strand. As used herein, “nucleic acid sequencing data,” “nucleic acid sequencing information,” “nucleic acid sequence,” “nucleotide sequence,” “genome sequence,” “gene sequence,” or “fragment sequence,” or “nucleic acid sequencing read” refers to any information or data indicating the order of nucleotide bases (e.g., adenine, guanine, cytosine, and thymine or uracil) within a molecule of nucleic acid such as DNA or RNA (e.g., whole genome, whole transcript, exome, oligonucleotide, polynucleotide, or fragment). It should be understood that this lesson considers sequence information obtained using all available various techniques, platforms, or technologies, including but not limited to capillary electrophoresis, microarrays, linkage-based systems, polymerase-based systems, hybridization-based systems, direct or indirect nucleotide identification systems, thermal sequencing, ion-based or pH-based detection systems, and electronic signature-based systems.

[0026] "Optional" or "optional" means that a subsequently described event, situation, or substance may or may not occur or exist, and that the description includes cases where said event, situation, or substance occurs or exists, and cases where it does not occur or does not exist.

[0027] Throughout the detailed description and claims of this specification, the word “comprising” and variations thereof such as “comprising” and “comprising” means “comprising but not limited thereto,” and is not intended to exclude, for example, other additives, components, integers, or steps. In particular, in a method described as comprising one or more steps or operations, each step is specifically considered to include what is enumerated (unless the step contains a limiting term such as “consisting of,” which means that each step is not intended to exclude, for example, other additives, components, integers, or steps not enumerated in the steps.

[0028] "Exemplary" means "an example of" and is not intended to indicate a desirable or ideal configuration. "Such as" is not used in a restrictive sense but is used for descriptive purposes.

[0029] A range may be expressed herein as "about" one specific value and / or "about" another specific value. When such a range is expressed, what is specifically considered and disclosed is also the range from one specific value and / or another specific value, unless the context otherwise specifically indicates otherwise. Similarly, when a value is expressed as an approximation, by the preceding use of "about," the specific value will be understood to form another specifically considered embodiment that is to be considered disclosed, unless the context otherwise specifically indicates otherwise. It will be further understood that each endpoint of the range is significant not only in relation to other endpoints but also independently of other endpoints, unless the context otherwise specifically indicates otherwise. Finally, it should be understood that all individual values ​​and sub-ranges of values ​​included within an explicitly disclosed range are also to be specifically considered and disclosed, unless the context otherwise specifically indicates otherwise. The foregoing applies regardless of whether some or all of these embodiments are explicitly disclosed in a particular case.

[0030] B. Method for identifying reliable receptor-pMHC binding and its uses

[0031] In some embodiments, the described method and system can identify reliable TCR-pMHC binding by analyzing multi-omics high-throughput binding data. The method and system may be referred to herein as ICON (Integrative Context-specific Normalization).

[0032] A method is disclosed comprising the steps of: receiving single-cell sequencing data including single-cell sequence data, dextramer sequence data, and single-cell T cell receptor (TCR) sequence data; filtering data associated with low-quality cells from the dextramer sequence data based on the single-cell sequence data; adjusting the dextramer sequence data based on background noise measurements; filtering data based on the presence or absence of specific receptor sequences from the dextramer sequence data based on the single-cell TCR data; and identifying data remaining in the normalized filtered dextramer sequence data as associated with reliable TCR-pMHC binding events.

[0033] Single-cell sequence data and corresponding receptor sequence data may be derived from various cell types, including T cells (αβ or γδ) and B cells. Accordingly, a method is disclosed comprising, as an example, receiving single-cell sequencing data including single-cell sequence data, dextramer sequence data, and single-cell T cell receptor (TCR) sequence data; filtering data associated with low-quality cells from the dextramer sequence data based on the single-cell sequence data; adjusting the dextramer sequence data based on background noise measurements; filtering data based on the presence or absence of α-chains or β-chains from the dextramer sequence data based on the single-cell TCR data; and identifying data remaining in the normalized filtered dextramer sequence data as associated with reliable TCR-pMHC binding events.

[0034] 1. Data acquisition

[0035] A method for acquiring, receiving, and / or determining multi-omics high-throughput combined data is disclosed. Fig. 1 As illustrated in, the system (100) is a single-cell immune profiling platform (102) It may include a single-cell immune profiling platform. (102) It is multi-omics high-throughput combined data (e.g., sequence data (104) It can be configured to generate ). In one aspect, multi-omics high-throughput combined data may include one or more of single-cell sequence data, dextramer sequence data, and / or single-cell receptor sequence data. Single-cell sequence data may include, for example, RNA-seq data. Dextramer sequence data may include, for example, dCODE-dextramer-seq and / or cell surface protein expression sequencing, also referred to as CITE-seq (cell indexing of transcripts and epitopes by sequencing). Single-cell receptor sequence data may include TCR-seq data, for example, such as paired αβ chain (or γδ chain) single-cell TCR-seq data.

[0036] In some aspects, multi-omics high-throughput combined data may be incorporated into a previously generated and disclosed method. In some aspects, multi-omics high-throughput combined data may be generated as part of the disclosed method.

[0037] In some aspects, Fig. 2 As illustrated in, single-cell immune profiling platform (102)It can be configured to label peripheral blood mononuclear cells (PBMCs) from a healthy human donor for sorting on cell type, such as T cells or B cells. In some aspects, the cells may be T cells (e.g., CD4+ or CD8+ cells). In some aspects, the T cells may be αβ T cells or γδ T cells. In some aspects, the cells may be B cells. Therefore, when labeling for sorting, the label may be a CD4, CD8, or B cell-specific label.

[0038] In some respects, once the cell type of interest is classified, the classified cells can be categorized based on whether they bind to a specific peptide-major histocompatibility complex (MHC) (pMHC). In some respects, the cells are a set of dextramers, for example, dCODE TM It can be combined with dextramer. In some aspects, dCODE TMDextramer® technology may be used. The dextramer may comprise two or more MHCs, peptides presented by each MHC, and a DNA barcode. In some aspects, a dextramer pool is used. In some aspects, the dextramer pool may comprise 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, or 100 single dextramers, each comprising a different pMHC, but is not limited thereto. In some aspects, the pool of dextramers comprises two or more of each of the single dextramers containing different pMHCs. In some aspects, two or more MHCs on a single dextramer are identical and therefore present the same peptide. In some aspects, the MHC may be MHC class I (MHC I) or MHC class II (MHC II). In some aspects, the DNA barcode includes one or more primer sequences, a peptide-MHC (pMHC) specific barcode, and a unique molecular identifier. In some aspects, the dextramer may additionally include a label. For example, the label may be a fluorescent label. In some aspects, cells that bind to a specific pMHC are classified based on the label on the dextramer. In some aspects, cells that bind to a specific pMHC are classified based on a labeled antibody specific to the dextramer.

[0039] In some aspects, cell classification for specific cell types and cell classification for cells that recognize dextramers can be performed simultaneously or sequentially.

[0040] In some aspects, after classifying cells bound to dextramers containing pMHC, each cell and its corresponding dextramer can be sequenced. In some aspects, both the cell sequence and the dextramer sequence (e.g., the DNA barcode sequence from the dextramer) have a common sequencing barcode, which allows determining which cell sequence is associated with which dextramer sequence. In some aspects, Next GEM technology can be used for sequencing. The common sequencing barcode is different from the DNA barcode found in the dextramer.

[0041] In some aspects, sequencing of cells bound to dextramers containing pMHCs is sequence data that may include single-cell sequence data, dextramer sequence data, and single-cell receptor sequence data. (104) The present invention provides... In some aspects, single-cell sequence data includes sequences derived from the whole cell genome or transcriptome. Thus, in some aspects, single-cell sequence data includes gene expression data. In some aspects, dextramer sequence data includes DNA barcode sequences. In some aspects, single-cell receptor sequence data includes sequences of specific receptors. For example, single-cell receptor sequence data includes single-cell TCR or B cell receptor (BCR) sequence data. In some aspects, single-cell TCR sequence data includes paired TCR sequence data. In some aspects, paired TCR sequence data includes sequence data for the α chain and β chain (if present) for each cell. In some aspects, paired TCR sequence data includes sequence data for the γ chain and δ chain (if present) for each cell. Thus, for each method and example described herein, sequencing of the alpha chain and beta chain may be exchanged for sequencing of the gamma chain and delta chain.

[0042] Fig. 1 The system depicted in(100) Returning to the point, in terms of work, sequence data (104) is a computing device (106) It can be provided to. Computing device (106) It can be, for example, a smartphone, tablet, laptop computer, desktop computer, server computer, etc. computing device (106) It may include a group of one or more servers. Computing device (106) is sequence data (102) It may be configured to create, store, maintain, and / or update various data structures, including databases for storing one or more of the computing devices. (106) The ICON (Integrative COntext-specific Normalization) module (108) and / or prediction module (110) It can be configured to operate one or more application programs such as the ICON module. (108) and prediction module (110) It can be stored and / or configured to operate separately on the same computing device or on a separate computing device.

[0043] In some aspects, the ICON module (108) is the received sequence data (104) It can be configured to analyze (e.g., multi-omics high-throughput combined data, single-cell sequence data, dextramer sequence data, single-cell receptor sequence data, etc.). Sequence data (104) It can include not only sequence information but also meta-information. Sequence data (104)As known to those skilled in the art, it may be saved in any suitable file format, including, for example, VCF files, FASTA files, or FASTQ files. FASTA and FASTQ are common file formats used to store raw sequence reads of high-throughput sequencing. FASTQ files store the identifier, sequence, and quality score string for each sequence read. FASTA files store only the identifier and sequence. Other file formats are considered.

[0044] In some aspects, Fig. 3 As shown in the ICON module (108) The stage 310 sequence data (104) Step of filtering low-quality cells from (e.g., dextramer sequence data), step 320 Sequence data regarding background noise (104) The step of adjusting, step 330 sequence data (104) Step of selecting T cells having paired αβ chains, step 340 sequence data (104) Step of applying dextramer signal correction, step 350 sequence data (104) Steps for performing cell-specific and / or pMHC-specific dextramer signal normalization and binder identification [ / g14] and step 360 A method comprising the step of identifying data remaining in normalized dextramer sequence data as associated with reliable TCR-pMHC binding events. 300 It can be configured to perform. In one embodiment, the ICON data processing may be performed in a donor, cell, and / or dextramer-specific context.

[0045] step 310 sequence data (104)The step of filtering low-quality cells from may include single-cell RNA-seq-based filtering of low-quality cells. ICON module (108) It can be configured to filter out low-quality cells, such as dichromats and dead cells. Cells with an unexpectedly large number of genes among the detected T cells (e.g., > 2,500 genes per cell) can be classified as dichromats, and cells with a high fraction of mitochondrial gene expression (e.g., ratio of mitochondrial gene expression UMI to total gene expression > 0.4) or too few detected genes (< 200 genes per cell) can be classified as dead cells. Data associated with low-quality cells is sequence data (104) (for example, It can be removed from dextramer sequence data.

[0046] In one embodiment, step 310 In the sequence (104) The step of filtering low-quality cells from may include, for each cell indicated in the dextramer sequence data, determining the number of genes based on single-cell sequence data; removing from the dextramer sequence data data data associated with cells having a number of genes outside a gene threshold range (the gene threshold range may be, for example, about 200 to about 2,500 genes); for each cell indicated in the dextramer sequence data, determining the fraction of mitochondrial gene expression based on single-cell sequence data; and removing from the dextramer sequence data data data associated with cells having a fraction of mitochondrial gene expression exceeding a gene expression threshold. The gene expression threshold may be about 40% of the total number of unique molecular identifiers.

[0047] step 320 Sequence data for background noise in (104)The adjustment step may include single-cell dCODE-dextramer-seq-based background adjustment. In one aspect, two types of background noise controls designed for dextramer binding assays are dextramer stained and negative control dextramer (NC_dex, from classified CD8+ T cells. nc Indicated by), and without sorting for dextramer (Dex_unsorted, du It includes dextramer-stained CD8+ T cells (indicated by ). To examine signal and noise distributions, the maximum dextramer signal at the Unique Molecular Identifier (UMI) of each cell can be selected to represent the optimal binding of each cell. Specifically, the non-specific dextramer binding signal of the cell is Max ( nc 1 , ..., nc n It can be expressed as ), n The maximum dextramer signal of the negative control dextramer included the dextramer pool. Dextramer stained and sorted samples (Dex_sorted, ds The dextramer binding signal of the cell from (indicated by ) is, m The maximum dextramer signal at the UMI of the test dextramer, Max ( ds 1 , 쪋, ds m It can be expressed as ). Similarly, the dextramer binding signal of cells from Dex_unsorted samples is Max ( du 1 , ... , du m ) It can be expressed as. The non-specific dextramer binding signal in UMI P 99.9 can be selected as a non-specific dextramer binding cutoff (absolute specificity of negative dextramer control can be excluded).

[0048] To estimate potential noise introduced by the cell sorting process, a cutoff for dextramer sorting efficiency can be determined by comparing the cumulative distribution of dextramer binding signals between Dex_sorted and Dex_unsorted samples. The Kolmogorov-Smirnov test (KS test) p-value can be calculated by comparing the cumulative curves of dextramer-sorted and dextramer-unsorted samples using each data point (dextramer UMI) as a sliding window. The dextramer UMI that defines the largest difference in dextramer binding signals between Dex_sorted and Dex_unsorted is ( argmax D s,u ) It can be used as a threshold value for estimating dextramer classification efficiency. Measure of estimated background noise of dextramer classified samples ( d ) can be defined as follows:

[0049] d = Max(P 99.9 , argmaxD s,u )

[0050] The dextramer signal (UMI) for each test dextramer of the classified cells is a measure of the estimated background noise ( d It can be corrected by subtracting ):

[0051] E c = E s - d

[0052] In one embodiment, step 320The step of adjusting data for background noise may include the step of determining classified dextramer sequence data and unclassified dextramer sequence data based on dextramer sequence data. The classified dextramer sequence data may include classified test dextramer sequence data (dex_sorted) and negative control dextramer sequence data (nc_dex). The unclassified dextramer sequence data may include unclassified test dextramer sequence data (dex_unsorted). Method 300 silver, stage 320 In this, for each cell indicated in the dextramer sequence data, based on the negative control dextramer sequence data (nc_dex), the maximum negative control dextramer signal Can determine ). Method 300 silver, stage 320 In this, for each cell indicated in the dextramer sequence data, based on the classified test dextramer sequence data (dex_sorted), the maximum classified dextramer signal Can determine ). Method 300 silver, stage 320 In, for each cell indicated in the dextramer sequence data, based on the unsorted test dextramer sequence data (dex_unsorted), the maximum unsorted dextramer signal ) can be determined.

[0053] method 300 silver, stage 320 In, based on the maximum negative control dextramer signal, dextramer binding background noise ( ) can be estimated, and based on the maximum classified dextramer signal and the maximum unclassified dextramer signal, the dextramer classification gate efficiency It can be estimated. The dextramer classification gate efficiency is, for example, of the classified test dextramer sequence data and unclassified dextramer sequence data It can be determined by the maximum difference between them.

[0054] method 300 silver, stage 320 In, dextramer combined background noise ( ) and dextramer classification gate efficiency Based on, the measured value of background noise ( d Determine ) and for each cell indicated in the dextramer sequence data, the measurement of background noise from the dextramer signal associated with each cell ( d ) can be deducted ( ).

[0055] In one embodiment, step 330 sequence data (104) The step of selecting T cells having paired α-β chains may include, for each cell indicated in the dextramer sequence data, determining the presence or absence of at least one α-chain and at least one β-chain based on single-cell TCR sequence data, and removing data associated with cells having only an α-chain, only a β-chain, or multiple α- or β-chains from the dextramer sequence data based on the presence or absence of at least one α-chain and at least one β-chain. 330 It may include a step of removing random data from single paired γδ chains and dextramer sequence data not associated with cells. Therefore, step 320 The same step for adjusting background noise in can be performed in relation to the presence or absence of γ chains and / or δ chains.

[0056] step 330 sequence data (104) The step of selecting T cells having paired αβ chains involves the step of removing random data from single paired αβ chains and dextramer sequence data not associated with cells. It may include. Single-cell receptor sequence data (e.g., single-cell TCR-seq data) can be used to determine data associated with T cells having only α-chains, only β-chains, and multiple α- or β-chains, and such data is sequence data (104) It can be removed from (e.g., dextramer sequence data). For T cells in which multiple α- or β-chains are detected, the α- or β-chain with the highest UMI number can be assigned to each T cell. For example, if 4 α-chains and 4 β-chains are detected in one T cell, the β-chain with the highest UMI can be selected from the list of all β-chains. The same applies to α-chains. The α- or β-chain selected from this process can be assigned to a cell.

[0057] method 300 The stage 340 In, sequence data (104) It may include a step of applying dextramer signal correction. Step 340 In, sequence data (104) The dextramer signal in can be corrected, and corrected dextramer sequence data can be generated. Although each dextramer has optimal binding conditions, it is impossible to arrange experimental conditions so that the multiplexed dextramer binding assay is optimal for all dextramers. This generates multiple dextramer bindings to the same T cell / clone. To correct for this effect, the following technique can be used to penalize the dextramer signal when binding simultaneously to the same T cell / clone.

[0058] Binding to dextramer Background noise-subtracted dextramer signal for T cells If defined as, Regarding T cells The fraction of the dextramer signal resulting from dextramer binding is further expressed as follows:

[0059]

[0060] TCR clonal type of T cells clones that bind to dextramer j, and The number of T cells belonging to What is represented as, Clone type that binds to dextramer The fraction of T cells belonging to is expressed as follows:

[0061]

[0062] Using these amounts, Binding to dextramer The corrected dextramer signal for T cells is calculated as follows:

[0063] .

[0064] method 300 silver, stage 350 In this step, for each cell indicated in the dextramer sequence data, cell-specific normalization of the dextramer signal associated with each cell may be performed and / or pMHC-specific normalization may be performed for each cell indicated in the dextramer sequence data to normalize the corrected dextramer sequence data. Such normalization may result in normalized dextramer sequence data. Step 350 It may additionally include binder identification. To make all dextramer binding signals similar, the corrected dextramer binding signals can be log-ratio normalized across 44 test dextramers within the cell. pMHC-specific normalization can subsequently be performed based on the Log-Rank distribution. A normalized dextramer UMI > 0 was empirically selected as a cutoff for pMHC-specific binders.

[0065] In one embodiment, the corrected dextramer sequence data is step 350 It can be normalized. For example, cell-specific normalization can be performed based on the log-rank distribution for each cell and / or pMHC-specific normalization can be performed to make the dextramer binding signals equivalent to one another. Adjusted dextramer binding signals of classified cells E c It is normalized across the test dextramer and then can be normalized across all cells as the following expression:

[0066]

[0067]

[0068] >= 0.9 can be empirically determined as a cutoff for pMHC-specific binders.

[0069] method 300 is , step 360 In this case, the data remaining in the normalized dextramer sequence data can be further identified as associated with reliable TCR-pMHC binding events. This data can be considered as part of the training dataset for use in the machine learning process. Generated processed sequence data (104) (e.g., training data set) is a prediction module (110) It can be provided to.

[0070] C. Method for using reliable receptor-pMHC binding for machine learning

[0071] Now, referring to FIG. 4, a prediction module (110) is described. The prediction module (110) may be configured to train at least one ML module (430) configured to predict binding affinity for a given receptor sequence based on the analysis of one or more training data sets (410) by a training module (420) using machine learning ("ML") technology.

[0072] training data set (410)It may include one or more receptor sequences, one or more genetic identifiers, binding status, and an identifier of the peptide to which the receptor sequence is bound (if present). The binding status may indicate "Yes" for a receptor sequence bound to a peptide, or "No" for a receptor sequence not bound to a peptide. In the case of a receptor sequence bound to a peptide, the peptide identifier can be used to identify the antigen associated with the peptide. This data is from the ICON module (108) Sequence data processed by (104) It may be derived wholly or partially from. In one embodiment, the TCR-CDR3 amino acid sequence is sequence data comprising associated V, D, and J gene identifiers, a label indicating binding status (yes, no), and an identifier of the peptide to which the TCR-CDR3 amino acid sequence is bound. (104) It can be determined from. The TCR-CDR3 amino acid sequence can be numerically encoded to represent 20 possible amino acids. Padding can be applied to the sequence as needed. V and J genetic identifiers can be one-hot encoded to provide a categorical and discrete representation of the genetic identifiers in numeric space. The encoded TCR-CDR3 amino acids and V and J genetic identifiers can be concatenated to represent a single TCR record and may be associated with a label (yes, no) indicating the binding status. The label may additionally indicate the specific peptide to which the TCR is bound. One or more TCR records are combined to form a training dataset (410) It can generate.

[0073] A subset of TCR records is the training data set (410)Alternatively, it may be randomly assigned to the test data set. In some implementations, the assignment of data to the training or test data set may not be completely random. In this case, one or more criteria may be used during the assignment. Generally, any appropriate method may be used to assign data to the training or test data set while ensuring that the distributions of Yes and No Labels are somewhat similar in the training and test data sets.

[0074] Training Module (420) ...a training data set according to one or more feature selection techniques (410) By extracting a set of feature parts from multiple TCR records (e.g., labeled as yes) within the ML module (430) Can train. Training module (420) A training data set containing statistically significant features of positive examples (e.g., labeled "Yes") and statistically significant features of negative examples (e.g., labeled "No"). (410) ML module by extracting a set of feature parts from (430) It can train.

[0075] Training Module (420) training data sets in various ways (410) A set of feature parts can be extracted from. Training module (420) It can perform feature extraction multiple times by using different feature extraction techniques each time. In one example, the feature sets generated using different techniques are each different machine learning-based classification models (440) It can be used to generate. In one example, a feature set with the highest quality metrics can be selected for use in training. Training module (420)is one or more machine learning-based classification models configured to indicate whether a new receptor sequence (e.g., having an unknown binding state) is likely to bind to a peptide or pMHC. (440A-440N) You can use feature set(s) to build it.

[0076] training data set (410) is the training data set (410) Any dependency, association, and / or correlation between the features within and the yes / no labels may be analyzed. The identified correlations may take the form of a list of features associated with different yes / no labels. As used herein, the term “feature” may refer to any characteristic of a data item that can be used to determine whether the data item belongs to one or more specific categories. For example, the features described herein may include one or more sequence patterns, amino acid sequences of one or both of the α and β chains, and names of the v and j gene segments of one or both of the α and β chains.

[0077] The feature selection technique may include one or more feature selection rules. One or more feature selection rules may include feature generation rules. The feature generation rules are based on a training dataset. (410) It may include determining which feature part occurs above a threshold number and identifying the feature part satisfying the threshold as a candidate feature part.

[0078] A single feature selection rule or multiple feature selection rules can be applied to feature selection. Feature selection rules can be applied in a cascading manner, or they can be applied in a specific order and as a result of a previous rule. For example, a feature occurrence rule is applied to the training data set. (410)It can be applied to generate a first list of feature parts. The final list of candidate feature parts may be analyzed according to additional feature selection techniques to determine one or more candidate feature group members (e.g., feature group members that can be used to predict combinations). Any appropriate computational technique may be used to identify candidate feature group members using any feature selection technique, such as filters, wrappers, and / or embedded methods. One or more candidate feature group members may be selected according to a filter method. Filter methods include, for example, Pearson correlation, linear discriminant analysis, analysis of variance (ANOVA), chi-square, combinations thereof, etc. The selection of feature parts according to the filter method is independent of any machine learning algorithm. Instead, feature parts may be selected based on scores from various statistical tests regarding correlation with the outcome variable (e.g., yes / no).

[0079] As another example, one or more candidate feature groups may be selected according to a wrapper method. The wrapper method may be configured to train a machine learning model using a subset of features. Based on inferences derived from the previous model, features may be added and / or removed from the subset. Wrapper methods include, for example, forward feature selection, backward feature removal, recursive feature removal, and combinations thereof. As an example, forward feature selection may be used to identify one or more candidate feature groups. Forward feature selection is an iterative method that starts a machine learning model without features. In each iteration, features that best improve the model are added until the addition of new variables no longer improves the performance of the machine learning model. As an example, backward removal may be used to identify one or more candidate feature groups. Backward removal is an iterative method that starts a machine learning model with all features. In each iteration, minimal significant features are removed until no improvement is observed from the removal of features. Recursive feature removal can be used to identify one or more candidate feature groups. Recursive feature removal is a greedy optimization algorithm that aims to find the subset of features with the best performance. Iterative feature removal iteratively generates models and, in each iteration, sets aside the best or worst performing features. Recursive feature removal constructs the next model with remaining features until all features are exhausted. Then, recursive feature removal ranks the features based on the order of removal.

[0080] As another example, one or more candidate feature groups can be selected according to an embedded method. The embedded method combines the qualities of filter and wrapper methods. Embedded methods include, for example, the Least Absolute Shrinkage and Selection Operator (LASSO) and Ridge regression, which implement a penalty function to reduce overfitting. For example, LASSO regression performs L1 regularization by adding a penalty corresponding to the absolute value of the coefficient size, and Ridge regression performs L2 regularization by adding a penalty corresponding to the square of the coefficient size.

[0081] Training Module (420) After generating this feature set(s), the training module (420) is a machine learning-based classification model based on feature set(s) (440) It can generate. A machine learning-based classification model can refer to a complex mathematical model for data classification generated using machine learning techniques. In one example, a machine learning-based classification model (440) It may include a map of support vectors representing boundary features. For example, the boundary features may be selected from a set of features and / or may represent the highest-ranking feature in the set of features.

[0082] In one embodiment, the training module (420) is the training data set (410) A machine learning-based classification model for each classification category (e.g., Yes, No) using a set of feature parts extracted from (440A-440N) It can be built. In some examples, machine learning-based classification models (440A-440N) is a single machine learning-based classification model (440A-440N) It can be combined as. Similarly, the ML module (430) is a single or multiple machine learning-based classification model (440)A single classifier and / or single or multiple machine learning-based classification models containing (440) It can represent multiple classifiers containing.

[0083] The extracted feature parts (e.g., one or more candidate feature parts) can be combined with a classification model trained using machine learning approaches, such as discriminant analysis; decision trees; nearest neighbor (NN) algorithms (e.g., k-NN models, replicator NN models, etc.); statistical algorithms (e.g., Bayesian networks, etc.); clustering algorithms (e.g., k-means, mean-shift, etc.); neural networks (e.g., reservoir networks, artificial neural networks, etc.); Support Vector Machines (SVM); logistic regression algorithms; linear regression algorithms; Markov models or chains; principal component analysis (PCA) (e.g., for linear models); multi-layer perceptron (MLP) ANNs (e.g., for non-linear models); replicator reservoir networks (e.g., for non-linear models, typically for time series); random forest classification; combinations thereof and / or others. The generated ML module (430) It may include decision rules or mappings for each candidate feature region to assign a binding state to a new receptor sequence.

[0084] In one embodiment, the training module (420) is a machine learning-based classification model as a Convolutional Neural Network (CNN) (440) It can be trained. The CNN may include at least one synthetic feature layer and three fully connected layers leading to a final classification layer (softmax). The final classification layer may be finally applied to combine the outputs of the fully connected layers using a softmax function as is known in the art.

[0085] Candidate feature(s) and ML module (430)It can be used to predict the binding state (and associated peptide) of multiple TCR records in a test dataset. In one example, the result for each TCR record includes a confidence level corresponding to the likelihood or probability that the receptor sequence binds to the peptide. The confidence level may be a value between 0 and 1, which may indicate the likelihood that the receptor sequence belongs to a yes / no binding state in relation to one or more peptides. In one example, when there are two states (e.g., yes and no), the confidence level is a value p It can correspond to, which refers to the probability that a specific receptor sequence belongs to the first state (e.g., yes). In this case, the value 1- p This may refer to the probability that a specific receptor sequence belongs to the second state (e.g., No). Generally, multiple confidence levels may be provided for each test receptor sequence and for each candidate feature when there are more than two states. Top-performing candidate features can be determined by comparing the results obtained for each test receptor sequence with the known Yes / No binding states for each test receptor sequence. Generally, the top-performing candidate features will have results that closely match the known Yes / No binding states.

[0086] Top-performing candidate feature(s) can be used to predict the yes / no binding status of receptor sequences for one or more peptides. For example, a new TCR sequence can be determined / received. The new TCR sequence is an ML module ( 430 It can be provided in ), which, based on the top-performing candidate feature, can classify the new TCR sequence as either bound (yes) or unbound (no) and indicate the bound peptide(s).

[0087] Fig. 5 is a training module (420) ML module using (530)An exemplary training method for generating (500) It is a flowchart illustrating. Training module (420) supervised, unsupervised, and / or semi-supervised (e.g., reinforcement-based) machine learning-based classification models (440) It can be implemented. Fig. 5 The method illustrated in (500) This is an example of a supervised learning method; variations of this example of a training method are discussed below, but other training methods can be similarly implemented to train unsupervised and / or semi-supervised machine learning models.

[0088] Training methods (500) The stage 510 ICON module (108) The first sequence data processed by can be determined (e.g., access, reception, retrieval, etc.). The sequence data may include a set of labeled receptor sequences. The label may correspond to the binding state (e.g., yes or no) and identification of the peptide(s) to which the receptor sequence is bound.

[0089] Training methods (500) The stage 520 In this, a training data set and a test data set can be generated. The training data set and the test data set can be generated by randomly assigning labeled receptor sequences to the training data set or the test data set. In some embodiments, assigning labeled receptor sequences as training or test samples may not be entirely random. For example, most of the labeled receptor sequences may be used to generate the training data set. For example, 75% of the labeled receptor sequences may be used to generate the training data set, and 25% may be used to generate the test data set.

[0090] Training methods (500) silver, stage 530In this, for example, one or more features (e.g., extraction, selection, etc.) that can be used by a classifier to distinguish between different classifications of binding states (e.g., yes vs. no) for one or more peptides can be determined. As an example, a training method (500) A set of feature parts can be determined from labeled receptor sequences. In additional examples, the set of feature parts can be determined from labeled receptor sequences different from those in either the training dataset or the test dataset. That is, the labeled receptor sequences can be used for feature determination rather than for training a machine learning model. These labeled receptor sequences can be used to determine an initial set of feature parts, which can be further reduced using the training dataset.

[0091] Training methods (500) The stage 540 In this step, one or more machine learning models can be trained using one or more feature parts. In one example, the machine learning model can be trained using supervised learning. In another example, other machine learning techniques, including unsupervised learning and semi-supervised learning, may be used. Step 540 Machine learning models trained in can be selected based on different criteria depending on the problem to be solved and / or the data available in the training dataset. For example, machine learning classifiers may experience varying degrees of bias. Therefore, one or more machine learning models 540 Can be trained at, stages 550 It can be optimized, improved, and cross-validated.

[0092] Training methods (500) The stage 560 You can build a predictive model by selecting one or more machine learning models. The predictive model can be evaluated using a test data set. The predictive model analyzes the test data set and, step 570A predicted combination state can be generated in. The predicted combination state is step 580 These values ​​can be evaluated to determine whether the desired level of accuracy has been achieved. The performance of the prediction model can be evaluated in multiple ways based on the classification of multiple true positives, false positives, true negatives, and / or false negatives of multiple data points indicated by the prediction model.

[0093] For example, a false positive in a predictive model may refer to the number of times the predictive model incorrectly classifies a receptor sequence as bound when it is not actually bound. Conversely, a false negative in a predictive model may refer to the number of times the machine learning model classifies a receptor sequence as unbound when it is actually bound. True negatives and true positives may refer to the number of times the predictive model correctly classifies one or more receptor sequences as bound or unbound. The concepts of recall and precision are associated with this measurement. Generally, recall refers to the ratio of true positives to the sum of true positives and false negatives, which quantifies the sensitivity of the predictive model. Similarly, precision refers to the ratio of the sum of true positives to false positives. When these desired accuracy levels are reached, the training phase ends and the predictive model (e.g., ML module) (430) this step 590 It can be output from; however, when the desired accuracy level is not reached, training methods with variations, such as considering larger sequence datasets, for example (500) The subsequent iteration of is step 510 It can be performed starting from.

[0094] In one embodiment, a flexible framework for studying TCR-pMHC specificity, referred herein as TCRAI, is provided. In one embodiment, TCRAI may utilize TensorFlow 2. TCRAI is highly modular and allows for adjustments to the model architecture. Any number of V(D)J genes and CDR regions of TCRs may be defined as inputs to a model in text form. Through a "processor" object that converts text into a numerical representation, one may choose how to process these inputs in numerical form in a non-trainable manner. Then, these numerical inputs may be further processed in a trainable manner through an "extractor" object that forms a block of the neural network and provides an output vector representation of the input data, referred herein as a TCRAI fingerprint. A TCRAI fingerprint may be linked to a single TCRAI fingerprint that describes the input TCR through a single numerical vector. Then, the TCRAI fingerprint is passed through "closer" objects that form the final blocks of the neural network architecture, which can generate predictions for the input TCRs. TCRAI thus provides pre-built processors, extractors, and closers. TCRAI can be configured to perform binomial, multinomial, regression, and / or other operations by choosing to configure different closer objects. In one embodiment, TCRAI can be used to build a model for predicting whether a given TCR can bind to a specific pMHC complex.

[0095] In one embodiment, TCRAI may use 1D convolution and batch normalization for CDR3 sequences and low-dimensional representations for genes, which leads to model regularization and forces the model to learn stronger gene associations.

[0096] In one embodiment, the input information of the TCR can be processed in a numeric format. For each CDR3 sequence, the amino acid can be converted into an integer, and the integer vector can be encoded for one-time use. For V and J genes, a dictionary of gene types in integers can be constructed for each V and J gene and can be used to convert each gene into an integer.

[0097] The neural network architecture applied to the processed input information may include an embedding layer and a convolutional network. Specifically, the processed CDR3 residues may be embedded in a 16-dimensional space through learned embeddings, and the generated numeric CDR3 may be fed through one or more (e.g., 3) 1D convolutional layers. In one embodiment, a filter with dimensions [64,128,256], kernel widths [5,4,4], and intervals [1,3,3] may be used. Each convolution may be activated by dropout and batch normalization following exponential linear unit activation. Following these three convolutional blocks, global max pooling may be applied to the final feature, and this process encrypts each CDR3 by a vector of length 256, "CDR3 fingerprint". The processed gene input for each gene is one-hot encoded via learned embeddings and embedded in a reduced-dimensional space (e.g., 16 for the V gene and 8 for the J gene) to provide the "gene fingerprint" of each gene as a vector. Then, the fingerprints of all selected CDR3s and genes can be concatenated together into a single vector, the "TCRAI fingerprint." The TCRAI fingerprint passes through a final fully connected layer to provide binomial prediction (single output value, sigmoid activation), regression prediction (single output, no activation), or multinomial prediction (multiple output values, softmax activation).

[0098] In one embodiment, TCR sequencing files may be collected as multi-omics high-throughput combined data in raw CSV format. Sequencing files may be parsed to obtain the amino acid sequence of CDR3 after removing unproductive sequences. Clones having different nucleotide sequences but identical matching amino acid sequences from the CDR3 V, D, and J genes may be aggregated together under a single TCR. Thus, each TCR record may include a single pair of α and β TCR chains having the CDR3 amino acid sequence and the V and J genes for each chain.

[0099] The data can be split for each model into a training set (e.g., 76.5%), a validation set (e.g., 13.5%), and a left-out test set (e.g., 10%), followed by 5-fold Monte Carlo cross-validation (MCCV) on the training set. Models can be trained by minimizing cross-entropy loss through the Adam optimizer, and cross-entropy loss can be weighted for each class by a weight of 1 / (number of classes * fraction of samples within that class). With the left-out validation dataset, early stopping may be introduced to prevent overfitting; if the validation loss increases for more than 5 time points, the model stops training, and the weights of the model with the minimum validation loss are restored. When training multiple models, only the training rate and batch size need to be adjusted during cross-validation. After cross-validation, the hyperparameters that perform optimally can be selected, and the model can be retrained on the entire training set using the validation set to control for early stopping. Then, the retrained model can be evaluated on the left-out test set.

[0100] The TCRAI model can generate both predictions for TCRs that bind to a specific pMHC (or, in the case of polynomials, one of many pMHCs) and numerical vectors (TCRAI fingerprints) describing TCRs in the context of the question of whether they can bind to pMHCs (e.g., by encoding paired αβ chain CDR3 amino acid sequences and the V and J genes of each TCR as one-dimensional input vectors).

[0101] In one embodiment, the distribution of fingerprints can be analyzed to identify groups of TCRs having different combination patterns. Fingerprints can be reduced to a two-dimensional space, for example, using UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. When using a model trained on one dataset to infer fingerprints for another invisible dataset, the UMAP projector can be fitted to the TCRs from the training dataset and the projector can be used to transform the TCRs from the invisible set.

[0102] When clustering TCR fingerprints, the fingerprints of all TCRs in the dataset can be projected into a two-dimensional space as described above, and then TCRs with strong true values ​​(STP, binomial prediction >0.95) can be selected. Then, these STPs can be clustered in the two-dimensional space, for example, using a k-means classifier. Other clustering algorithms may be used. Then, the TCRs within each cluster can be collected and used to construct CDR3 motif logos (using weblogo), gene-utilization, and / or UMI distributions by pairing the unique TCR clones within the cluster with all repeated clones in the high-throughput data.

[0103] D. How to use

[0104] In one aspect, a trained predictive model (e.g., a machine learning classifier) ​​can be used to predict the binding status of a TCR sequence to one or more peptides. TCR sequences can be presented to the machine learning classifier. The machine learning classifier can predict the likelihood that a TCR sequence will bind to one or more specific peptides. Similarly, multiple TCR sequences can be presented to the machine learning classifier. For each of the multiple TCR sequences, the machine learning classifier can predict the likelihood that each TCR sequence will bind to one or more specific peptides. In one aspect, the machine learning classifier can generate a TCR-peptide map as shown in the exemplary output below.

[0105]

[0106] Therefore, the generated TCR-peptide map can be used to rapidly identify peptides likely to bind to the subject's TCR sequence. Biological samples (e.g., blood) are obtained from the subject, cells are isolated, and sequenced. The subject's TCR sequence can be identified and compared with the TCR-peptide map to identify the peptide most likely to bind to the subject's TCR sequence.

[0107] In some aspects, identifying and evaluating antigen-specific T cells can be used to better understand the activity of drugs in the context of monotherapy and combination therapy, to identify features of potent anti-tumor T cells, to screen immunogenic epitopes in a haplotype-related manner, to develop new vaccines and TCR therapies, and to develop peptide binding algorithms based on TCR sequence features.

[0108] In some aspects, a method for identifying a subject using the binding pattern of the subject's TCR is disclosed. For example, blood may be collected (first blood collection), cells from the blood may be processed through a single-cell-based immune profiling platform, and the generated data may be processed according to the ICON method described herein. In some aspects, cells are exposed to various dextramers containing pMHCs from a wide range of immunogens. After performing the ICON method as described herein, a reliable TCR binding pattern can be determined. In some aspects, the TCR binding pattern indicates the specificity of the TCR to the immunogen on the dextramer. Then, blood may be collected at a different time point (day, week, month, year) from the first blood collection (second blood collection). In some aspects, the second blood collection is about 10 15 Since there are several possible TCR sequences, there is a possibility that T cells with TCRs having sequences different from those present in the first blood collection may be included, but the TCR binding pattern is expected to be unlikely to change. Cells from the second blood collection may be exposed to the same dextramer used in the first blood collection, and the generated data can be analyzed according to the ICON method. Regardless of the different TCR sequences, the binding data from the first and second blood collections can be compared and used to determine whether they originated from the same subject.

[0109] In some aspects, a method for identifying a subject using machine learning to predict the binding pattern of the subject's TCR is disclosed. Reliable TCR binding data can be identified according to the ICON method as described herein. In some aspects, reliable TCR binding data can be used to train a machine learning classifier as described herein. The trained machine learning classifier can be used to predict the specific TCR binding pattern of the subject. In some aspects, blood can be collected (first blood collection), and the trained machine learning classifier can be used to predict the TCR binding pattern. Then, blood can be collected at a different time point from the first blood collection (day, week, month, year later) (second blood collection). In some aspects, the second blood collection is about 10 15 Since there are several possible TCR sequences, there is a possibility that T cells with TCRs having sequences different from those present in the first blood collection may be included, but the TCR binding pattern is expected to be unlikely to change. Regardless of the different TCR sequences, a trained machine learning classifier can be used to predict the second TCR binding pattern using data derived from the second blood collection. It is possible to predict that the second blood collection was performed from the same subject as the first blood collection based on the TCR signature.

[0110] In some aspects, TCR or BCR binding patterns can be established using the described methods. In some aspects, having reliable TCR data identified using the methods described herein enables someone, such as a medical professional, to infer the subject's antigenic or vaccination history. In some aspects, reliable TCR data identified using the ICON methods described herein enables someone, such as a medical professional, to infer the pathogens to which the subject was exposed or the countries the subject visited. For example, the presence of TCR binding data for pathogens found only in Africa may indicate that the subject was in Africa and was exposed to these pathogens.

[0111] In some aspects, reliable TCR data identified using the ICON method described herein can be used to assess the subject's current immune status. For example, blood may be collected (first blood collection), cells from the blood may be processed through a single-cell-based immune profiling platform, and the generated data may be processed according to the ICON method described herein, thereby producing TCR binding data. In some aspects, the dextramer used to establish the TCR binding data contains tumor-specific pMHCs. Thus, once the TCR binding data has been normalized using the ICON method and reliable TCR binding data has been established, the presence of predicted tumor-specific TCRs can be determined. For example, the reliable TCR data can be used in the disclosed machine learning (CNN) method, and thus the blood from the subject can be analyzed for the presence of predicted tumor-specific TCRs. Thus, the presence of tumor-specific TCRs can lead to the early detection of cancer before any tumor or cancer symptoms are detected.

[0112] In some aspects, a method for selecting T cells for T cell-based therapy is disclosed. In some aspects, training data may be accumulated using the disclosed machine learning classification method. In some aspects, the classifier may assign a pMHC binding probability to each tested TCR sequence. In some aspects, the tested TCR sequence is associated with a T cell, where the T cell may be derived from a primary or secondary cell culture. This avoids the need to perform binding analysis on all T cells under test to determine whether each T cell possesses a TCR specific to a different pMHC. Instead, the classifier relies on determining the probability of TCR-pMHC binding. Then, these TCRs classified as highly selective for the specific pMHC, and the T cells containing the TCRs accordingly, can be used in T cell therapy. In some aspects, T cells identified through the machine learning classifier may provide safer cell therapy than T cells identified through binding analysis, because only the most reliable binding data was used to generate the training data used to classify the TCRs associated with the selected T cells.

[0113] In some aspects, a method for monitoring immunity is disclosed. In some aspects, blood may be collected from a subject receiving immunotherapy (e.g., vaccine therapy; immune checkpoint therapy), and cells, particularly T cells, may be classified as having specificity for an epitope of interest based on training data established in the disclosed machine learning approach. In some aspects, if it is determined that T cells have specificity for an epitope of interest, it can be inferred that the subject will or is responding to the immunotherapy. For example, if the immunotherapy is a vaccine that induces an immune response to a cancer-specific antigen, T cells obtained from the subject will be classified based on their binding probability to the cancer-specific antigen. If T cells are selected as having a high binding probability to the cancer-specific antigen based on training data obtained using single-cell immune profiling techniques and ICON, the subject will be considered a responder to the immunotherapy (e.g., vaccine).

[0114] In some aspects, a TCR epitope mapping method using the disclosed method is disclosed. In some aspects, TCR epitope mapping is a term referring to a process of identifying the specific (in some cases the shortest) amino acid sequence of an epitope of a specific antigen that is recognized by T cell (CD4+ and / or CD8+) receptors and is likely to simultaneously stimulate a long-lasting cytotoxic immune response. While performing the disclosed single-cell immune profiling platform technology, a dextramer may be used, wherein all different epitopes from one or more antigens of interest may be presented on the dextramer. That is, a single dextramer may contain pMHC, wherein the peptide of the pMHC is a single epitope from one or more antigens of interest, and a sufficient amount of dextramer is used so that all epitopes of one or more antigens of interest are present on the pMHC on the dextramer. T cells may be exposed to dextramers in the disclosed single-cell immune profiling platform, wherein the dextramer contains a single epitope from one or more antigens of interest, and sufficient dextramer is used so that all epitopes of one or more antigens of interest are present on the pMHC on the dextramer. Single-cell sequence data, dextramer sequence data, and single-cell TCR sequence data obtained from single-cell immune profiling may provide data regarding T cells bound to different dextramers (e.g., epitopes). The single-cell immune profiling data is then processed using an ICON as described herein to yield binding data for cells having the most reliable binding to one or more epitopes of one or more antigens of interest. In some aspects, machine learning classification of TCRs binding to one or more epitopes of one or more antigens of interest may be used to predict which T cells from the subject may respond to a specific antigen (e.g., tumor antigen).

[0115] E. Kit

[0116] In addition to the materials described above, other materials may be packaged together in any suitable combination as a kit useful for carrying out the disclosed method or assisting in its performance. It is useful when the kit components within a given kit are designed and adapted for use together in the disclosed method. For example, a kit for generating single-cell sequencing data is disclosed, and the kit includes a reagent for single-cell immune profiling. In some aspects, the kit may include one or more of the disclosed dextramers, including pMHC. In some aspects, the kit may include the following GEM sequencing materials. In some aspects, the kit may include multi-omics high-throughput combined data comprising one or more of single-cell sequence data, dextramer sequence data, and / or single-cell receptor sequence data.

[0117] example

[0118] The following examples illustrate the present method and system related to the detection of colorectal cancer. The following examples are not intended to limit the use thereof.

[0119] A. Example 1

[0120] 1. Result

[0121] i. Multi-omics high-throughput TCR-pMHC coupling data.

[0122] 10x Genomics recently generated an extensive and publicly available TCR-pMHC binding dataset. In their initial report, four HLA haplotype healthy donors ( Fig. 19 Binding profiles of over 150,000 CD8+ T cells from ) were evaluated across 44 pMHC dextramers by simultaneously sequencing T cell αβ chain pairs and transcripts, while using a single-cell-based immune profiling platform to directly detect antigen binding to T cells ( Fig. 2 The dextramer pool consists of epitopes with known common viral and oncolytic responses across eight HLA alleles ( Fig. 20 ).

[0123] Described herein is a highly multiplexed dextramer binding dataset generated at the single-cell level. 10x Genomics used a simple approach to determine pMHC-binding TCRs by applying a total cutoff for background noise and non-specific dextramer binding to all donors. However, from the TCR-pMHC binding events identified by this approach, we discovered an unexpectedly high number of indiscriminate cross-HLA and cross-peptide associations, particularly in donors 3 and 4 ( Fig. 11a ). During additional testing, Donor 3's data was excluded from this study due to data quality issues ( Fig. 11b ).

[0124] To robustly identify reliable coupling events from such high-throughput TCR-pMHC coupling data, ICON, an integrative text-specific normalization method, was developed. Fig. 6a , Fig. 12and Methods). The ICON data normalization process was performed in a donor-specific context by individually taking multi-omics high-throughput binding data from each donor as input data. Briefly, good quality cells (live and unicellular) were selected using single-cell transcriptome data. Then, negative control dextramers (n = 6) and dextramer-unclassified samples were used as background controls for each donor to empirically estimate the background binding noise for each donor. Subsequently, the raw dextramer binding signal was corrected by subtracting the background noise individually estimated for each donor. Next, the corrected dextramer signal was normalized across cells and pMHC to generate a directly similar dextramer binding signal. The distribution of the ICON-normalized dextramer binding signal and the binding specificity of the extended T cell clones indicate that the signal-to-noise ratio of the high-throughput TCR-pMHC binding data was significantly increased ( Figures 6a and 6b and Fig. 12b and Fig. 13 ).

[0125] ii. TCR-pMHC binding events identified from 10x Genomics high-throughput data.

[0126] When ICON was applied, a total of 20,843 CD8+ T cells were identified from 1,514 indigenous T cell clones binding to 29 pMHCs from 3 donors ( Fig. 7a , Fig. 21and methods). The number of unique TCR-pMHC interactions identified from this high-throughput dataset is similar in size to the total number of paired αβ TCRs in VDJdb. Among pMHC-binding TCRs, 98.9% of the total TCRs (94.7% of the unique TCRs) bind to 7 pMHCs: B*08:01_RAKFKQLL_BZLF1_EBV, A*02:01_GILGFVFTL_Flu-MP_Influenza, A*11:01_IVTDFSVIK_EBNA-3B_EBV, A*03:01_KLGGALQAK_IE-1_CMV, A*11:01_AVFDRKSDAK_EBNA-3B_EBV, A*02:01_GLCTLVAML_BMLF1_EBV and A*02:01_ELAGIGILTV_MART-1_Cancer ( Fig. 7b and Fig. 16 and Fig. 17 ).

[0127] Dextramer pool ( Fig. 14 and Fig. 15 Donors 1 and 2, having the most common HLA haplotype (A*02:01) in ), share a significant portion of the intrinsic TCR-pMHC response (n = 38). Fig. 7c Donor 4 is A*02:01-negative and has an HLA haplotype different from Donors 1 and 2 ( Fig. 19 The shared pMHC binding TCR sequence between the binding of donor 4 and donors 1 and 2 was not observed ( Fig. 7c This indicates that the TCR-pMHC binding pattern is most likely HLA restriction.

[0128] Interestingly, 37% of TCRs with a shared β-chain pair with a different α-chain. This rate is slightly lower for shared TCR α-chains (30.9%). While the majority (~92%) of TCRs with shared α- or β-chains bind to sample pMHCs, ~8% of them recognize different pMHCs ( Fig. 7dThis indicates that αβ pairing information is necessary for accurate inference of TCR functionality.

[0129] The dual specificity (specificity versus degeneracy) of TCRs has been proposed as an important feature of the immune response mechanism that sufficiently distinguishes itself from foreign peptides to avoid autoimmune reactivity while maintaining broad antigen coverage. Indeed, highly specific yet indiscriminate TCR-pMHC interactions have been observed. 98.7% of native TCRs bind to a single specific pMHC, while the remaining TCRs interact with two or three pMHCs ( Fig. 7e & a Although TCRs capable of interacting with more than one epitope have been observed, these TCR-pMHC interactions generally follow HLA type-specific patterns. Over 99.3% of binding events are HLA-matched, of which 11.6% are associated with cross-recognition between HLA A*03-supertype family members HLA A*03:01 and A*11:01, which share similar major anchor sites of the presented peptide. However, 0.7% of binding events are cross-HLA type interactions.

[0130] iii. Synthetic Neural Network (CNN)-based T cell antigen-specific classification.

[0131] Using such large and diverse TCR-pMHC binding datasets requires more robust functional classifiers to computationally validate or prioritize these binding events. Recent research has demonstrated that synthetic neural networks (CNNs) can learn high-dimensional information from TCR sequences and thus can robustly predict TCR-pMHC binding. A CNN-based framework was adapted to validate and / or predict TCR-pMHC binding. Briefly, the V and J genes of each TCR, as well as the paired αβ chain CDR3 amino acid sequences, were encoded as one-dimensional input vectors. Specifically, CDR3 amino acid sequences were encoded using trainable embeddings, and V and J gene segments were transformed into vectors. The CNN structure may include three fully connected layers leading from a single convolutional feature layer to a final classification layer ( Fig. 8a and method). To address potential biases that may be introduced by having an unbalanced number of coupled and uncoupled TCRs for a given pMHC, a class-weighted cost function was used in training (method).

[0132] To evaluate the performance of this CNN-based model, 11 pMHC-specific binding T cell repertoires generated by traditional single multimer binding and antigen re-exposure analysis were collected as the gold standard dataset ( Fig. 23 Each selected pMHC binding repertoire was split into training, validation, and test sets. The CNN-based model had an average area under the curve (AUC) of 0.90 ((AUC) With a state of = 0.90), the antigen binding specificity of the selected TCRs could be classified ( Fig. 8b The CNN-based classifier was compared with the TCR sequence similarity distance-based classifier. The CNN-based classifier outperforms the distance-based prediction model ( Fig. 8c), especially in the case of a very diverse pMHC repertoire ( Fig. 14 The difference in classification performance between CNN-based and distance-based classifiers is ( AUC) is positively correlated with the diversity of pMHC-bound T cell repertoire measured by Shannon entropy ( Fig. 8d ).

[0133] iv. Classification of pMHC binding repertoires identified from 10x Genomics high-throughput data.

[0134] Next, a CNN-based classifier was applied to the top 7 pMHC binding repertoires identified from the 10x Genomics binding data ( Fig. 7b and Fig. 15 ). Average (AUC) of 7 pMHC repertoires Classified as = 0.89( Fig. 9a In these data, as with the selected dataset, CNN-based classifiers outperform distance-based models ( Fig. 16 To further computationally validate these binding TCRs, four pMHC repertoires (A*02:01_ELAGIGILTV_MART-1, A*02:01_GILGFVFTL_Flu-MP, A*02:01_GLCTLVAML_BMLF1_EBV, and A*11:01_AVFDRKSDAK_EBNA-3B_EBV) binding to TCRs in a selected dataset were also used. A CNN-based classifier was trained using four repertoires identified from the 10x Genomics dataset to predict additional A*02:01_ELAGIGILTV_MART-1 binding repertoires from in-house independent antigen re-exposure experiments (Methods), in addition to the four selected repertoires. Fig. 9b It shows prediction results comparable to high performance on the training set.

[0135] Historically, TCR β-chain sequencing has often been used to infer T cell antigen binding specificity due to its higher combinatorial potential compared to α-chains. To quantitatively evaluate the contribution of TCR α- and β-chains to predicting TCR-pMHC interactions, α-chains or β-chains were used as inputs to a CNN-based classifier instead of paired αβ chains. The performance of paired αβ chains is better than that of α- or β-chains alone, with an average 16% increase in AUC ( Fig. 9c Disproportionate α- and β-chain contributions to the prediction of TCR-pMHC-specific recognition were observed. For example, the contribution of the β-chain was dominant in the A*02:01_GILGFVFTL_Flu-MP_Influenza repertoire, whereas the α-chain was more important in the prediction of A*11:01_AVFDRKSDAK_EBNA-3B_EBV and A*02:01_ELAGIGILTV_MART-1_Cancer-specific binders ( Fig. 9c Similarly, different levels of conservation of TCR VJ gene usage were observed between the α-chains and β-chains of these seven pMHC repertoires ( Fig. 9d In addition, V gene usage was generally more conserved in the α-chain than in the β-chain, excluding the dominant TRBV19 usage in the A*02:01_GILGFVFTL_Flu-MP_Influenza repertoire, which may partially explain the unbalanced classification performance between the α-chain and the β-chain. Again, these results comprehensively demonstrate the importance of αβ pairing for accurately inferring TCR-pMHC interactions.

[0136] To further understand the conserved TCR sequence features underlying the classification, motif conservation of CDR3 amino acid sequences was explored from the 10 most predictive TCR sequences for each of these 7 pMHC repertoires ( Fig. 9e). Aligned with VJ gene usage, motif conservation is generally more evident in α-chain CDR3 than in β-chain CDR3( Figs. 9e and 9d For the four pMHC repertoires in which VDJdb also possesses the CDR3 amino acid motif, the motifs identified from 10x Genomics data are similar to the motifs from VDJdb ( Fig. 9e and Fig. 17a Taken together, the results indicate that pMHC-specific TCRs identified from high-throughput datasets are likely reliable binding partners, and that CNN-based models can capture key conserved TCR sequence features.

[0137] v. Immunophenotype of pMHC-bound CD8+ T cells.

[0138] Combined information on antigen specificity and T cell phenotype has been reported to be critical for the clinical success of immunotherapies, such as vaccination. Multi-omics data generated by the 10x Genomics immune profiling platform enables the association of T cell antigen specificity with various T cell phenotypes. Using gene (single-cell RNA-seq) and surface protein (CITE-seq) expression levels from these multi-omics datasets, pMHC-bound CD8+ T cells were separated into subpopulations (Methods and Fig. 18 Then, the identified subpopulations were annotated according to the previously described 32 CD8+ T cell subtype marker genes: untreated cells (CD45RA+CD45RO-CD62LhiCD127hi), central memory cells (Tcm, CD45RA-CD45RO+CD62L+), T effector memory cells (Tem, CD45RA-CD45RO+CD62L-), peripheral memory cells (Tpm, CD62L+CD127hi), terminal differentiation effector cells (Temra, CD45RA+CD45RO-CD127loGZMBhi), and other memory cells (CD43loKLRG1hiCD127) ( Figures 10a and 10b ).

[0139] 98.6% of pMHC-binding T cells were memory cells enriched from extended T cell clones ( Fig. 10d ), indicating that these T cells were selected by a specific immune response and are therefore likely to be reactive and reliable binders. Most of these memory T cells bound to common viral epitopes (e.g., influenza, EBV, CMV), and CD8+ pMHC-bound T cells from each donor exhibited different distributions of memory cell subsets. For example, Donor 1 had mainly Tpm and Tcm cells, while Donor 2 had Tem and Tpm cells, and Donor 4 had mostly Temra cells ( Figs. 10c and 10d ).

[0140] Most pMHC-binding T cells expressed a memory phenotype, but 1.3% of them were untreated cells. These untreated cells had more diverse pMHC interactions than untreated cells and often bound to endogenous antigens, tumor-associated antigens (e.g., MART-1), or antigens derived from viruses known to be seronegative of the donor (e.g., HIV). Fig. 10c and Fig. 20 Interestingly, the proportion of untreated T cells with cross-HLA type binding was significantly higher than that of untreated cells ( Fig. 10e These results indicate that the healthy donor T cell repertoire, particularly untreated cells, may react to antigens they have not yet encountered or are rare, and possess cross-reactivity. Further testing is required to evaluate whether these cells can mount functional T cell responses.

[0141] 2. argument

[0142] A method (ICON) capable of identifying reliable TCR-pMHC interactions was developed by significantly increasing the signal-to-background ratio in highly multiplexed 10x Genomics TCR-pMHC binding data. To accurately estimate background noise, a factor proven essential for reliably identifying TCR-pMHC binding events, it is essential to have appropriate controls (negative control dextramers and dextramer unclassified T cell samples). Although ICON was developed from a single dataset consisting of a single pool of multiplexed dextramers, this method can be generalized to query pMHC-TCR binding data from a wider range of pMHC dextramer pools as more multiplexed datasets are generated.

[0143] In this study, the robustness of this CNN-based classifier was demonstrated in predicting TCR-pMHC specific binding, indicating that such computational prediction can potentially be used to virtually study T cell antigen-specific recognition (compared to experimental comparisons). Immunological monitoring of T cell antigen-specific recognition has been applied to determine possible correlations between immune responses to specific antigens (e.g., tumor-specific antigens and peptide vaccines) and clinical outcomes in patients undergoing immunotherapy. However, experimentally mapping TCR sequences to antigen specificity is costly and labor-intensive. With appropriate training data for specific pMHCs, the classifier presented herein can assign pMHC binding probabilities to each TCR sequence of interest without performing binding tests. In this study, the multinomial prediction mode of this classifier was validated ( Fig. 17b ), and potentially used to select highly specific TCRs for safe T cell-related therapy.

[0144] The results indicate that a large portion (>30%) of TCRs binding to specific pMHCs share a single chain and differ in the second chain, emphasizing that T cell clonality should be determined by data with paired αβ chains. Additionally, 8% of these single-chain-shared TCRs can bind to different pMHCs. This is consistent with the predictive power of TCR antigen specificity using paired TCR chains, which is 16% greater than using only one chain. Therefore, single-cell paired αβ chain sequencing has the potential to be more powerful for accurately investigating T cell repertoire clonality and TCR-pMHC binding specificity.

[0145] The ability to assess biologically relevant T cell responsiveness is important for investigating and monitoring immune responses to pathogens and other disease states. We observed that the majority of recovered T cell responsiveness (98.6%) matched appropriate HLA types / supertypes, and that the phenotype of multimer-positive cells was mostly restricted to the memory T cell compartment, indicating that related memory responsiveness from previous functional T cell responses can be addressed by this technology. Paired αβ TCR sequencing revealed multiple TCR sequences specific to individual multimers, enhancing broad antigenic immune responses to common viral attacks.

[0146] Although low-grade HLA mismatch reactivity was recovered, these were significantly abundant in unexpanded, untreated T cells compared to the memory subset, potentially revealing antigen-specific interactions for previously unexposed targets or targets that did not peak in functional T cell responses. Additionally, the range of TCR binding intensity is expected to be recovered in these experiments, which may contribute to the detection of unexpected binding patterns. Dextramers are highly multimeric and have the potential to detect a wider range of TCR binding intensity than traditional tetrameric reagents. Furthermore, since the range of fluorescent dextramer intensity was categorized by multimeric-positive gating, low-frequency, low-intensity TCR interactions were captured in this high-sensitivity single-cell assay.

[0147] 3. method

[0148] i. 10x Genomics Single-Cell Immunological Profiling Dataset

[0149] The 10x Genomics data used in this study was downloaded from support.10xgenomics.com / single-cell-vdj / datasets.

[0150] ii. Single-cell RNA-seq data QC

[0151] CD8+ cells from each donor were selected for downstream analysis according to the following criteria: number of RNA features detected per cell <= 2500 and > 200 genes, and mitochondrial percentage less than 40% of the total number of UMIs (Unique Molecular Identifiers).

[0152] iii. pMHC-linked T cell classification

[0153] The Seurat V3 single-cell sequencing analysis R packages 33 and 34 were used for classification analysis based on single-cell RNA-seq data. Since a significant enrichment of TCR VJ gene usage was observed in identified pMHC-binding T cells, TCR genes were excluded from classification. Therefore, cell clusters would not be dominated by their shared VJ gene usage. Then, all other gene expressions in the identified binding T cells were normalized and scaled using Seurat V3 default parameters. PCA was performed on the number of transformed UMIs normalized on variably expressed genes. The top 10 PCs were used for cell classification. UMAP was used for classification visualization ( Fig. 17 ).

[0154] iv. Generation of CDR3 motifs from the most predictive pMHC-coupled TCR pairs

[0155] The CDR3 amino acid sequences of the α and β chains from the 10 most predictive TCRs were aligned using COBALT (www.ncbi.nlm.nih.gov / tools / cobalt / cobalt.cgi). The aligned CDR3 amino acid sequences were input into WebLogo35 with default parameters to generate motifs.

[0156] v. Selection of reported pMHC-specific binding pair TCRs

[0157] Raw files were downloaded from VDJdb28 (vdjdb.cdr3.net / ) and the pathology-related TCR database36 (friedmanlab.weizmann.ac.il / McPAS-TCR / ). Data were processed to obtain pMHC TCR pairs according to the following criteria: for VDJdb, paired α- or β-chain CDR3 amino acid sequences were required for each "complex.id"; TCRs annotated with "source" were removed from 10x genomics; and data were filtered for "species" = "human". For McPAS-TCR, known "Epitope.ID" were required for the entire data and contained "CDR3.alpha.aa" and "CDR3.beta.aa"; similarly, for VDJdb, human TCRs were filtered.

[0158] vi. Normalization of TCR-pMHC binding data

[0159] We developed an integrated CONtext-specific normalization (ICON) method. This method takes multi-omics single-cell sequencing data generated from the 10x Genomics Immune Map platform as input and performs TCR-pMHC binding specificity data normalization to identify reliable binding events. The multi-omics dataset includes single-cell RNA-seq, paired αβ-chain single-cell TCR-seq, dCODE-Dextramer-seq, and cell surface protein expression sequencing—also named CITE-seq (Cellular Indexing of Transcriptomes and Epitopes by Sequencing). ICON includes the following key steps ( Fig. 6a and Fig. 12 ):

[0160] Single-cell RNA-seq-based filtering of low-quality cells. This filters out low-quality cells, such as duplexes and dead cells. Cells with an unexpectedly high number of genes for detected T cells (e.g., > 2500 genes per cell) were classified as duplexes, and cells with a high fraction of mitochondrial gene expression (e.g., ratio of mitochondrial gene expression UMI to total gene expression UMI > 0.4) or too few detected genes (< 200 genes per cell) were classified as dead cells. Fig. 12a ).

[0161] Single-cell dCODE-Dextramer-seq-based background adjustment. There are two types of background noise controls designed for the dextramer binding assay and used in the analysis: one is negative control dextramer derived from dextramer-stained and sorted CD8+ T cells (n = 6) (indicated as NC_dex, nc), and the other is dextramer-stained CD8+ T cells without sorting for dextramer (indicated as Dex_unsorted, du). To examine signal and noise distributions, the maximum dextramer signal within the Unique Molecular Identifier (UMI) of each cell was selected to represent the best binding of each cell. Specifically, the non-specific dextramer binding signal of the cell is Max ( nc 1,..., nc Indicated by 6), the maximum dextramer signals of the six negative control dextramers included the dextramer pool. The dextramer binding signals of cells from dextramer-stained and sorted samples (Dex_sorted, ds Indicated by ), Max(which is the maximum dextramer signal at UMI among 44 test dextramers ds 1,..., ds 44 It is indicated as ). Similarly, the dextramer binding signal of cells from Dex_unsorted samples is Max( du ,..., du 44 It is indicated as ). The distribution of these three types of dextramer signals prior to the ICON process is Fig. 12b It is shown in the upper panel. P of the non-specific dextramer binding signal within UMI 99.9 (Excluding absolute outliers of negative dextramer control) was selected as a non-specific dextramer binding cutoff for each donor.

[0162] To estimate the potential noise introduced by the cell sorting process, the cumulative distribution of dextramer binding signals was compared between Dex_sorted and Dex_unsorted samples to determine the cutoff for dextramer sorting efficiency ( Fig. 12c For each donor, Kolmogorov-Smirnov test (KS test) p-values ​​were calculated by comparing the cumulative curves of dextramer-classified and dextramer-unclassified samples using each data point (dextramer UMI) as a sliding window. The S-shaped decreasing p-value curve indicates the abundance of dextramer binding signals in dextramer-classified samples compared to dextramer-unclassified samples, whereas the V-shaped curve suggests a loose cell classification gate ( Fig. 12d The dextramer UMI defining the largest difference in dextramer binding signals between Dex_sorted and Dex_unsorted(argmax D_(s,u) ) was used as a threshold to estimate the dextramer classification efficiency for V-shaped samples. Finally, the background noise of the dextramer-classified samples was defined as follows:

[0163] d=Max(P 99.9 , argmaxDs,u )

[0164] The dextramer signal (UMI) for each of the 44 test dextramers of the classified cells was corrected by subtracting the estimated background ( Fig. 12e ):

[0165] E c = E s - d

[0166] Then, cell-specific normalization was performed based on the Log-Rank distribution for each cell. pMHC-wide normalization was performed to make the dextramer binding signals comparable. The adjusted dextramer binding signal E_c of the classified cells was normalized across 44 test dextramers and then normalized across all cells using the following equation. E_c^' >= 0.9 was empirically selected as the cutoff for pMHC-specific binders ( Fig. 12f ).

[0167]

[0168]

[0169] T cells with single-paired αβ chains were selected based on single-cell TCR-seq. T cells with only α-chains, only β-chains, and multiple α- or β-chains were eliminated. In this study, only T cells with single-paired αβ chains were used.

[0170] The ICON normalization process was performed separately for each donor.

[0171] vii. Antigen-specific T cell expansion and antigen re-exposure to identify MART-1 binding T cells

[0172] Peripheral blood mononuclear cells (PBMCs) from HLA A*02:01 individuals were isolated by Ficoll-Paque Plus gradient isolation. PBMCs were seeded into culture plates of T cell medium (CellGenix dendritic cell medium, cat#20801-0500 + 5% human serum AB (Sigma, cat#H3667)) + 1% penicillin / streptomycin / L-glutamine (ThermoFisher, cat#10378-016), and T cells were supported with cytokines IL-7 and IL-15 at 5 ng / ml (CellGenix, cat# 1410-050 and 1413-050, respectively), IL-2 at 10 U / ml (Peprotech, cat# 200-0), and the A*02:01-restricted MART-1 epitope ELAGIGILTV at 10 µg / ml (Genscript). Fresh medium and cytokines were supplied to the cultures every 2 days for 1 week. On day 7 of culture, cells were stained with fluorescently tagged dextramer HLA-A*02:01 MART-1 ELAGIGIILT (Immudex, cat#WB2162-PE), and antigen-specific CD8+ T cell expansion was evaluated by flow cytometry. For antigen re-exposure analysis, a peptide was added to the T cell expansion culture 7 days after expansion. At 24 hours after restimulation, cells were collected and stained with fluorescently labeled antibodies for CD3 (BD ​​Biosciences, cat#612750), CD8 (BD Biosciences, cat#612889), CD69 (BD Biosciences, cat#564364), CCR7 (Biolegend, cat#353218), CD45RO (Biolegend, cat#304238), CD137 (Biolegend, cat#309828), and CD25 (Biolegend, cat#356104).Using the Astrios cell sorter (Beckman Coulter), fluorescence-activated cell sorting (FACS) gating was set up on forward scatter plots, side scatter plots, and fluorescence channels to select live cells while excluding debris and duplexes. Single CD3+CD8+CD45RO+CD137+ cells were sorted for further processing using a 100 μm nozzle.

[0173] Then, the classified cells were loaded onto a Chromium Single Cell 5' Chip (10x Genomics, cat#) and processed using a Chromium Controller to generate GEM (Gel Bead in Emulsion). The RNA-Seq library was prepared using the Chromium Single Cell 5' Library & Gel Bead Kit (10x Genomics, cat#) according to the manufacturer's protocol.

[0174] viii. Regeneron Oligo-Tagged Dextramer Staining and Classification for 10x Genomics Donor 3 and Donor 4

[0175] 10x Genomics provided cryopreserved Donor 3 and Donor 4 PBMCs for use in re-evaluating CD8+ T cell dextramer binding ability. CD8+ T cells were enriched using Miltenyi CD8+ T cell negative enrichment (Mitenyi). Then, the cells were incubated with benzonase (Millipore) and dasatinib (Axon) for 45 minutes, followed by oligo-tagged dextramer pool (Immudex, Fig. 21Cells were stained with ) at room temperature for 30 minutes. Then, cells were stained with fluorescent labels for CD3 (BD ​​Biosciences, cat#612750), CD4 (BD Biosciences, cat#563919), CD8 (BD Biosciences, cat#612889), CCR7 (Biolegend, cat#353218), and CD45RO (Biolegend, cat#304238), and stained with CITE-seq on ice for an additional 30 minutes. Using an Astrios cell sorter (Beckman Coulter), fluorescence-activated cell sorting (FACS) gating was set up on forward scatter plots, side scatter plots, and fluorescence channels to select viable cells while excluding debris and diploids. Single CD3+CD8+dextramer+ cells were sorted for further processing using a 100 μm nozzle ( Fig. 11 ).

[0176] TCR sequence similarity distance-based classification recently reported TCRdist, a weighted Hamming distance-based method for predicting TCR-pMHC binding specificity based on the sequence space of the TCR CDR region guided by structural information regarding pMHC binding. To measure receptor density within repertoires, nearest-neighbor (NN) distances (the average TCRdist between a receptor within a repertoire and its nearest receptor) were additionally calculated. For each pMHC repertoire, a binder was defined as a TCR that binds to a given pMHC. The NN-distance between each binding TCR and each set of pMHC binders from which the given TCR was removed was calculated. The NN distances were separated based on the known specificity of each TCR. Receiver Operational Characteristic (ROC) curves and the Area Under the ROC Curve (AUC) were calculated for the binary classifiers of each pMHC using the plotROC R package38. Simply put, ROC curves were generated for each classifier by calculating sensitivity and specificity at various NN distance thresholds—classifying the TCR as a pair for a given pMHC when their NN distance fell below a given threshold.

[0177] ix. CNN-based classification

[0178] A weighted binary classifier was adopted based on a deep learning framework, which includes three main steps adjusted to accommodate specific requirements.

[0179] x. Input data formatting

[0180] TCR sequencing files were collected from 10x Genomics as raw CSV format files. The sequencing files were parsed to obtain the amino acid sequence of CDR3 after removing unproductive sequences. Clones having different nucleotide sequences but identical matching amino acid sequences from CDR3 and the V, D, and J genes were aggregated together under a single TCR. Therefore, each TCR record used herein contains single paired α and β TCR amino acid sequences of the CDR3, V, and J genes. For the model run with only the α-chain, the TCRB-CDR3 amino acid sequence and the β-chain gene were removed from the input. A similar removal was performed for the model run with only the β-chain.

[0181] xi. Data conversion

[0182] Each TCR-CDR3 amino acid sequence was numerically encoded to represent 20 possible amino acids. Only sequences compliant with IUPAC (International Union of Pure and Applied Chemistry) amino acids were retained. For TCRs of different lengths, zero-padding was applied up to a length of 40. Trainable embedded layers were used to further extract feature regions from the amino acid sequences. V and J genes were one-hot encoded to provide categorical and discrete representations of gene names in numerical space. The encoded sequences and gene names were concatenated to represent a single TCR record. This data transformation process was applied before training all networks.

[0183] xii. Single TCR sequence classifier

[0184] This method was applied by providing a general conventional neural network architecture for training TCRs and focusing on sample or repertoire-level prediction. The focus was on optimizing single TCR sequence prediction. To achieve this, T cell clone sizes were removed from the input data. Additionally, a single translational invariant layer was applied to the sequence, followed by three fully connected synthetic layers in the final output layer. The network was trained using the Adam Optimizer (learning rate = 0.001) to minimize the cross-entropy loss between the soft-maxed-logits and one-hot cryptographic representations of the network's discrete categorical outputs. This approach was modified using a biologically meaningful kernel size of 439 to capture latent motifs. To account for imbalanced class representations in the training data, a weighted cross-entropy loss function was applied using the following formula:

[0185]

[0186] w c is a weight calculated using the inverse frequency of TCR sequences for each class. C represents one class; n c is the total TCR within a class; n is the total number of TCRs; , represents the predicted actual class for each TCR sequence.

[0187] Monte Carlo cross-validation (MCCV) training was performed with a fixed number of TCRs for validation and testing, respectively. An early stopping algorithm was implemented using the validation group of the sequences. Here, Monte Carlo sampling was repeated 20 times. The receiver operating characteristic (ROC) curve for the sequence classifier was computed based on the test set after averaging all MCCV predictions.

[0188] B. Example 2

[0189] 1. result

[0190] i. Identification of pMHC-specific binding TCRs from high-throughput binding data

[0191] 10x Genomics recently generated an extensive and publicly available TCR-pMHC binding dataset. In their initial report, binding profiles of over 150,000 CD8+ T cells derived from four HLA haplotype healthy donors (see Table 1, Donors 1 through 4) were evaluated across 44 pMHC dextramers by simultaneously sequencing T cell αβ chain pairs and transcripts, while using a single-cell-based immunoprofiling platform to directly detect antigen binding to T cells (see Figure 2). The dextramer pool consists of epitopes with known common viral and oncological responses across eight HLA alleles (see Table 2).

[0192] [Table 1] Information on T cell donors used in this study

[0193]

[0194] [Table 2] List of dCODE dextramer reagents used in the study.

[0195]

[0196]

[0197]

[0198] Described herein is a highly multiplexed dextramer binding dataset generated at the single-cell level with paired T cell α- and β-chain sequences. 10x Genomics identified pMHC-binding TCRs by applying a total cutoff for background noise and non-specific dextramer binding to all donors and dextramers (18). Surprisingly, an unexpectedly large number of indiscriminate TCR-pMHC binding events were found to be provided by 10x Genomics ( Fig. 24 ICON was developed to robustly identify reliable coupling events from such high-throughput TCR-pMHC coupling data ( Fig. 25a , Figs. 26a-d and Materials and Methods). The ICON data processing is performed in donor, cell, and dextramer-specific contexts. Briefly, good quality cells (live and unicellular) were selected using single-cell transcriptome data. Then, background binding noise for each donor was empirically estimated using negative control dextramers (n = 6). Subsequently, the raw dextramer binding signal was corrected by subtracting the background noise individually estimated for each donor. Since previous studies have shown that αβ binding synergistically induces TCR-pMHC recognition, T cells with paired αβ chains were selected as candidates for pMHC-binding T cells. The T cell dextramer binding signal was further corrected by penalizing dextramers that bind simultaneously to the same T cell / clone. Finally, the dextramer binding signals were normalized across cell and pMHC to make them directly comparable ( Fig. 25a , Figs. 26a-d and methods). To evaluate ICON performance, the pMHC binding specificity of CD8+ T cells from another healthy donor (Donor V) was evaluated using the same dextramer panel ( Fig. 27and Materials and Methods). ICON was able to link 91% of sequenced T cells with paired βαβ chains to their antigenic targets. To estimate the specificity of ICON, 21 individual dextramer binding assays were performed using T cells from the same donor, Donor V ( ee and materials and methods). Flow cytometry results show that they are consistent with the relative abundance of T cells binding to these 21 dextramers identified from ICON ( Fig. 25c ).

[0199] Applying ICON identified a total of 53,062 CD8+ T cells belonging to 5,721 unique T cell clones binding to 37 pMHCs from 5 donors ( Fig. 25b , Fig. 29 The dual specificity (specificity versus degeneracy) of TCRs has been proposed as an important feature of the immune response mechanism that sufficiently distinguishes itself from foreign peptides to avoid autoimmune reactivity while maintaining broad antigen coverage. Indeed, 99.6% of native TCRs bind to one specific pMHC, and the remaining TCRs interact with two pMHCs ( Fig. 25b Furthermore, these TCR-pMHC interactions generally follow HLA type-specific patterns. 94% of binding events are HLA-matched, of which 6% are associated with cross-recognition between HLA A*03-supertype family members HLA A*03:01 and A*11:01, which share similar major anchor sites of the presented peptide. Donors 1 and 2, possessing the most common HLA haplotype (A*02:01) in the dextramer pool (Tables 1 and 2), share a significant portion (n = 44) of the intrinsic TCR-pMHC interactions ( Fig. 25d, Fig. 25g), this supports the possibility that the TCR-pMHC binding pattern is most likely HLA-restricted. However, 6% of binding events are cross-HLA type interactions. HLA-type mismatched bound T cells tend to have smaller clones or be unicellular (antigen-untreated).

[0200] Of all pMHC-binding TCRs, 99% of the total TCRs (96% of native TCRs) bind to 9 pMHCs: B*08:01_RAKFKQLL_BZLF1_EBV (T cell #:18,468 / native TCR #:479), A*02:01_GILGFVFTL_Flu-MP_Influenza (T cell #:8,365 / native TCR #:1,095), A*11:01_IVTDFSVIK_EBNA-3B_EBV (T cell #:5,438 / native TCR #:149), A*03:01_KLGGALQAK_IE-1_CMV (T cell #:3,899 / native TCR #:2,865), A*11:01_AVFDRKSDAK_EBNA-3B_EBV (T Cells #:1,579 / Intrinsic TCR #:95), A*02:01_GLCTLVAML_BMLF1_EBV (T cells #:1,886 / Intrinsic TCR #:117), A*02:01_ELAGIGILTV_MART-1_Cancer (T cell count:297 / Intrinsic TCR #:293), B*35:01_IPSINVHHY_pp65_CMV (T cell count:6,986 / Intrinsic TCR #:280) and A*02:01_NLVPMVATV_pp65_CMV (T cells #:5,612 / Intrinsic TCR #:164)( Fig. 25eTo further understand the conserved TCR sequence features that form the basis of classification, the use of TCR VJ genes for these nine pMHC repertoires was investigated. In addition to the abundance reported by previous studies, such as TRBV19 and TRAV27 in the influenza repertoire, TRAV5 and TRBV20-1 in the BMLF1_EBV repertoire, and TRBV6-5 in NLVPMVATV_pp65_CMV, TRAV12-2 in the MART-1_Cancer repertoire, TRAV21, TRAV35, TRBV11-2, and TRBV6-6 in the IVTDFSVIK_EBNA-3B_EBV repertoire, TRAV8-3, TRAV13-1, and TRBV28 in AVFDRKSDAK_EBNA-3B_EBV, TRAV13-1, TRAV13-2, and TRBV12-3 in the BZLF1_EBV repertoire, TRAV12-1, TRAV41, and TRBV2 in IPSINVHHY_pp65_CMV, and TRBV20-1, and TRAV23 / D6 and TRBV12-4 in NLVPMVATV_pp65_CMV were found ( Fig. 25f Consistent with the use of conserved VJ genes, the Shannon diversity index and TCR clone size distribution suggested that each pMHC-binding T cell repertoire experiences different degrees of expansion in responding to their target peptides ( Figs. 30a & b ).

[0201] ii. TCRAI: Neural network classifier of T cell antigen specificity

[0202] As large and diverse TCR-pMHC binding events are identified, powerful functional classifiers are required to rapidly validate these binding events. Recent studies have demonstrated that neural networks can learn high-dimensional information from TCR sequences and thus can strongly predict TCR-pMHC binding.

[0203] TCRAI, a Python package, was developed using Tensorflow 2 and provides a flexible framework for TCR-pMHC specificity studies. Fig. 31a Using the highly modular TCRAI package allows for easy customization of the model's architecture. Briefly, the TCRAI framework works as follows: An arbitrary number of V(D)J genes and the CDR regions of the TCRs can be defined as inputs to the model in their text form. Then, through "Processor" objects that convert the text into numerical representations, one can choose how to process these inputs numerically in a non-trainable manner. These numerical inputs can then be further processed in a trainable manner through "Extractor" objects that form blocks of the neural network and provide output vector representations of the input data, referred to as fingerprints. These fingerprints are linked to a single TCRAI fingerprint that describes these input TCRs via a single numeric vector. Then, these TCRAI fingerprints are passed through "Closer" objects that form the final blocks of the neural network architecture to generate predictions for the input TCRs. The TCRAI package provides several such pre-built processors, extractors, and closures. This also enables the performance of binary, multinomial, regression, or other operations by choosing to configure different closure objects.

[0204] To evaluate the performance of TCRAI, a literature search was conducted on currently available methods (Table 3), and the classifier was compared with four major methods in the field: GLIPH2, DeepTCR, NetTCR, and TCRdist. For comparison, an eight-cell repertoire of pMHC-specific binding T cells was collected as the gold-standard dataset, consisting of at least 50 unique paired αβ-chain TCRs generated by traditional monomer binding or antigen re-exposure assays (Table 4 and Materials and Methods). Three of the methods—DeepTCR, NetTCR, and TCRdist—are predictive models, similar to TCRAI. The area under the ROC (Receiver Operator Attribute) curve (AUROC / AUC), a standard measure of classification success for these predictive models, indicates that TCRAI and DeepTCR, which use similar neural network frameworks, outperform TCRdist and NetTCR. Overall, TCRAI demonstrates more consistent and better performance than DeepTCR ( Fig. 31e and Fig. 32b Since GLIPH2 is designed to cluster TCR sequences into distinct groups of shared specificity, the sensitivity and specificity of these four prediction models (calculated at a model threshold that maximizes the geometric mean of the two) were measured for comparison with GLIPH2. The comparison results demonstrated that TCRAI possesses the best balance of sensitivity and specificity. Fig. 33 Several methods with purposes different from those of TCRAI were not included in the comparison. For example, ALICE is intended to detect homologous / extended TCR populations. TcellMatch uses cell-specific covariates (e.g., gene expression) as input but not TCR sequences alone, and its performance was tested on 10x Genomics Immune Map data with a high noise-to-signal ratio without further processing.

[0205] [Supplementary Table 3]Summary of the method for linking TCR-antigen specificity

[0206]

[0207] * TCRex: A web tool for academic, impersonal research only.

[0208] [Table 4] Summary of 8 pMHC repertoires collected from VDJdb and McPAS (Methods)

[0209]

[0210] iii. Classification of pMHC-binding TCRs identified from high-throughput data

[0211] Then, TCRAI was applied to the 9 most abundant pMHC binding repertoire ICONs identified from high-throughput data ( Fig. 25e TCRs from these 9 pMHC repertoires were classified with an average AUC of 0.88 using TCRAI in binary mode. Similar predictive performance was also observed using TCRAI in multinomial mode ( Fig. 34a and Fig. 35 (Hereafter, TCRAI results are derived from binary mode unless otherwise specified). Historically, TCR β-chain sequencing has often been used to infer T cell antigen binding specificity due to its higher combinatorial potential compared to α-chains. To quantitatively evaluate the contribution of TCR α- and β-chains to predicting TCR-pMHC interactions, α-chains or β-chains were used as inputs for TCRAI instead of paired αβ chains. The performance of paired αβ chains is better than that of α- or β-chains alone, with an average increase of approximately 0.2 in AUC ( Fig. 34bConsistent with previous studies, these results comprehensively demonstrate the importance of αβ pairing for accurately inferring TCR-pMHC interactions. Predictive performance for the β-chain is not always better than that for the α-chain, which indicates the importance of the α-chain in TCR-pMHC-specific recognition, which has often been previously overlooked.

[0212] To further validate the performance of TCRAI, four pMHC repertoires coupled to TCR from selected public datasets (A*02:01_ELAGIGILTV_MART-1, A*02:01_GILGFVFTL_Flu-MP, A*02:01_GLCTLVAML_BMLF1_EBV, and A*02:01_NLVPMVATV_pp65_CMV) were also used. TCRAI was trained using four repertoires identified from high-throughput datasets to predict four selected repertoires. Fig. 34c This shows that the prediction results are generally similar to the performance on the training set. However, when inferring A*02:01_NLVPMVATV_pp65_CMV, TCRAI's performance was significantly worse than the other three pMHCs. To understand the performance difference, the model's TCRAI fingerprint space was examined (Materials and Methods). A*02:01_ELAGIGILTV_MART-1_Cancer, and the other two pMHCs ( Fig. 36a In the case of ), TCR combinations from high-throughput datasets and selected datasets spatially overlap in fingerprint space, whereas the overlap is significantly worse in the case of pp65_CMV( Fig. 34d and Fig. 36b This poor overlap is attributed to a 98.2% pp65_CMV combined TCR in high-throughput datasets from a single donor ( Fig. 29), while this represents a small subspace of possible binding TCRs, the open data contains TCRs from various donors representing a larger range of TCR spaces. This result also highlights the importance of large and diverse datasets for training robust TCR-antigen prediction models.

[0213] iv. Analysis of pMHC-specific TCR characteristics

[0214] To investigate the characteristics of TCRs binding to a given pMHC, we analyzed how the TCRAI classifier model arranges TCRs within its fingerprint space (Materials and Methods). TCR fingerprints from the classifier model enable the discovery of specific groups of TCRs possessing conserved gene usage and CDR3 motifs. These groups often exhibit different binding capabilities and diverse structural binding patterns.

[0215] Clustering TCR with A*02:01_GILGFVFTL_Flu-MP_Influenza yields two well-separated clusters in the TCRAI fingerprint space ( Fig. 37a The constructed α- and β-CDR3 motifs and gene usage indicate that Cluster 0 has strongly conserved xRSx motifs and TRB19 and TRAJ42 gene usage in the β-chain, while the smaller group Cluster 1 has highly conserved gene usage TRBV19 / TRBJ1-2 / TRAV38-1 / TRAJ52 ( Fig. 37c The dextramer signal distribution (in UMI, unique molecular identifiers) showed that TCRs in Cluster 0 have a stronger binding to Flu dextramer than TCRs in Cluster 1 ( Fig. 37bThe results are consistent with the well-known strong conservation of the CDR3 motif and TCRBV19 gene utilization in A*02:01_GILGFVFTL_Flu reactive T cells, which were thought to be linked to "featureless" pMHC complexes. Further comparison with the recently identified classes of A*02:01_GILGFVFTL_Flu binding TCRs revealed that clusters 0 and 1 were linked to their groups I (typical) and II (novel), respectively. Additionally, it was found that group I TCRs exhibit stronger binding than group II TCRs. The 3D structures of the proposed TCR-pMHC binding complexes suggest that, due to highly conserved motifs / residues, the two groups of these TCRs possess different binding modes, which leads to different Phe-5 ring rotations of the Flu peptide in these two complexes ( Fig. 37d ).

[0216] The binding of TCRs to the other eight pMHCs was also characterized. The results for the A*02:01_GLCTLVAML_BMLF1_EBV binding TCR are particularly interesting. In previous studies, a dominant open TCR composed of TRBV20-1 / TRBJ1-2 / TRAV5 / TRAJ31 was observed. However, previous analyses of the TCR population binding to this pMHC focused on the TRAV5 TCR, which is significantly biased in the population. The current experiment unbiasedly identified five TCR clusters in the TCRAI fingerprint space ( Fig. 37e Clusters 1 and 2 were split into two clusters based on β-chain gene usage but represent the classic HLA*02:01_GLCTLVAML open TCR( 37gCluster 0 contains a β-chain CDR3 motif not presented in the TCR or elsewhere following gene use (TRBV2 / TRBJ2-2). TCRs belonging to this novel group exhibit different binding abilities to regular TCR clusters (clusters 1 and 2), as indicated by the reduced dextramer UMI numbers ( Fig. 37f ) and this indicates lower affinity, which will partially explain why these TCR groups have not yet received attention.

[0217] v. Immunophenotype of pMHC-bound CD8+ T cells

[0218] Combined information on antigen specificity and T cell phenotype has been reported to be critical for the clinical success of immunotherapy, such as vaccination. Multi-omics data generated by immune map platforms enables the association between T cell antigen specificity and T cell phenotype. Using gene (single-cell RNA-seq) and surface protein (CITE-seq, cell indexing by sequencing of transcripts and epitopes) expression from these multi-omics datasets, pMHC-bound CD8+ T cells were grouped into subpopulations ( Fig. 38a and Materials and Methods). Then, the identified subpopulations were annotated according to the aforementioned CD8+ T cell subtype marker genes: untreated cells (CD45RA+CD62LhiCD127hi), central memory cells (Tcm, CD45RA-CD62L+CD127+EOMEShighTBETlow), T effector memory cells (Tem, CD45RA-CD62LlowCD127+GZMB+), peripheral memory cells (Tpm, CD62L+CD127hiGZMB+), terminally differentiated memory cells (Temra, CD45RA+CD127loGZMBhi) and other memory cells (CD43loKLRG1hiCD127-) ( Figs. 38a & b ).

[0219] 96% of pMHC-coupled T cells were memory cells and were enriched in expanded T cell clones ( Fig. 38e % d This indicates that these T cells were selected by a specific immune response and are therefore likely to be reactive and reliable binders. Most of these memory T cells bound to common viral epitopes (e.g., influenza, EBV, CMV), and pMHC-bound T cells from each donor exhibited different distributions of memory cell subsets. For example, Donors 1 and 2 mainly had Tpm, while Donor V had Tem, and Donors 3 and 4 mostly had Temra cells ( Figs. 38c & d ).

[0220] Most pMHC-binding T cells expressed a memory phenotype, but 4% of them were untreated cells. These untreated cells had more diverse pMHC interactions than untreated cells and often bound to tumor-associated antigens (e.g., MART-1), endogenous antigens, or antigens derived from viruses known to be seronegative of the donor (e.g., HPV). Fig. 38c Interestingly, the proportion of untreated T cells with cross-HLA type binding was significantly higher than that of untreated cells ( Fig. 38f These results indicate that the healthy donor T cell repertoire, particularly untreated cells, may react to antigens they have not yet encountered or are rare, and possess cross-reactivity. Further testing is required to evaluate whether these cells can mount functional T cell responses.

[0221] 2. argument

[0222] High-throughput TCR-pMHC binding data presents an attractive pathway for enhancing the understanding of TCR antigen recognition. However, this type of data is often associated with a high noise-to-signal ratio. The present invention provides a framework of computational tools including ICON, a novel method capable of identifying reliable TCR-pMHC interactions by significantly increasing the signal-to-noise ratio in highly multiplexed TCR-pMHC binding data with good sensitivity and specificity. ICON computes noise-corrected dextramer signals in a parameterless manner, allowing pMHC-TCR binding data to be easily generalized from a wider range of pMHC dextramer pools and potentially extended to the normalization of protein binding signals in a single-cell space, such as CITE-seq.

[0223] In this study, the Python package TCRAI was developed, demonstrating the robustness of deep learning classifiers in predicting TCR-pMHC specific binding. Due to the importance of the CDR3 region in determining the specificity of a TCR for a given antigen, there is a temptation to build a predictive model that utilizes only this information, as is often the case. However, due to the use of highly conserved genes for many pMHCs, VJ gene usage was found to be a significant predictive factor for TCRAI, particularly when there are few unique pMHC-binding TCRs in the dataset. The predictive performance of models receiving CDR3 information outperformed gene-level-only models when there were at least 100 pMHC-binding TCR sequences or more ( Fig. 39 This indicates that the volume of such data for these models is required to extract useful sequence motifs from CDR3.

[0224] TCRAI was found to be capable not only of performing state-of-the-art classification of TCR-pMHC-specific binding but also of identifying groups of TCRs with different binding profiles. By partnering dextramer UMIs with TCR sequence information, it was possible to investigate the different binding capabilities between these groups. This finding indicates that as the volume of high-throughput TCR-pMHC binding data increases, the ability to discover new TCR motifs and pair them with UMIs as well as broader multi-omics data will grow. For example, the ability to investigate the different transcriptional regulation of T cell receptor signaling between TCR groups with different binding mechanisms is highly interesting for a wide range of scientific questions as well as for the development of T cell therapies.

[0225] T cell antigen-specific recognition can potentially be studied virtually (compared to experimentally) using TCRAI. Immunological monitoring of T cell antigen-specific recognition has been applied to determine immune responses to specific antigens (e.g., SARS-COV2, tumor-specific antigens, and peptide vaccines) and potential correlations with disease severity and clinical outcomes in patients receiving immunotherapy. However, experimentally mapping TCR sequences to antigen specificity is costly and labor-intensive. With appropriate training data for specific pMHCs, the TCRAI classifier presented herein can assign pMHC binding probabilities to each TCR sequence of interest without performing binding tests. In this study, the multinomial prediction mode of this classifier was validated ( Fig. 35 This means that it can be used to select highly specific TCRs for safe T cell-related therapies.

[0226] The ability to assess biologically relevant T cell responsiveness is important for investigating and monitoring immune responses to pathogens and other disease states. The majority of recovered T cell responsiveness (94%) matched appropriate HLA types / supertypes, and the phenotypes of multimer-positive cells were mostly restricted to the memory T cell compartment, indicating that related memory responses from previous functional T cell responses can be resolved by this technology. Paired αβ TCR sequencing revealed multiple TCR sequences specific to individual multimers, enhancing broad antigenic immune responses to common viral attacks.

[0227] Although low-grade HLA mismatch reactivity was recovered, these were significantly abundant in unexpanded, untreated T cells compared to the memory subset, potentially revealing antigen-specific interactions for previously unexposed targets or targets that did not peak in functional T cell responses. Additionally, the range of TCR binding intensity could be recovered in these experiments, which may contribute to the detection of unexpected binding patterns. Dextramers are highly multimeric and have the potential to detect a wider range of TCR binding intensity than traditional tetrameric reagents. Furthermore, since the range of fluorescent dextramer intensity was categorized by multimer-positive gating, low-frequency, low-intensity TCR interactions were captured in this high-sensitivity single-cell assay.

[0228] 3. Materials and Methods

[0229] i. 10x Genomics Single-Cell Immunological Profiling Dataset

[0230] The 10x Genomics data used in this study was downloaded from support.10xgenomics.com / single-cell-vdj / datasets.

[0231] ii. Identification of pMHC-binding T cell phenotypes

[0232] The Seurat V3 single-cell sequencing analysis R package was used for classification analysis based on single-cell RNA-seq data. Since a significant enrichment of TCR VJ gene usage was observed in identified pMHC-binding T cells, TCR genes were excluded from classification. Therefore, cell clusters would not be dominated by their shared VJ gene usage. Then, all other gene expressions in the identified binding T cells were normalized and scaled using Seurat V3 default parameters. PCA was performed on the number of normalized and transformed UMIs on variable-expressed genes. The top 10 PCs were used for cell classification. UMAP was used for classification visualization.

[0233] iii. Selection of reported pMHC-specific binding pair TCRs

[0234] Raw files were downloaded from VDJdb (42) (vdjdb.cdr3.net / ) and the pathology-related TCR database (friedmanlab.weizmann.ac.il / McPAS-TCR / ). Data were processed to obtain pMHC TCR pairs according to the following criteria: for VDJdb, a paired α- or β-chain CDR3 amino acid sequence was required for each "complex.id"; TCRs annotated with "source" were removed from 10x genomics; and "species" = "human" were filtered. For McPAS-TCR, known "Epitope.ID" were required from the entire dataset and had "CDR3.alpha.aa" and "CDR3.beta.aa"; similarly, for VDJdb, human TCRs were filtered.

[0235] iv. Normalization of high-throughput TCR-pMHC combined data

[0236] Reliable TCR-pMHC interactions were identified by developing ICON, an integrated text-specific normalization method. It takes as input multi-omics single-cell sequencing data generated from multiplexed multimer coupling platforms, such as the 10x Genomics Immunoma, which include single-cell RNA-seq, paired αβ-chain single-cell TCR-seq, dCODE-Dextramer-seq, and cell surface protein expression sequencing—also named CITE-seq. ICON includes the following key steps ( Fig. 25a and Fig. 26 ):

[0237] Step 1: Single-cell RNA-seq-based filtering of low-quality cells

[0238] This filters out low-quality cells, such as duplexes and dead cells. T cells with an unexpectedly high number of genes (e.g., > 2,500 genes per cell) were classified as duplexes, and cells with a high fraction of mitochondrial gene expression (e.g., ratio of mitochondrial gene expression to total gene expression > 0.2) or too few gene detections (< 200 genes per cell) were classified as dead cells. Fig. 26a ).

[0239] Step 2: Single-cell dCODE-Dextramer-seq-based background estimation

[0240] Six negative control dextramers were designed to estimate background noise from multiplexed dextramer binding assays. To examine signal and noise distributions, the maximum dextramer signal within the UMI (Unique Molecular Identifier) ​​of the negative control dextramers and the worst-case noise and best-case dextramer binding of each T cell were represented using the test dextramers for each cell. The density distributions of these two types of dextramer signals are Fig. 26b It appears in. Background cutoff( Fig. 26bThe gray dotted line) was empirically selected for each donor.

[0241] Step 3: Select T cells with paired αβ chains based on single-cell TCR-seq

[0242] T cells with only a single chain were removed. For detected T cells with multiple α- or β-chains, the cell with the highest UMI number was assigned to each T cell.

[0243] Step 4: Dextramer Signal Correction

[0244] Although each dextramer possesses optimal binding conditions in itself, it is impossible to arrange experimental conditions so that a multiplexed dextramer binding assay is optimal for all dextramers. This results in multiple dextramer bindings to the same T cell / clone, as observed in these high-throughput datasets. Fig. 26c To compensate for this effect, the following technique was used to penalize the dextramer signal when simultaneously binding to the same T cell / clone.

[0245] j 번째 Binding to dextramer i 번째 E ij If defined as, i 번째 Regarding T cells j 번째 The fraction of the dextramer signal resulting from dextramer binding is further expressed as follows:

[0246]

[0247] i 번째 TCR clonal type of T cells k i Low, and dextramer j clone type that combines with k i The number of T cells belonging to T_( k ij Represented by ), j 번째 Clone type that binds to dextramer k i The fraction of T cells belonging to is expressed as follows:

[0248]

[0249] Using these amounts, j 번째 Binding to dextramer i 번째 The corrected dextramer signal for T cells is calculated as follows:

[0250]

[0251] Step 5: Cell- and pMHC-specific dextramer signal normalization and binder identification

[0252] To make all dextramer binding signals similar, the corrected dextramer binding signals were log-denormalized across 44 test dextramers within the cell. Subsequently, pMHC-specific normalization was performed based on a log-rank distribution. A normalized dextramer UMI > 0 was empirically selected as a cutoff for pMHC-specific binders.

[0253] v. Regeneron Oligo-Tagged Dextramer Staining and Classification

[0254] CD8+ T cells were enriched from healthy donor PBMCs using Miltenyi CD8+ T cell negative enrichment (Mitenyi). Then, the cells were incubated with benzonase (Millipore) and dasatinib (Axon) for 45 minutes, and then stained with oligo-tagged dextramer pool (Immudex, see Table 2) at room temperature for 30 minutes. Then, cells were stained with fluorescent labels for CD3 (BD ​​Biosciences, cat#612750), CD4 (BD Biosciences, cat#563919), CD8 (BD Biosciences, cat#612889), CCR7 (Biolegend, cat#353218), and CD45RA (Biolegend, cat#304238), and stained with CITE-seq antibodies for an additional 30 minutes on ice. Using an Astrios cell sorter (Beckman Coulter), fluorescence-activated cell sorting (FACS) gating was set up on forward scatter plots, side scatter plots, and fluorescence channels to select viable cells while excluding debris and diploids. Single CD3+CD8+dextramer+ cells were sorted for further processing using a 100 μm nozzle.

[0255] vii. Building a neural network-based classifier TCRAI

[0256] While TCRAI provides a flexible framework for designing TCR classifiers, a specific and consistent architecture was used throughout this work, which is described in detail below. In addition to the flexible architecture, several key differences from the DeepTCR architecture are the use of 1D convolution and batch normalization for CDR3 sequences, and lower-dimensional representations of genes. These changes provide improved model regularization and enable the model to learn stronger gene associations.

[0257] To process the input information of the TCR into a numeric format, the following method was applied. For each CDR3 sequence, the amino acids are first converted into integers, and subsequently, these integer vectors are encoded for one-time use. For V and J genes, a dictionary of gene types in integers is constructed individually for each V and J gene, and these are used to convert each gene into an integer.

[0258] The neural network architecture applied to the processed input information includes an embedding layer and a convolutional network. Specifically, the processed CDR3 residues are embedded in a 16-dimensional space through learned embeddings, and the generated numeric CDR3s are fed through three 1D convolutional layers with filters of dimension, kernel width, and spacing. Each convolution is activated by exponential linear unit activation, followed by dropout and batch normalization. Following these three convolutional blocks, global max pooling is applied to the final feature layer, and this process encodes each CDR3 into a vector of length 256, the "CDR3 fingerprint." For each gene, the processed gene input is one-hot encoded through learned embeddings and embedded in a reduced-dimensional space (16 for V genes, 8 for J genes) to provide the "fingerprint" of each gene as a vector. All selected CDR3s and gene fingerprints are concatenated together into a single vector, the "TCRAI fingerprint." The TCRAI fingerprint passes through a single final fully connected layer to provide binomial prediction (single output value, sigmoid activation), regression prediction (single output, no activation), or multinomial prediction (multiple output values, softmax activation). Binomial and multinomial prediction are the focus of this work.

[0259] TCR sequencing files were collected from 10x Genomics as raw CSV format files. The sequencing files were parsed to obtain the amino acid sequence of CDR3 after removing unproductive sequences. Clones having different nucleotide sequences but identical matching amino acid sequences from CDR3 and the V, D, and J genes were aggregated together under a single TCR. Thus, each TCR record used herein contains a single pair of α and β TCR chains containing the CDR3 amino acid sequence and the V and J genes for each chain.

[0260] The data is split into training (76.5%), validation (13.5%), and left-out test sets (10%) for each model, followed by 5-fold Monte Carlo cross-validation (MCCV) on the training set. Models are trained using the Adam optimizer to minimize cross-entropy loss, which is weighted by 1 / (number of classes * fraction of samples in that class) for each class. On the left-out validation dataset, early stopping is introduced to prevent overfitting; if the validation loss increases for more than 5 time points, the model stops training, and the weights of the model with the minimum validation loss are restored. Since the number of models being trained is large, only the training rate and batch size are adjusted during cross-validation. After cross-validation, the optimally performing hyperparameters are selected, and the model is retrained on the entire training set using the validation set to control early stopping. Then, the retrained model is evaluated on the left-out test set.

[0261] viii. TCRAI fingerprint analysis

[0262] The TCRAI model generates both predictions for TCRs binding to a specific pMHC (or one of many pMHCs in the multinomial case) and numerical vector "fingerprints" describing the TCRs in the context of whether the TCRs can bind to that pMHC. To understand how the model works and to identify groups of TCRs with different binding patterns, the distribution of these fingerprints is analyzed. UMAP is used to reduce the fingerprints to a two-dimensional space. When using a model trained on one dataset to infer fingerprints for another unseen dataset, the UMAP projector can be fitted to the TCRs from the training dataset and used to transform the TCRs from the unseen set.

[0263] When clustering TCR fingerprints, the fingerprints of all TCRs in the dataset are projected into a 2D space as described above, and then TCRs with strong true values ​​(STP, binomial prediction >0.95) are selected. Then, these STPs are clustered in the 2D space using a k-means classifier. Then, the TCRs within each cluster are collected and used to construct CDR3 motif logos (weblogo use), gene-uses, and UMI distributions by pairing the unique TCR clones within the cluster with all repeated clones in the high-throughput data.

[0264] viii. DeepTCR variant

[0265] The DeepTCR method was applied to construct a binary classifier adjusted as described below.

[0266] For each TCR record, a single pair of α and β TCR chains was used, consisting only of the CDR3 amino acid sequence and the V and J genes for each chain, according to the input provided in the TCRAI package. That is, clonalization, MHC, or the use of the D gene were not included in the DeepTCR model. The final output layer was tuned to provide a single binomial output, and the model's hyperparameters were optimized for the problem at hand within the context of the DeepTCR framework.

[0267] Fig. 41 The network (4104) computing unit connected through (4101) (e.g., arithmetic unit (106) ) and server (4102) an environment including non-limiting examples of (4100) It is a block diagram illustrating [it]. In one aspect, any part or all of the described method may be performed in an arithmetic unit as described herein. Arithmetic unit (4101) is sequence data (104) (e.g., single-cell sequence data, dextramer sequence data, and single-cell receptor sequence data), training data (410) (e.g., labeled receptor sequence data), ICON module (108) , prediction module (110) It may include one or more computers configured to store one or more of the following. Server (1402) is sequence data (104) It may include one or more computers configured to store. Multiple servers (4102) is network (4104) through the arithmetic unit (4101) It can communicate with. In one embodiment, the server (1402) is a single-cell immune profiling platform (102) It may include a repository for data generated by.

[0268] computing unit (4101) and server (4102)In terms of hardware architecture, generally, a processor (4108) , memory system (4110) , input / output (I / O) interface (4112) , and network interface (4114) It may be a digital computer including. These parts ( 4108 , 4110 , 4112 , and 4114 ) is a local interface (4116) It is combined to enable communication through the local interface. (4116) ...may be, for example, one or more buses or other wired or wireless connections as known in the art, but are not limited thereto. Local interface (4116) It may have additional elements, such as a controller, buffer (cache), driver, repeater, and receiver, which are omitted for simplification to enable communication. Additionally, the local interface may include address, control, and / or data connections to enable proper communication between the aforementioned components.

[0269] processor (4108) is, especially a memory system (4110) It may be a hardware device for executing software stored in. Processor (4108) Any custom-made or commercially available processor, central processing unit (CPU), or arithmetic unit (4101) and server (4102) Among the various processors associated with it, it may be an auxiliary processor, a semiconductor-based microprocessor (in the form of a microchip or chipset), or any device generally for executing software instructions. Arithmetic Unit (4101) and / or server (4102) When it is operating, the processor (4108) is a memory system (4110) Execute software stored within, and the memory system (4110) Communicates data and, depending on the software, the computing unit (4101)and server (4102) It can be configured to roughly control the operation of.

[0270] I / O interface (4112) It may be used to receive user input from one or more devices or components and / or provide system output to them. User input may be provided, for example, via a keyboard and / or mouse. System output may be provided via a display device and a printer (not shown). I / O interface (41412) It may include, for example, a serial port, a parallel port, a small computer system interface (SCSI), an infrared (IR) interface, a radio frequency (RF) interface, and / or a universal serial bus (USB) interface.

[0271] Network interface (4114) is network (4104) arithmetic unit on (4101) and / or server (4102) It can be used to transmit and receive from. Network interface (4114) It may include, for example, a 10BaseT Ethernet adapter, a 100BaseT Ethernet adapter, a LAN PHY Ethernet adapter, a Token Ring adapter, a wireless network adapter (e.g., WiFi, cellular, satellite), or any other suitable network interface device. Network interface (4114) is network (4104) It may include address, control, and / or data connections to enable appropriate communication on.

[0272] The memory system (4110) may include any one or a combination of volatile memory elements (e.g., random access memory (RAM such as DRAM, SRAM, SDRAM, etc.)) and non-volatile memory elements (e.g., ROM, hard drive, tape, CDROM, DVDROM, etc.). Additionally, the memory system (4110) may incorporate electronic, magnetic, optical and / or other types of storage media. It is noted that the memory system (4110) may have a distributed architecture in which various components are located remotely from one another but can be accessed by the processor (4108).

[0273] memory system (4110) The software within may include one or more software programs, each of which includes a sequence of executable instructions for implementing a logical function. Fig. 41 In the example, the arithmetic unit (4101) memory system (4110 The software within ) is sequence data (104) , training data (410) , ICON module (108) , prediction module (110) , and a suitable operating system (O / S) (4118) It may include. Fig. 41 In the example of, the server (4102) memory system (4110) The software within is sequence data (104) , and a suitable operating system (O / S) (4118) It may include operating systems (4118) It essentially controls the execution of other computer programs and provides scheduling, input-output control, file and data management, memory management, and communication control and related services.

[0274] For example, applications and operating systems ( 4118Although other executable program components such as ) are illustrated in this specification as separate blocks, these programs and components are arithmetic units ( 4101 ) and / or server (4102) It is recognized as being able to reside in different storage components at various times. Training module (220) An implementation of the method may be stored on a computer-readable medium of a certain form or transmitted through it. Any disclosed method may be performed by computer-readable instructions implemented on a computer-readable medium. A computer-readable medium may be any available medium accessible by a computer. By example, but not to be limiting, a computer-readable medium may include "computer storage media" and "communication media." "Computer storage media" may include volatile and non-volatile, removable and non-removable media implemented by any method or technique for storing information such as computer-readable instructions, data structures, program modules, or other data. Exemplary computer storage media may include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital multi-purpose disc (DVD) or other optical storage device, magnetic cassette, magnetic tape, magnetic disk storage device or other magnetic storage device, or any other medium that can be used to store desired information and can be accessed by a computer.

[0275] In one embodiment, the ICON module (108) and / or prediction module (110) The method (4200) It can be configured to perform, Fig. 42 It is shown in. Method (4200) It can be performed wholly or partially by a single computing unit, multiple electronic devices, etc. Method (4200) The stage 4201The method may include the step of receiving single-cell sequence data, dextramer sequence data, and single-cell T cell receptor (TCR) sequence data. The single-cell sequence data may include RNA-seq data, the dextramer sequence data may include dCODE-Dextamer-seq data, and the single-cell T cell receptor (TCR) sequence data may include TCR-seq data.

[0276] method (4200) silver, stage 4202 In this, for each cell indicated in the dextramer sequence data, the method may include the step of determining the number of genes based on single-cell sequence data.

[0277] method (4200) The stage 4203 The method may include a step of removing data associated with cells having a number of genes outside a gene threshold range from the dextramer sequence data. For example, the gene threshold range may be about 200 genes to about 2,500 genes.

[0278] method (4200) silver, stage 4204 In this, for each cell indicated in the dextramer sequence data, the method may include the step of determining the fraction of mitochondrial gene expression based on single-cell sequence data.

[0279] method (4200) The stage 4205 The method may include a step of removing data associated with cells having a fraction of mitochondrial gene expression exceeding a gene expression threshold from the dextramer sequence data. The gene expression threshold may be about 40% of the total number of unique molecular identifiers.

[0280] method (4200) The stage 4206The method may include a step of determining unclassified dextramer sequence data based on dextramer sequence data. Classified dextramer sequence data may include classified test dextramer sequence data and negative control dextramer sequence data. Unclassified dextramer sequence data may include unclassified test dextramer sequence data.

[0281] method (4200) silver, stage 4207 In this, for each cell indicated in the dextramer sequence data, the method may include the step of determining the maximum negative control dextramer signal based on the negative control dextramer sequence data. The maximum negative control dextramer signal is It can be expressed as ), where n is the number of negative control dextramers.

[0282] method (4200) silver, stage 4208 In this, for each cell indicated in the dextramer sequence data, the method may include a step of determining the maximum classified dextramer signal based on the classified test dextramer sequence data. The maximum classified dextramer signal is It can be expressed as ), where m is the number of test dextramers.

[0283] method (4200) silver, stage 4209 For each cell indicated in the dextramer sequence data, the method may include a step of determining the maximum unclassified dextramer signal based on the unclassified test dextramer sequence data. The maximum unclassified dextramer signal is It can be expressed as ), where m is the number of test dextramers.

[0284] method (4200) silver, stage 4210In this, it may include a step of estimating dextramer binding background noise based on the maximum negative control dextramer signal. The dextramer binding background noise is ( It may include a step of determining ).

[0285] method (4200) The stage 4211 In this, the method may include a step of estimating the dextramer classification gate efficiency based on the maximum classified dextramer signal and the maximum unclassified dextramer signal. The dextramer classification gate efficiency is The dextramer classification gate efficiency can be expressed as ) and It can be determined as the maximum difference between ).

[0286] method (4200) The stage 4212 Based on the dextramer binding background noise and the dextramer classification gate efficiency, the method may include a step of determining a measured value of background noise. The measured value of background noise is ( d It can be expressed as ).

[0287] method (4200) silver, stage 4213 In this, for each cell indicated in the dextramer sequence data, the method may include a step of subtracting a measurement of background noise from the dextramer signal associated with each cell. The step of subtracting a measurement of background noise from the dextramer signal associated with each cell includes ( It may include a step of evaluating ).

[0288] The method (4200) may include, in step 4214, a step of performing cell-specific normalization on the dextramer signal associated with each cell for each cell indicated in the dextramer sequence data. The step of performing cell-specific normalization may include a step of evaluating the following:

[0289]

[0290] method (4200) The stage 4215 In this, for each cell indicated in the dextramer sequence data, the method may include a step of performing pMHC-specific normalization. The step of performing pMHC-specific normalization may include a step of evaluating the following:

[0291]

[0292] method (4200) silver, stage 4216 In this, for each cell indicated in the dextramer sequence data, the method may include the step of determining the presence or absence of at least one α-chain and at least one β-chain based on single-cell TCR sequence data.

[0293] method (4200) The stage 4217 In this case, the method may include the step of removing data associated with cells having only an α-chain, only a β-chain, or multiple α- or β-chains based on the presence or absence of at least one α-chain and at least one β-chain from normalized dextramer sequence data.

[0294] method (4200) The stage 4218 It may include a step of identifying data remaining in the normalized dextramer sequence data as associated with reliable TCR-pMHC binding events.

[0295] method (4200) The method may additionally include a step of training a predictive model based on data associated with reliable TCR-pMHC binding events. (4200) It may additionally include a step of predicting the binding state of a newly presented receptor sequence according to a trained prediction model.

[0296] In one embodiment, the ICON module (108) and / or prediction module (110) The method (4300) It can be configured to perform, Fig. 43 It is shown in. Method (4300) It can be performed wholly or partially by a single computing unit, multiple electronic devices, etc. Method (4300) The stage 4310 The method may include the step of receiving single-cell sequencing data including single-cell sequence data, dextramer sequence data, and single-cell T cell receptor (TCR) sequence data. The single-cell sequence data may include RNA-seq data, the dextramer sequence data may include dCODE-Dextamer-seq data, and the single-cell T cell receptor (TCR) sequence data may include TCR-seq data.

[0297] method (4300) The stage 4320 The method may include a step of filtering data associated with low-quality cells from dextramer sequence data based on single-cell sequence data. The step of filtering data associated with low-quality cells from dextramer sequence data based on single-cell sequence data may include, for each cell indicated in the dextramer sequence data, determining the number of genes based on single-cell sequence data; removing data associated with cells having a number of genes outside a gene threshold range from the dextramer sequence data; for each cell indicated in the dextramer sequence data, determining the fraction of mitochondrial gene expression based on single-cell sequence data; and removing data associated with cells having a fraction of mitochondrial gene expression exceeding a gene expression threshold from the dextramer sequence data. The gene threshold range may be about 200 genes to about 2,500 genes. The gene expression threshold may be about 40% of the total number of unique molecular identifiers.

[0298] method (4300) The stage 4330The method may include a step of adjusting dextramer sequence data based on a measurement of background noise. (4300) The method may further include a step of determining classified dextramer sequence data based on dextramer sequence data, wherein the classified dextramer sequence data includes classified test dextramer sequence data, negative control dextramer sequence data, and unclassified dextramer sequence data, wherein the unclassified dextramer sequence data includes unclassified test dextramer sequence data. (4300) The method may further include the step of determining the maximum negative control dextramer signal based on negative control dextramer sequence data for each cell indicated in the dextramer sequence data, determining the maximum classified dextramer signal based on classified test dextramer sequence data for each cell indicated in the dextramer sequence data, and determining the maximum unclassified dextramer signal based on unclassified test dextramer sequence data for each cell indicated in the dextramer sequence data. The maximum negative control dextramer signal is It can be expressed as ), where n is the number of negative control dextramers. The maximum classified dextramer signal is It can be expressed as ), where m is the number of test dextramers. The maximum unclassified dextramer signal is It can be expressed as ), where m is the number of test dextramers.

[0299] The step of adjusting dextramer sequence data based on background noise measurements comprises estimating dextramer binding background noise based on the maximum negative control dextramer signal, estimating dextramer classification gate efficiency based on the maximum classified dextramer signal and the maximum unclassified dextramer signal, and, based on the dextramer binding background noise and dextramer classification gate efficiency, the measurement of background noise ( d The method may further include the step of determining ) and, for each cell indicated in the dextramer sequence data, subtracting a measurement of background noise from the dextramer signal associated with each cell. The measurement of background noise is ( d It can be expressed as ). The step of subtracting the measurement of background noise from the dextramer signal associated with each cell ( It may include a step of evaluating ). Method (4300) The method may further include a step of normalizing dextramer sequence data. The step of normalizing dextramer sequence data may include a step of performing cell-specific normalization on the dextramer signal associated with each cell for each cell indicated in the dextramer sequence data, and / or a step of performing pMHC-specific normalization for each cell indicated in the dextramer sequence data. The step of performing cell-specific normalization may include a step of evaluating the following:

[0300] The step of performing pMHC-specific normalization may include a step of evaluating the following:

[0301]

[0302] method (4300) The stage 4340The method may include the step of filtering data based on the presence or absence of an α-chain or a β-chain from the dextramer sequence data, based on single-cell TCR data. The step of filtering data based on the presence or absence of an α-chain or a β-chain from the dextramer sequence data, based on single-cell TCR data, may include determining the presence or absence of at least one α-chain and at least one β-chain for each cell indicated in the dextramer sequence data, based on the single-cell TCR sequence data, and removing data associated with cells having only an α-chain, only a β-chain, or multiple α- or β-chains from the normalized dextramer sequence data, based on the presence or absence of at least one α-chain and at least one β-chain.

[0303] method (4300) The stage 4350 It may include a step of identifying data remaining in the normalized filtered dextramer sequence data as being associated with reliable TCR-pMHC binding events.

[0304] method (4300) The method may additionally include a step of training a prediction model based on the data remaining in the normalized filtered dextramer sequence data. (4300) It may additionally include a step of predicting the binding state of a newly presented receptor sequence according to a trained prediction model.

[0305] In one embodiment, the ICON module (108) and / or prediction module (110) The method (4400) It can be configured to perform, Fig. 44 It is shown in. Method (4400) It can be performed wholly or partially by a single computing unit, multiple electronic devices, etc. Method (4400) The stage 4410The method may include a step of performing TCR-pMHC binding specificity data normalization on dextramer sequence data to identify multiple TCR-pMHC binding events. The step of performing TCR-pMHC binding specificity data normalization on dextramer sequence data to identify multiple TCR-pMHC binding events is a method (4200) and / or methods (4300) It may include part or all of.

[0306] method (4400) The stage 4420 Based on normalized dextramer sequence data, the method may include the step of determining a training dataset containing multiple TCR sequences, wherein each TCR sequence is associated with a binding affinity. The step of determining a training dataset containing multiple TCR sequences based on normalized dextramer sequence data, wherein each TCR sequence is associated with a binding affinity, may include the step of determining a paired αβ chain CDR3 amino acid sequence, a V gene identifier, and a J gene identifier for each of the multiple TCR sequences, and encoding the paired αβ chain CDR3 amino acid sequence, the V gene segment sequence, and the J gene segment sequence for each of the multiple TCR sequences into a one-dimensional input vector. The step of encoding a paired αβ chain CDR3 amino acid sequence for each of the multiple TCR sequences includes the step of converting each αβ symbol of an amino acid into a numerical representation of an amino acid. For each of the multiple TCR sequences, the step of encoding the V gene identifier and the J gene identifier includes one-hot encoding to generate a categorical and discrete representation of the gene name in a numeric space.

[0307] method (4400)It may additionally include a step of clustering a one-dimensional input vector into one or more clusters. The step of clustering a one-dimensional input vector into one or more clusters includes a step of applying a KNN clustering algorithm to the one-dimensional input vector. One or more clusters represent the bond strength.

[0308] method (4400) The stage 4430 In this, the method may include a step of determining multiple feature parts for a prediction model based on multiple TCR sequences. The prediction model may include a weighted binary classifier or a synthetic neural network (CNN).

[0309] method (4400) The stage 4440 In this, the method may include a step of training a prediction model based on a plurality of features based on a first portion of a training dataset. The step of training a prediction model based on a plurality of features based on a first portion of a training dataset includes training a synthetic neural network (CNN). The step of training a prediction model based on a plurality of features based on a first portion of a training dataset includes applying a class-weighted cost function.

[0310] method (4400) The stage 4450 In this, it may include a step of testing a prediction model based on a second part of the training dataset.

[0311] method (4400) The stage 4460 In this, it may include a step of outputting a prediction model based on the test.

[0312] method (4400) It may further include the step of presenting an unknown TCR sequence to a trained prediction model and the step of predicting binding affinity by the trained prediction model.

[0313] In one embodiment, the ICON module (108) and / or prediction module(110) The method (4500) It can be configured to perform, Fig. 45 It is shown in. Method (4500) It can be performed wholly or partially by a single computing unit, multiple electronic devices, etc. Method (4500) The stage 4510 The method may include a step of presenting an unknown TCR sequence to a trained predictive model, wherein the trained predictive model is trained based on a training dataset derived according to TCR-pMHC binding specificity data normalization. (4500) The stage 4510 The method may include a step of performing TCR-pMHC binding specificity data normalization on dextramer sequence data to identify multiple TCR-pMHC binding events. The step of performing TCR-pMHC binding specificity data normalization on dextramer sequence data to identify multiple TCR-pMHC binding events is a method (4200) and / or methods (4300) It may include part or all of.

[0314] method (4500) silver, stage 4520 The method may include a step of predicting the coupling affinity using a prediction model trained in the method. The prediction model may include a weighted binary classifier or a synthetic neural network (CNN).

[0315] method (4500) The method may include the step of determining a training dataset containing multiple TCR sequences based on normalized dextramer sequence data, wherein each TCR sequence is associated with a binding affinity. The training dataset may include multiple TCR sequences, wherein each TCR sequence is associated with a binding affinity. The training dataset may include paired αβ chain CDR3 amino acid sequences, V gene identifiers, J gene identifiers, and binding affinities (e.g., yes / no).

[0316] method (4500) The method may include training a prediction model based on a plurality of features based on a first portion of a training dataset. The step of training a prediction model based on a plurality of features based on a first portion of a training dataset includes training a synthetic neural network (CNN). The step of training a prediction model based on a plurality of features based on a first portion of a training dataset includes training a synthetic neural network (CNN) in which one translational invariant layer is applied to the sequence and then three fully connected synthetic layers are applied to the final output layer. The step of training a prediction model based on a plurality of features based on a first portion of a training dataset includes applying a class-weighted cost function. Based on a first part of a training dataset, the step of training a prediction model according to a plurality of feature parts includes training a neural network by embedding one-hot encoded V and J genes of each chain of TCR sequences through learned embeddings, and concatenating these embeddings with the output of a synthetic neural network for each CDR3 to which the embedded CDR3 is supplied, forming a 1D numeric vector representing the TCR, and then passing each numeric TCR sequence through a final fully connected layer.

[0317] In one embodiment, the ICON module (108) and / or prediction module (110) The method (4400) It can be configured to perform, Fig. 44 It is shown in. Method (4400) It can be performed wholly or partially by a single computing unit, multiple electronic devices, etc. Method (4400) silver 4601 It may include the step of receiving single-cell sequence data, dextramer sequence data, and single-cell T cell receptor (TCR) sequence data.

[0318] method (4400) silver,4602 In this, for each cell indicated in the dextramer sequence data, the method may include the step of determining the number of genes based on single-cell sequence data.

[0319] method (4400) silver 4603 The method may include a step of removing data associated with cells having a number of genes outside the gene threshold range from the dextramer sequence data.

[0320] method (4400) silver, 4604 In this, for each cell indicated in the dextramer sequence data, the method may include the step of determining the fraction of mitochondrial gene expression based on single-cell sequence data.

[0321] method (4400) silver 4605 The method may include a step of removing data associated with cells having a fraction of mitochondrial gene expression exceeding a gene expression threshold from dextramer sequence data.

[0322] method (4400) silver 4606 Based on the dextramer sequence data, the method may include a step of determining classified dextramer sequence data, wherein the classified dextramer sequence data includes classified test dextramer sequence data and negative control dextramer sequence data.

[0323] method (4400) silver, 4607 In this, for each cell indicated in the dextramer sequence data, the method may include the step of determining the maximum negative control dextramer signal based on the negative control dextramer sequence data.

[0324] method (4400) silver, 4608In this, for each cell indicated in the dextramer sequence data, the method may include the step of determining the maximum classified dextramer signal based on the classified test dextramer sequence data.

[0325] method (4400) silver, 4609 In this, the method may include a step of estimating dextramer binding background noise based on the maximum negative control dextramer signal and the maximum classified dextramer signal.

[0326] method (4400) silver, 4610 In this, for each cell indicated in the dextramer sequence data, the method may include the step of determining the presence or absence of at least one α-chain and at least one β-chain based on single-cell TCR sequence data.

[0327] method (4400) silver 4611 In this, the method may include the step of removing data associated with cells having only an α-chain, only a β-chain, or multiple α- or β-chains based on the presence or absence of at least one α-chain and at least one β-chain from the dextramer sequence data.

[0328] method (4400) silver, 4612 In this, for each dextramer binding to a given cell indicated in the dextramer sequence data, the method may include a step of determining the ratio of the sum of intracellular dextramer signals to all dextramer bindings to the cell (a measure of dextramer binding specificity to the cell). For each dextramer binding to a given cell indicated in the dextramer sequence data, the step of determining the ratio of the sum of intracellular dextramer signals to all dextramer bindings to the cell comprises: Binding to dextramer Background noise-subtracted dextramer signal for T cells The step of determining, and evaluating the following Regarding T cells It may include a step of determining the fraction of the dextramer signal resulting from the binding of the dextramer:

[0329]

[0330] method (4400) silver, 4613 In this method, for each dextramer binding to a given TCR clone type of each cell indicated in the dextramer sequence data, the method may include the step of determining the fraction of T cells within the clone for a specific dextramer (a measure of the dextramer binding specificity for the clone type to which the cell belongs). For each dextramer binding to a given TCR clone type of each cell indicated in the dextramer sequence data, the step of determining the fraction of T cells within the clone for a specific dextramer comprises: T cell, TCR clonal type Determining, clone type binding to dextramer j The number of T cells belonging to, Determining, and evaluating the following Clone type that binds to dextramer It may include determining the fraction of T cells belonging to:

[0331]

[0332] method (4400) silver, 4641In this, for each dextramer binding to a given cell indicated in the dextramer sequence data, the method may include the step of determining a corrected dextramer signal associated with each dextramer binding to a cell based on a measurement of dextramer binding specificity for the cell and a measurement of dextramer binding specificity for the clonal type to which the cell belongs. For each dextramer binding to a given cell indicated in the dextramer sequence data, the step of determining a corrected dextramer signal associated with each dextramer binding to a cell based on a measurement of dextramer binding specificity for the cell and a measurement of dextramer binding specificity for the clonal type to which the cell belongs may include evaluating the following: Binding to dextramer It may include determining the corrected dextramer signal for T cells:

[0333] .

[0334] method (4400) The method may include the step of performing cell-specific normalization of the dextramer signal associated with each cell for each cell indicated in the dextramer sequence data;

[0335] method (4400) silver 4615 In this, for each cell indicated in the dextramer sequence data, the step of performing pMHC-specific normalization may be included.

[0336] method (4400) silver 4616 Based on a threshold value, the method may include a step of identifying data remaining in the normalized dextramer sequence data as being associated with reliable TCR-pMHC binding events.

[0337] A person skilled in the art will recognize or be convinced of many equivalents to specific embodiments of the methods and compositions described herein through experiments not exceeding the ordinary scope. Such equivalents are intended to be included in the following claims.

Claims

Claim 1 A method performed by a computing device, comprising: receiving single-cell sequencing data including single-cell sequence data, dextramer sequence data, and single-cell T-cell receptor (TCR) sequence data by the computing device; filtering data associated with low-quality cells by removing, based on the single-cell sequence data, from the dextramer sequence data, data associated with cells having a number of genes outside a gene threshold range or a fraction of mitochondrial gene expression exceeding a gene expression threshold; by subtracting a measurement of background noise from the dextramer signal associated with each cell for each cell indicated in the dextramer sequence data by the computing device; and filtering data according to the presence or absence of α-chains or β-chains by removing, based on the single-cell TCR sequence data, from the dextramer sequence data, data associated with cells having only α-chains, only β-chains, or multiple α- or β-chains. A method comprising the step of identifying, by the computing device, the data remaining in the filtered dextramer sequence data as being associated with a reliable TCR-pMHC binding event. Claim 2 A method according to claim 1, wherein the step of filtering data associated with low-quality cells from the dextramer sequence data based on the single-cell sequence data by the computing device comprises: the step of determining the number of genes based on the single-cell sequence data for each cell indicated in the dextramer sequence data by the computing device; and the step of determining the fraction of mitochondrial gene expression based on the single-cell sequence data for each cell indicated in the dextramer sequence data by the computing device. Claim 3 A method according to claim 1 or 2, further comprising: a step of determining, by the computing device based on the dextramer sequence data, the classified dextramer sequence data, which includes classified test dextramer sequence data and negative control dextramer sequence data as classified dextramer sequence data, and the unclassified dextramer sequence data, which includes unclassified test dextramer sequence data as unclassified dextramer sequence data; a step of determining, by the computing device, a maximum negative control dextramer signal for each cell indicated in the dextramer sequence data, based on the negative control dextramer sequence data; a step of determining, by the computing device, a maximum classified dextramer signal for each cell indicated in the dextramer sequence data, based on the classified test dextramer sequence data; and a step of determining, by the computing device, a maximum unclassified dextramer signal for each cell indicated in the dextramer sequence data, based on the unclassified test dextramer sequence data. Claim 4 In claim 3, the step of subtracting a measurement of background noise from a dextramer signal associated with each cell for each cell indicated in the dextramer sequence data by the computing device comprises: a step of estimating dextramer binding background noise based on the maximum negative control dextramer signal by the computing device; a step of estimating dextramer classification gate efficiency based on the maximum classified dextramer signal and the maximum unclassified dextramer signal by the computing device; and a step of determining a measurement of background noise based on the dextramer binding background noise and the dextramer classification gate efficiency by the computing device. Claim 5 A method according to claim 1 or 2, wherein the step of filtering data according to the presence or absence of an α-chain or a β-chain from the dextramer sequence data based on the single-cell TCR sequence data by the computing device comprises: determining the presence or absence of at least one α-chain and at least one β-chain based on the single-cell TCR sequence data for each cell indicated in the dextramer sequence data by the computing device. Claim 6 A method according to claim 5, further comprising: a step of determining, by the computing device, for each dextramer binding to a given cell indicated in the dextramer sequence data, the ratio of the intracellular dextramer signal to the sum of all dextramers binding to the cell as a measure of dextramer binding specificity for the cell; a step of determining, by the computing device, for each dextramer binding to a given TCR clone type indicated in the dextramer sequence data, the fraction of T cells within the clone binding to a specific dextramer as a measure of dextramer binding specificity for the clone type to which the cell belongs; and a step of determining, by the computing device, for each dextramer binding to a given cell indicated in the dextramer sequence data, a corrected dextramer signal associated with each dextramer binding to the cell based on the measure of dextramer binding specificity for the cell and the measure of dextramer binding specificity for the clone type to which the cell belongs. Claim 7 The method according to claim 1 or 2 further comprises the step of training a prediction model based on data remaining in the filtered dextramer sequence data by the computing device, wherein the step of training a prediction model based on data remaining in the filtered dextramer sequence data comprises: determining a training dataset containing a plurality of TCR sequences based on data remaining in the filtered dextramer sequence data by the computing device, wherein each TCR sequence is associated with a binding affinity; determining a plurality of feature parts for a prediction model based on the plurality of TCR sequences by the computing device; training a prediction model according to the plurality of feature parts based on a first portion of the training dataset by the computing device; testing the prediction model based on a second portion of the training dataset by the computing device; and outputting the prediction model based on the test by the computing device. Claim 8 In claim 7, the method comprises the step of determining a training dataset containing a plurality of TCR sequences based on data remaining in the filtered dextramer sequence data by the computing device, wherein each TCR sequence is associated with binding affinity: the step of determining, by the computing device, for each TCR sequence of the plurality of TCR sequences, a paired αβ chain CDR3 amino acid sequence, a V gene segment sequence, and a J gene segment sequence; and the step of encoding, by the computing device, for each TCR sequence of the plurality of TCR sequences, the paired αβ chain CDR3 amino acid sequence, the V gene segment sequence, and the J gene segment sequence into a one-dimensional input vector. Claim 9 In claim 8, for each of the plurality of TCR sequences, the step of encoding the paired αβ chain CDR3 amino acid sequence comprises the step of converting each alphabetic representation of the amino acid into a numerical representation of the amino acid. Claim 10 In claim 8, for each of the plurality of TCR sequences, the step of encoding the V gene segment sequence and the J gene segment sequence comprises the step of one-hot encoding to generate a categorical and discrete representation of a gene name in a number space. Claim 11 In claim 10, the step of training a prediction model according to a plurality of feature parts based on a first part of the training dataset comprises: training a neural network by embedding one-hot encoded V and J genes of each chain of TCR sequences through learned embeddings, and concatenating these embeddings with the output of a synthetic neural network for each CDR3 to which the embedded CDR3 is supplied, forming a 1D numeric vector representing the TCR, and then passing each numeric TCR sequence through a final fully connected layer. Claim 12 In claim 8, the method further comprises the step of clustering the one-dimensional input vector into one or more clusters by the computing device, wherein the step of clustering the one-dimensional input vector into one or more clusters includes: applying a KNN clustering algorithm to the one-dimensional input vector, and wherein the one or more clusters represent a bond strength. Claim 13 A method according to claim 7, further comprising the steps of: presenting an unknown TCR sequence to a trained prediction model by the computing device; and predicting binding affinity by the trained prediction model. Claim 14 A method according to claim 7, further comprising: a step of presenting target TCR sequence data to a prediction model by the computing device; a step of determining a target TCR binding pattern by the prediction model based on the target TCR sequence data; and a step of determining by the computing device the possibility that a target associated with the TCR sequence data has moved to one or more locations based on a repository of antigen locations and the target TCR binding pattern. Claim 15 A method according to claim 1 or 2, further comprising: generating a TCR binding pattern for a subject based on data remaining in filtered dextramer sequence data associated with a reliable TCR-pMHC binding event by the computing device; receiving, by the computing device at a subsequent time, second single-cell sequence data, second dextramer sequence data, and second single-cell T cell receptor (TCR) sequence data for the subject; determining a second TCR binding pattern based on the second single-cell sequence data, second dextramer sequence data, and second single-cell T cell receptor (TCR) sequence data for the subject by the computing device; and identifying the subject based on a comparison of the TCR binding pattern for the subject and the second TCR binding pattern by the computing device. Claim 16 delete Claim 17 delete Claim 18 delete Claim 19 delete Claim 20 delete Claim 21 delete Claim 22 delete Claim 23 delete Claim 24 delete Claim 25 delete Claim 26 delete Claim 27 delete Claim 28 delete Claim 29 delete Claim 30 delete Claim 31 delete Claim 32 delete Claim 33 delete Claim 34 delete Claim 35 delete Claim 36 delete Claim 37 delete Claim 38 delete Claim 39 delete Claim 40 delete Claim 41 delete Claim 42 delete Claim 43 delete Claim 44 delete Claim 45 delete Claim 46 delete Claim 47 delete Claim 48 delete Claim 49 delete

Images