Antigen prediction of infectious disease-derived epitopes

Abstract

Disclosed herein are systems and methods for determining allele, antigen, and infectious disease based vaccine compositions, as determined based on HLA alleles expressed by a patient. In addition, unique infectious disease-derived vaccines are also described herein.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of and priority to U.S. Provisional Patent Application No. 63 / 329,259, filed on April 8, 2022, the entire disclosure of which is hereby incorporated by reference in its entirety for all purposes. Background Art

[0003] The therapeutic vaccine of viral infection has great prospects for personalized treatment. Accurately predicting the viral epitopes that HLA alleles may present can be valuable for developing therapeutic vaccines. The therapeutic vaccine comprising predicted viral epitopes can more effectively trigger antiviral immune responses. However, unlike the efforts in cancers with a large number of mass spectrometry (MS) human immunopeptidome data sets before, infectious disease-derived antigens lack such MS immunopeptidome data sets. This shows that there are great limitations in constructing a model that predicts that viral epitopes may present. Therefore, it is necessary to be able to effectively utilize new models of other types of data (e.g., in addition to MS immunopeptidome data sets). Summary of the invention

[0004] Disclosed herein is an optimization method for determining and selecting infectious disease derived antigens for personalized infectious disease (ID) vaccines. In general, the method involves applying a multi-part presentation model to generate an independent allele (per-allele) presentation possibility, indicating whether a single HLA allele (e.g., HLA alleles expressed by a patient) may present an epitope (e.g., an infectious disease derived epitope). In various embodiments, the multi-part presentation model includes a first part and a second part, the first part including a pan-allele model part (also referred to as a pan-specific model part), and the second part including one or more allele-specific model parts, also referred to herein as an independent allele model. Typically, the pan-allele model part allows information to be shared across similar alleles, so that the pan-allele model part can generate a presentation possibility across various HLA alleles, including HLA alleles that the pan-allele model part has not previously encountered. Each allele-specific model part is represented as a network model for generating a presentation possibility for a specific HLA allele. Therefore, the pan-allele model part and the allele-specific model part both generate independent allele presentation possibilities. By combining the outputs of the pan-allele model part and the allele-specific model part, the multi-part presentation model outputs an improved independent allele presentation probability for a single peptide sequence (eg, an infectious disease-derived peptide sequence).

[0005] In various embodiments, the multipart presentation model is trained using training data derived from one of the following: 1) binding affinity data between training peptide sequences and HLA alleles, and 2) mass spectrometry elution peptide data representing presentation of training peptide sequences and HLA alleles. In specific embodiments, the multipart presentation model is trained using training data derived from both of the following: 1) binding affinity data between training peptide sequences and HLA alleles, and 2) elution peptide data from mass spectra representing presentation of training peptide sequences and HLA alleles. Here, it may be preferred to use both types of data to train the multipart presentation model, especially in the case where the amount of elution peptide data for infectious disease-derived peptide sequences generated by mass spectrometry is limited.

[0006] Disclosed herein is a method for determining one or more infectious disease-derived antigens that may be presented by cells of a subject, the method comprising: obtaining peptide sequences of a plurality of infectious disease-derived antigens; obtaining sequences of one or more MHC alleles of the subject; inputting the peptide sequences of the plurality of infectious disease-derived antigens and the sequences of the one or more MHC alleles of the subject into a multi-part presentation model to generate a set of numerical likelihoods that the plurality of infectious disease-derived antigens are presented by the one or more MHC alleles expressed on the surface of cells of the subject, wherein a first part of the multi-part presentation model comprises a pan-allele model part that receives as input the peptide sequences of the one or more infectious disease-derived antigens and the sequences of the one or more MHC alleles of the subject, or representations thereof, and wherein a second part of the multi-part presentation model comprises a plurality of allele-specific models, each of the allele-specific models receiving as input the peptide sequences of the plurality of infectious disease-derived antigens, or representations thereof; and selecting a subset of the plurality of infectious disease-derived antigens based on the set of numerical likelihoods to generate a set of selected antigens.

[0007] In various embodiments, the multi-part presentation model comprises a plurality of parameters generated using at least 1) mass spectrometry data and 2) binding affinity data determined from a plurality of samples. In various embodiments, the multi-part presentation model comprises a plurality of parameters generated using a training data set, the training data set comprising: training peptide sequences, and for one or more training peptide sequences, a label derived from the mass spectrometry data indicating whether the training peptide sequence is presented by one or more class I MHC alleles present in a plurality of samples. In various embodiments, the training peptide sequences are determined by performing mass spectrometry analysis on separated peptides eluted from MHC alleles present in a plurality of samples. In various embodiments, the multi-part presentation model comprises a plurality of parameters generated using a training data set, the training data set comprising: for one or more training peptide sequences, a label derived from the binding affinity data indicating whether the training peptide sequence binds to one or more class I MHC alleles present in a plurality of samples.

[0008] In various embodiments, the length of the training peptide sequence is in the range of k-mers, where k is between 8-15, inclusive. In various embodiments, the length of the training peptide sequence is in the range of k-mers, where k is between 8-11, inclusive. In various embodiments, the training peptide sequence is an infectious disease derived training peptide sequence.

[0009] In various embodiments, the pan-allele model portion includes a neural network. In various embodiments, a first set of layers of the neural network of the pan-allele model portion performs dimensionality reduction on the sequence of one or more MHC alleles of the subject. In various embodiments, a second set of layers of the neural network of the pan-allele model portion receives as input a peptide sequence representation of a plurality of infectious disease-derived antigens and a dimensionality-reduced representation of the sequence of one or more MHC alleles of the subject. In various embodiments, the representation of the peptide sequence of the plurality of infectious disease-derived antigens is generated by encoding the peptide sequence with a one-hot encoding scheme. In various embodiments, the second set of layers of the neural network models the interaction between the peptide sequence of the plurality of infectious disease-derived antigens and the sequence of one or more MHC alleles of the subject.

[0010] In various embodiments, one or more allele-specific models include a neural network. In various embodiments, the neural network of the allele-specific network receives a peptide sequence representation of a plurality of infectious disease-derived antigens as input and outputs an independent allele presentation probability of the allele. In various embodiments, the representation of the peptide sequence of a plurality of infectious disease-derived antigens is generated by encoding the peptide sequence through a unique hot encoding scheme. In various embodiments, each of the allele-specific models includes a neural network. In various embodiments, the second part of the multi-part presentation model includes ten or more allele-specific models.

[0011] In various embodiments, the numerical likelihood of an antigen is a combination of the output of a pan-allele model portion and the output of a plurality of allele-specific models. In various embodiments, the set of numerical likelihoods is further determined by a feature comprising at least one of: (a) a plurality of infectious disease derived antigenic peptide sequence flanking C-terminal sequences; and (b) a plurality of infectious disease derived antigenic peptide sequence flanking N-terminal sequences.

[0012] In various embodiments, the plurality of samples comprises at least one of: (a) one or more cell lines engineered to express a single MHC class I allele; (b) one or more cell lines engineered to express multiple MHC class I alleles; (c) one or more human cell lines obtained or derived from multiple patients; (d) fresh or frozen samples obtained from multiple patients; and (e) fresh or frozen tissue samples obtained from multiple patients. In various embodiments, the cells of the subject include cells infected with one of a pathogen, virus, bacteria, fungus, or parasite. In various embodiments, the infectious disease-derived antigen is derived from one of a pathogen, virus, bacteria, fungus, or parasite. In various embodiments, the infectious disease-derived antigen is derived from an infectious disease organism selected from the group consisting of severe acute respiratory syndrome-associated coronavirus (SARS), severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), Ebola virus, HIV, hepatitis B virus (HBV), influenza, hepatitis C virus (HCV), human papillomavirus (HPV), cytomegalovirus (CMV), chikungunya virus, respiratory syncytial virus (RSV), dengue virus, Orthomyxoviridae virus, tuberculosis, pan-coronavirus, herpes simplex virus infection (HSV), influenza, metapneumovirus (MPV) and parainfluenza virus (PIV).

[0013] Disclosed herein is a method for treating an infectious disease in a subject, comprising performing any of the methods disclosed herein, and further comprising obtaining a vaccine comprising a collection of selected antigens, and administering the vaccine to the subject. In various embodiments, the vaccine is administered to the subject prophylactically. In various embodiments, the vaccine is administered to the subject therapeutically. Disclosed herein is a method for manufacturing a vaccine, comprising performing any of the methods disclosed herein, and further comprising producing or having produced a vaccine comprising a collection of selected antigens. Disclosed herein is a vaccine comprising a collection of selected antigens selected by performing any of the methods disclosed herein. Disclosed herein is a method for treating an infectious disease in a subject, comprising performing any of the methods disclosed herein, and further comprising obtaining an isolated antigen-binding protein exhibiting binding specificity to one or more selected antigens, and administering the isolated antigen-binding protein to the subject. Disclosed herein is a method for treating an infectious disease in a subject, comprising performing any of the methods disclosed herein, and further comprising obtaining an isolated antigen-binding protein exhibiting binding specificity to one or more selected antigens, and administering the isolated antigen-binding protein to the subject. In various embodiments, the antigen-binding protein is an antibody or an antigen-binding fragment. In various embodiments, the antigen-binding protein is a T cell receptor or a chimeric antigen receptor.

[0014] In addition, a non-transitory computer-readable medium comprising instructions is disclosed herein, which, when executed by a processor, causes the processor to: obtain peptide sequences of a plurality of infectious disease-derived antigens; obtain sequences of one or more MHC alleles of a subject; input the peptide sequences of the plurality of infectious disease-derived antigens and the sequences of the one or more MHC alleles of the subject into a multi-part presentation model to generate a set of numerical likelihoods of presentation of the plurality of infectious disease-derived antigens by the one or more MHC alleles expressed on the surface of a cell of the subject, wherein the first part of the multi-part presentation model comprises a pan-allele model part, the pan-allele model part receiving the peptide sequences of the one or more infectious disease-derived antigens and the sequences of the one or more MHC alleles of the subject or a representation thereof as input, and wherein the second part of the multi-part presentation model comprises a plurality of allele-specific models, each of the allele-specific models receiving the peptide sequences of the plurality of infectious disease-derived antigens or a representation thereof as input; and select a subset of the plurality of infectious disease-derived antigens based on the set of numerical likelihoods to generate a set of selected antigens. In various embodiments, the multi-part presentation model comprises a plurality of parameters generated using at least 1) mass spectrometry data and 2) binding affinity data determined from a plurality of samples. In various embodiments, the multi-part presentation model comprises a plurality of parameters generated using a training data set, the training data set comprising: training peptide sequences, and for one or more of the training peptide sequences, a label derived from mass spectrometry data indicating whether the training peptide sequence is presented by one or more MHC class I alleles present in the plurality of samples. In various embodiments, the training peptide sequences are determined by performing mass spectrometry analysis on isolated peptides eluted from the MHC alleles present in the plurality of samples.

[0015] In various embodiments, the multi-part presentation model comprises a plurality of parameters generated using a training data set, the training data set comprising: for one or more training peptide sequences, a label derived from binding affinity data indicating whether the training peptide sequence binds to one or more class I MHC alleles present in a plurality of samples. In various embodiments, the length of the training peptide sequence is in the range of k-mers, where k is between 8-15, inclusive. In various embodiments, the length of the training peptide sequence is in the range of k-mers, where k is between 8-11, inclusive.

[0016] In various embodiments, the training peptide sequence is a training peptide sequence derived from an infectious disease. In various embodiments, the pan-allele model portion includes a neural network. In various embodiments, the first set of layers of the neural network of the pan-allele model portion reduces the dimension of the sequence of one or more MHC alleles of the subject. In various embodiments, the second set of layers of the neural network of the pan-allele model portion receives a plurality of peptide sequence representations of infectious disease-derived antigens and a reduced dimension representation of the sequence of one or more MHC alleles of the subject as input. In various embodiments, the representation of the peptide sequence of multiple infectious disease-derived antigens is generated by encoding the peptide sequence by a unique hot encoding scheme. In various embodiments, the second set of layers of the neural network models the interaction between the peptide sequence of multiple infectious disease-derived antigens and the sequence of one or more MHC alleles of the subject. In various embodiments, one or more allele-specific models include a neural network. In various embodiments, the neural network of the allele-specific network receives a plurality of peptide sequence representations of infectious disease-derived antigens as input and outputs the independent allele presentation possibility of the allele.

[0017] In various embodiments, the representation of the peptide sequences of multiple infectious disease derived antigens is generated by encoding the peptide sequences by a one-hot encoding scheme. In various embodiments, each of the allele-specific models includes a neural network. In various embodiments, the second part of the multi-part presentation model includes ten or more allele-specific models. In various embodiments, the numerical likelihood of the antigen is a combination of the output of the pan-allele model part and the output of multiple allele-specific models. In various embodiments, the set of numerical likelihoods is further determined by a feature comprising at least one of the following: (a) C-terminal sequences flanked by peptide sequences of multiple infectious disease derived antigens; and (b) N-terminal sequences flanked by peptide sequences of multiple infectious disease derived antigens. In various embodiments, multiple samples include at least one of the following: (a) one or more cell lines engineered to express a single MHC class I allele; (b) one or more cell lines engineered to express multiple MHC class I alleles; (c) one or more human cell lines obtained from or derived from multiple patients; (d) fresh or frozen samples obtained from multiple patients; and (e) fresh or frozen tissue samples obtained from multiple patients.

[0018] In various embodiments, the cells of the subject include cells infected by one of pathogens, viruses, bacteria, fungi or parasites. In various embodiments, infectious disease derived antigens are derived from one of pathogens, viruses, bacteria, fungi or parasites. In various embodiments, infectious disease derived antigens are derived from infectious disease organisms selected from the group consisting of severe acute respiratory syndrome-related coronavirus (SARS), severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), Ebola virus, HIV, hepatitis B virus (HBV), influenza, hepatitis C virus (HCV), human papillomavirus (HPV), cytomegalovirus (CMV), chikungunya virus, respiratory syncytial virus (RSV), dengue virus, orthomyxoviridae virus, tuberculosis, pan-coronavirus, herpes simplex virus infection (HSV), influenza, metapneumovirus (MPV) and parainfluenza virus (PIV). BRIEF DESCRIPTION OF THE DRAWINGS

[0019] These and other features, aspects and advantages of the present invention will be better understood with reference to the following description and accompanying drawings, in which:

[0020] Figure 1A is an overview of the circumstances for determining the likelihood of peptide presentation in a patient according to one embodiment.

[0021] Figure 1B and 1C An example cartridge design method according to an embodiment is described.

[0022] Figure 2A and 2B A method of acquiring presentation information according to an embodiment is shown.

[0023] Figure 3A is a high-level block diagram illustrating computer logic components of a presentation determination system according to one embodiment.

[0024] Figure 3B An example set of training data is shown according to one embodiment.

[0025] Figure 4A Shown is a process for implementing a multi-part presentation model according to one embodiment.

[0026] Figure 4B A network architecture for a multi-part presentation model according to one embodiment is shown.

[0027] Figure 5A An implementation of the allele-specific model portion of a multi-part presentation model according to one embodiment is shown.

[0028] Figure 5B An example network model related to MHC alleles is shown.

[0029] Figure 5C An example network model of MHC allele sharing is shown.

[0030] Figure 5D The generation of presentation probabilities of peptides associated with MHC alleles using an example network model is shown.

[0031] Figure 5E The generation of presentation probabilities of peptides associated with MHC alleles using an example network model is shown.

[0032] Fig. 5F The generation of presentation probabilities of peptides associated with MHC alleles using an example network model is shown.

[0033] Figure 5G The generation of presentation probabilities of peptides associated with MHC alleles using an example network model is shown.

[0034] Figure 5H The generation of presentation probabilities of peptides associated with MHC alleles using an example network model is shown.

[0035] Fig.5I The generation of presentation probabilities of peptides associated with MHC alleles using an example network model is shown.

[0036] Fig. 6A An implementation of the pan-allele model portion of the multi-part presentation model according to one embodiment is shown.

[0037] Figure 6B An example network model of MHC allele sharing is shown, according to one embodiment.

[0038] Figure 6C An example network model independent of MHC alleles is shown.

[0039] Fig.6D An example network model using MHC allele sharing is shown to generate presentation probabilities of peptides associated with MHC alleles.

[0040] Figure 7 Shown is a method for implementing Figure 1A and 3A An example computer for the entities shown in .

[0041] Fig. 8A The performance of various models for predicting HIV epitope presentation in the top 5 alleles is shown.

[0042] Figure 8B The performance of various models for predicting influenza A epitope presentation in the top 5 alleles is shown.

[0043] Figure 8C The performance of various models for predicting SARS-CoV-2 epitope presentation in the top 5 alleles is shown.

[0044] Fig. 9A Shown are the precision-recall curves for various models for predicting HIV epitope presentation in the top 25 alleles.

[0045] Fig. 9B Shown are the precision-recall curves for various models for predicting influenza A epitope presentation in the top 25 alleles. DETAILED DESCRIPTION

[0046] I. Definitions

[0047] In general, the terms used in the claims and the specification are intended to be interpreted as having the ordinary meaning as understood by one of ordinary skill in the art. Certain terms are defined below to provide additional clarity. If the ordinary meaning conflicts with the provided definition, the provided definition shall apply.

[0048] As used herein, the term "antigen" is a substance that induces an immune response. In various embodiments, "antigen" refers to an infectious disease-derived antigen derived from one of the pathogens, viruses, bacteria, fungi or parasites that can cause infectious diseases. Antigens can include polypeptide sequences or nucleotide sequences. In various embodiments, antigens can include mutations, such as frameshift or non-frameshift insertions and deletions, missense or nonsense substitutions, splice site changes, splice variants, genome rearrangements or gene fusions.

[0049] As used herein, the term "antigen-based vaccine" is a vaccine construct based on one or more antigens (eg, a plurality of antigens).

[0050] As used herein, the term "coding region" is the portion of a gene that encodes a protein.

[0051] As used herein, the term "coding mutation" is a mutation that occurs in the coding region.

[0052] As used herein, the term "indel" is an insertion or deletion of one or more nucleic acids.

[0053] As used herein, the term "epitope" is a specific portion of an antigen that is typically bound by an antibody or T-cell receptor.

[0054] As used herein, the term "immunogenicity" is the ability to elicit an immune response, for example, by T cells, B cells, or both.

[0055] As used herein, the terms "HLA binding affinity" and "MHC binding affinity" refer to the binding affinity between a specific antigen and a specific MHC allele.

[0056] As used herein, the term "polymorphism" is a germline variation, ie, a variation found in all DNA-bearing cells of an individual.

[0057] As used herein, the term "somatic variant" is a variant arising in non-germline cells of an individual.

[0058] As used herein, the term "allele" is one form of a gene, one form of a gene sequence, or one form of a protein.

[0059] As used herein, the term "HLA type" is the complementary sequence of the HLA gene alleles.

[0060] As used herein, the term "nonsense-mediated decay" or "NMD" is the degradation of mRNA by cells caused by premature stop codons.

[0061] As used herein, the term "exome" is a subset of a genome that encodes proteins. An exome can be a set of exons of a genome.

[0062] As used herein, the term "logistic regression" is a statistically derived regression model for binary data in which the logit of the odds that a dependent variable is equal to 1 is modeled as a linear function of the dependent variable.

[0063] As used herein, the term "neural network" is a machine learning model for classification or regression consisting of multiple layers of linear transformations followed by element-wise nonlinearities typically trained via stochastic gradient descent and back-propagation.

[0064] As used herein, the term "proteome" is the collection of all proteins expressed and / or translated by a cell, a population of cells, or an individual.

[0065] As used herein, the term "peptide group" is the collection of all peptides presented on the surface of a cell by MHC-I or MHC-II. The peptide group may refer to a property of a cell or a collection of cells.

[0066] As used herein, the term "ELISPOT" means enzyme-linked immunosorbent spot assay, which is a common method used to monitor immune responses in humans and animals.

[0067] As used herein, the term "tolerance or immune tolerance" is a state of immune unresponsiveness to one or more antigens (eg, self-antigens).

[0068] As used herein, the term "central tolerance" is tolerance that is effected in the thymus, either by deleting autoreactive T cell clones or by promoting the differentiation of autoreactive T cell clones into immunosuppressive regulatory T cells (Tregs).

[0069] As used herein, the term "peripheral tolerance" refers to tolerance that affects the periphery by downregulating or anergizing autoreactive T cells that survive central tolerance or by promoting the differentiation of these T cells into Tregs.

[0070] The term "sample" may include a single cell or multiple cells or cell fragments or an aliquot of body fluid taken from a subject by means including venipuncture, excretion, ejaculation, massage, biopsy, needle aspiration, lavage sample, scraping, surgical incision or intervention or other means known in the art.

[0071] The term "subject" is used interchangeably and encompasses a cell, tissue or organism, human or non-human, whether in vivo, ex vivo or in vitro, male / male or female / female. The terms subject and patient include mammals, including humans.

[0072] The term "mammal" encompasses both humans and non-humans, and includes, but is not limited to, humans, non-human primates, canines, felines, murines, bovines, equines, and porcines.

[0073] The term "clinical factor" refers to a measure of a subject's condition, such as disease activity or severity."Clinical factor" encompasses all markers of a subject's health status, including non-sample markers, and / or other features of the subject, such as but not limited to age and gender.Clinical factors can be a score, value, or set of values ​​that can be obtained by evaluating a sample (or sample group) from a subject or a subject under determined conditions.Clinical factors can also be predicted by markers and / or other parameters such as gene expression surrogates.

[0074] The term "antibody" herein is used in the broadest sense and includes polyclonal antibodies and monoclonal antibodies, including complete antibodies and functional (antigen-binding) antibody fragments, including fragment antigen-binding (Fab) fragments, F(ab')2 fragments, Fab' fragments, Fv fragments, recombinant IgG (rIgG) fragments, variable heavy chain (VH) regions capable of specific binding to antigens, single-chain antibody fragments, including single-chain variable fragments (scFv) and single-domain antibodies (e.g., sdAb, sdFv, nanobodies, camelid VHH, engineered or evolved human VH that does not require pairing with VL to achieve solubility or activity) fragments. The term encompasses genetic engineering and / or other modified forms of immunoglobulins, such as intrabodies, peptide antibodies, chimeric antibodies, fully human antibodies, humanized antibodies and heterologous conjugate antibodies, multispecific (e.g., bispecific) antibodies, double antibodies, three antibodies and four antibodies, two scFvs in series, three scFvs in series. Unless otherwise indicated, the term "antibody" should be understood to encompass its functional antibody fragments. The term also encompasses intact or full-length antibodies, including antibodies of any class or subclass, including IgG and its subclasses, IgM, IgE, IgA, and IgD.

[0075] Abbreviations: MHC: major histocompatibility complex; HLA: human leukocyte antigen or human MHC locus; NGS: next-generation sequencing; PPV: positive predictive value; FFPE: formalin-fixed, paraffin-embedded; NMD: nonsense-mediated decay; DC: dendritic cell.

[0076] It must be noted that, as used in the specification and the appended claims, the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise.

[0077] Any terms not directly defined herein should be understood to have the meanings commonly associated with them as understood in the field of the invention. Certain terms are discussed herein to provide additional guidance to practitioners in describing the compositions, devices, methods, etc. of the invention and how to make or use them. It should be understood that the same thing can be described in more than one way. Thus, alternative language and synonyms can be used for any one or more of the terms discussed herein. It is not important whether the term is described or discussed in detail herein. Some synonyms or alternative methods, materials, etc. are provided. The detailed description of one or more synonyms or equivalents does not exclude the use of other synonyms or equivalents unless explicitly stated. The use of examples, including term examples, is for illustrative purposes only and does not limit the scope and meaning of the aspects of the invention herein.

[0078] All references, issued patents, and patent applications cited in the body of this specification are hereby incorporated by reference in their entirety for all purposes.

[0079] II. Methods for Determining Antigens

[0080] Methods for determining antigens (e.g., antigens derived from infectious organisms) include determining antigens from infectious organisms, infections of subjects, or infected cells of subjects, which antigens may be presented on the cell surface of infected cells or immune cells (including professional antigen presenting cells, such as dendritic cells), and / or may be immunogenic. As an example, one such method may include the following steps: obtaining at least one of exome, transcriptome, or whole genome infectious organism nucleotide sequencing and / or expression data from infected cells of the subject, wherein the infectious organism nucleotide sequencing and / or expression data is used to obtain data representing a peptide sequence of each of a set of antigens (e.g., antigens derived from infectious organisms); inputting the peptide sequence of each of the antigens into one or more presentation models to generate a set of numerical likelihoods for each antigen to be presented by one or more MHC alleles on the cell surface of infected cells of the subject or cells present in the subject, the set of numerical likelihoods being determined based at least on the received mass spectrometry data; and selecting a subset of the set of antigens based on the set of numerical likelihoods to generate a set of selected antigens.

[0081] The presentation model (such as a multi-part presentation model) may include a statistical regression or machine learning (e.g., deep learning) model trained on a reference data set (also referred to as a training data set) comprising a set of corresponding labels, wherein the reference data set is obtained from each of a plurality of different subjects, wherein optionally some subjects may have an infection, and wherein the reference data set includes at least one of the following: data representing exome nucleotide sequences from infected tissues, data representing exome nucleotide sequences from normal tissues, data representing transcriptome nucleotide sequences from infected tissues, data representing proteome sequences from infected tissues, and data representing MHC peptide group sequences from infected tissues, and data representing MHC peptide group sequences from normal tissues. The reference data may also include mass spectrometry data, sequencing data, RNA sequencing data, expression profiling data, and proteomics data of monoallelic cell lines engineered to express predetermined MHC alleles that are subsequently exposed to synthetic proteins, normal human cell lines, and fresh and frozen primary samples, and T cell assays (e.g., ELISpot). In certain aspects, the set of reference data includes each form of reference data.

[0082] The presentation model may include a set of features derived at least in part from a reference data set, wherein the set of features includes at least one of an allele-dependent feature and an allele-independent feature. In certain aspects, each feature is included.

[0083] The method of identifying shared antigens also includes generating an output for constructing a personalized vaccine by identifying one or more antigens in one or more cells of the subject that may be presented on the surface of infected cells. As an example, one such method may include the following steps: obtaining at least one of exome, transcriptome, or whole genome nucleotide sequencing and / or expression data from infected cells and normal cells of a subject, wherein the nucleotide sequencing and / or expression data is used to obtain data representing a peptide sequence of each of a set of antigens determined by: comparing the nucleotide sequencing and / or expression data from infected cells with the nucleotide sequencing and / or expression data from normal cells, data of peptide sequences determined from normal cells of the subject; encoding the peptide sequence of each of the antigens into a corresponding numerical vector, each numerical vector including information about a plurality of amino acids constituting the peptide sequence and a set of amino acid positions in the peptide sequence; using a computer processor to input the numerical vector into a deep learning presentation model to generate a set of presentation probabilities for the set of antigens, each presentation probability in the set of antigens representing the probability that the corresponding antigen is presented on the surface of infected cells of the subject by one or more class II MHC alleles, the deep learning presentation model; selecting a subset of the set of antigens based on the set of presentation probabilities to generate a selected set of antigens; and generating an output for constructing a personalized vaccine based on the selected set of antigens.

[0084] Specific methods for determining antigens (e.g., infectious organism-derived antigens) are known to those of skill in the art, such as those described in more detail in International Patent Application Publications WO / 2017 / 106638, WO / 2018 / 195357, and WO / 2018 / 208856, each of which is incorporated herein by reference in its entirety for all purposes.

[0085] Disclosed herein is a method of treating a subject having an infection comprising performing the steps of any of the antigen determination methods described herein, and further comprising obtaining an infectious disease vaccine comprising a collection of selected antigens, and administering the infectious disease vaccine to the subject.

[0086] The methods disclosed herein may also include determining one or more T cells that have antigen specificity for at least one antigen in the subset. In some embodiments, determining includes co-culturing one or more T cells with one or more antigens in the subset under conditions of amplifying one or more antigen-specific T cells. In other embodiments, determining includes contacting one or more T cells with a tetramer containing one or more antigens in the subset under conditions that allow binding between T cells and tetramers. In even other embodiments, the methods disclosed herein may also include determining one or more T cell receptors (TCRs) of one or more determined T cells. In certain embodiments, determining one or more T cell receptors includes sequencing the T cell receptor sequences of one or more determined T cells. The methods disclosed herein may also include genetically engineering a plurality of T cells to express at least one of the one or more determined T cell receptors; culturing a plurality of T cells under conditions of amplifying a plurality of T cells; and infusing the amplified T cells into a subject. In some embodiments, genetically engineering a plurality of T cells to express at least one of the one or more determined T cell receptors includes cloning the T cell receptor sequences of one or more determined T cells into an expression vector; and transfecting each of a plurality of T cells with an expression vector. In some embodiments, the methods disclosed herein further comprise culturing the one or more defined T cells under conditions that expand the one or more defined T cells; and infusing the expanded T cells into the subject.

[0087] Also disclosed herein are isolated T cells having antigen specificity for at least one selected antigen in the subset.

[0088] Also disclosed herein is a method for making an infectious disease vaccine, comprising the following steps: obtaining at least one of exome, transcriptome or whole genome infectious disease organism nucleotide sequencing and / or expression data from infected cells of a subject, wherein the infectious disease organism nucleotide sequencing and / or expression data is used to obtain data representing a peptide sequence of each of a set of antigens (e.g., wherein the peptide is derived from any polypeptide known or found to have altered expression in infected cells or infected tissues compared to normal cells or tissues); inputting the peptide sequence of each of the antigens into one or more presentation models to generate a set of numerical likelihoods of each antigen being presented by one or more MHC alleles on the cell surface of infected cells of the subject, the set of numerical likelihoods being determined based at least on the received mass spectrometry data; and selecting a subset of the group of antigens based on the set of numerical likelihoods to generate a set of selected antigens; and producing or having produced an infectious disease vaccine comprising the set of selected antigens.

[0089] Also disclosed herein is an infectious disease vaccine, comprising selecting a set of selected antigens by performing a method comprising the following steps: obtaining at least one of exome, transcriptome or whole genome infectious disease organism nucleotide sequencing and / or expression data from infected cells of a subject, wherein the infectious disease organism nucleotide sequencing and / or expression data is used to obtain data representing a peptide sequence of each of the set of antigens, and wherein the peptide sequence of each antigen (e.g., derived from any polypeptide known or found to have altered expression in infected cells or infected tissues compared to normal cells or tissues); inputting the peptide sequence of each of the antigens into one or more presentation models to generate a set of numerical likelihoods of each antigen being presented by one or more MHC alleles on the cell surface of infected cells of the subject, the set of numerical likelihoods being determined based at least on the received mass spectrometry data; and selecting a subset of the group of antigens based on the set of numerical likelihoods to generate a set of selected antigens; and producing or having produced a set infectious disease vaccine comprising the selected antigens.

[0090] The vaccine may comprise one or more of a nucleotide sequence, a polypeptide sequence, RNA, DNA, a cell, a plasmid or a vector.

[0091] Vaccines may contain one or more antigens presented on the surface of infected cells.

[0092] Infectious disease vaccines may contain one or more antigens that are immunogenic in a subject.

[0093] An infectious disease vaccine may not include one or more antigens that induce an autoimmune response against a subject's normal tissues.

[0094] Infectious disease vaccines may contain an adjuvant.

[0095] Infectious disease vaccines may comprise excipients.

[0096] The methods disclosed herein may also include selecting antigens having an increased probability of being presented on the surface of infected cells relative to unselected antigens based on the presentation model.

[0097] The methods disclosed herein may also include selecting antigens having an increased likelihood of being able to induce an infectious organism-specific immune response in a subject relative to unselected antigens based on the presentation model.

[0098] The methods disclosed herein may further comprise selecting an antigen capable of being presented to naive T cells by a professional antigen presenting cell (APC) relative to an unselected antigen based on the presentation model, optionally wherein the APC is a dendritic cell (DC).

[0099] The methods disclosed herein may also include selecting antigens that have a reduced likelihood of being inhibited by central or peripheral tolerance relative to unselected antigens based on the presentation model.

[0100] The methods disclosed herein may also include selecting an antigen that has a reduced likelihood of being able to induce an autoimmune response to normal tissue in a subject relative to an unselected antigen based on the presentation model.

[0101] Exome or transcriptome nucleotide sequencing and / or expression data can be obtained by sequencing infected tissue.

[0102] The sequencing may be next generation sequencing (NGS) or any massively parallel sequencing method.

[0103] The set of numerical likelihoods can be further determined by at least an MHC allele interaction feature, the feature comprising at least one of the following: predicted affinity of binding of an MHC allele to an antigen-encoded peptide; predicted stability of an antigen-encoded peptide-MHC complex; sequence and length of the antigen-encoded peptide; probability of presentation of an antigen-encoded peptide of similar sequence in cells of other individuals expressing the particular MHC allele assessed by mass spectrometry proteomics or other methods; expression level of the particular MHC allele in the subject in question (e.g., as measured by RNA-seq or mass spectrometry); overall antigen-encoded peptide sequence-independent probability of presentation by a particular MHC allele in other different subjects expressing the particular MHC allele; overall antigen-encoded peptide sequence-independent probability of presentation by MHC alleles in the same molecular family (e.g., HLA-A, HLA-B, HLA-C, HLA-DQ, HLA-DR, HLA-DP) in other different subjects.

[0104] The set of numerical likelihoods is also determined by at least MHC-allele non-interaction features, including at least one of the following: C-terminal and N-terminal sequences flanking the antigen-encoding peptide within its source protein sequence; the presence of a protease cleavage motif in the antigen-encoding peptide, optionally weighted according to the expression of the corresponding protease in infected cells (as measured by RNA-seq or mass spectrometry); the turnover rate of the source protein as measured in an appropriate cell type; the length of the source protein, optionally taking into account a specific splice variant ("isoform") that is most highly expressed in infected cells, as measured by RNA-seq or proteomic mass spectrometry, or as predicted based on annotations of germline or somatic splicing mutations detected in DNA or RNA sequence data; the expression level of the proteasome, immunoproteasome, thymic proteasome or other protease in infected cells (which can be measured by RNA sequencing, proteomic mass spectrometry or immunohistochemistry); the expression of the source gene of the antigen-encoding peptide (e.g., as measured by RNA sequencing or mass spectrometry); the antigen during various stages of the cell cycle. Typical tissue-specific expression of source genes encoding peptides; a comprehensive catalog of features of source proteins and / or their domains, which can be found in, for example, uniProt or PDB http: / / www.rcsb.org / pdb / home / home.do; features describing the properties of domains of source proteins containing peptides, such as: secondary or tertiary structure (e.g., alpha helix vs. beta fold); alternative splicing; the probability of presenting a peptide from a source protein encoding the antigen-encoding peptide in question in other different subjects; the probability that a peptide will not be detected or over-represented by mass spectrometry due to technical bias; the expression of various gene modules / pathways as measured by RNASeq (which does not require a source protein containing the peptide), which can provide information about the state of infected cells, matrix or infected tissue; the number of copies of the source gene encoding the antigen-encoding peptide in infected cells; the probability of the peptide binding to TAP or the measured or predicted binding affinity of the peptide to TAP; the expression level of TAP in infected cells (which can be measured by RNA-seq, proteomic mass spectrometry, immunohistochemistry).Peptide presentation may be dependent on components of the antigen presentation machinery that are subject to the presence or absence of functional germline polymorphisms, including, but not limited to, genes encoding proteins involved in the antigen presentation machinery (e.g., B2M, HLA-A, HLA-B, HLA-C, TAP-1, TAP-2, TAPBP, CALR, CNX, ERP57, HLA-DM, HLA-DMA, HLA-DMB, HLA-DO, HLA-DOA, HLA-DOB, HLA-DP, HLA-DPA1, HLA-DPB1, HLA-DQ, HLA-DQA1, HLA-DQA2, HLA-DQB1, HLA-DQB2, HLA-DQA2, HLA-DQB2, HLA-DQA3, HLA-DQB4, HLA-DQB5, HLA-DQB6, HLA-DQB7, HLA-DQB8, HLA-DQB9, HLA-DQB10, HLA-DQB111, HLA-DQB12, HLA-DQB13, HLA-DQB14, HLA-DQB15, HLA-DQB16, HLA-DQB17, HLA-DQB18, HLA-DQB20, HLA-DQB21, HLA-DQB22, HLA-DQB19, HLA-DQB23, HLA-DQB24, HLA-DQB25, HLA-DQB19, HLA-DQ LA-DR, HLA-DRA, HLA-DRB1, HLA-DRB3, HLA-DRB4, HLA-DRB5 or any gene encoding a proteasome or immunoproteasome component); infection type (e.g., pathogen infection, viral infection, bacterial infection, fungal infection, and parasitic infection); clinical infection subtype (e.g., HIV infection, severe acute respiratory syndrome-related coronavirus (SARS) infection, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, Ebola infection, hepatitis B virus (HBV) infection, influenza infection, and hepatitis C virus (HCV) infection); typical expression of the source gene of the peptide in the relevant infection type or clinical subtype.

[0105] The methods disclosed herein may also include obtaining an infectious disease vaccine comprising a collection of selected antigens (eg, infectious organism-derived antigens) or a subset thereof, and optionally further comprising administering the infectious disease vaccine to a subject.

[0106] At least one antigen in the set of selected antigens (e.g., infectious organism-derived antigens) may include at least one of the following when in polypeptide form: binding affinity to MHC, IC50 value less than 1000 nM, 8-15, 8, 9, 10, 11, 12, 13, 14 or 15 amino acids in length for MHC class I polypeptides, 6-30, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 or 30 amino acids in length for MHC class II polypeptides, sequence motifs that promote proteasome cleavage in or near the polypeptide in the parent protein sequence, and sequence motifs that promote TAP transport. For MHC class II, the presence of sequence motifs in or near the peptide can promote cleavage by extracellular or lysosomal proteases (e.g., cathepsin) or HLA-DM catalyzed HLA binding.

[0107] Disclosed herein is a method for determining one or more antigens (e.g., infectious organism-derived antigens) that are likely to be presented on the cell surface of an infected cell, comprising performing the following steps: receiving mass spectrometry data, including data associated with a plurality of isolated peptides eluted from a major histocompatibility complex (MHC) derived from a plurality of fresh or frozen samples; obtaining a training data set by at least determining a set of training peptide sequences that are present in the sample and presented on one or more MHC alleles associated with each training peptide sequence; obtaining a set of training protein sequences based on the training peptide sequences; and using the training protein sequences and the training peptide sequences to train a set of numerical parameters of a presentation model, the presentation model providing a plurality of numerical likelihoods that a peptide sequence from an infected cell is presented by one or more MHC alleles on the surface of the infected cell.

[0108] The presentation model can represent the dependency between the presence of a specific pair of one MHC allele and a specific amino acid at a specific position in a peptide sequence and the probability of presenting this peptide sequence comprising a specific amino acid at a specific position on the surface of an infected cell by the specific one MHC allele of the pair.

[0109] The methods disclosed herein can also include selecting a subset of antigens (e.g., infectious organism-derived antigens), wherein the subset of antigens is selected because each antigen has an increased probability of being presented on the cell surface of an infected cell relative to one or more different antigens.

[0110] The methods disclosed herein may also include selecting a subset of antigens (e.g., infectious organism-derived antigens), wherein the subset of antigens is selected because each antigen has an increased likelihood of being able to induce a disease-specific immune response in a subject relative to one or more different antigens.

[0111] The methods disclosed herein may also include selecting a subset of antigens (e.g., infectious organism-derived antigens), wherein the subset of antigens is selected because each antigen has an increased probability of being able to be presented to naive T cells by a professional antigen presenting cell (APC), optionally wherein the APC is a dendritic cell (DC), relative to one or more different antigens.

[0112] The methods disclosed herein can also include selecting a subset of antigens (e.g., infectious organism-derived antigens), wherein the subset of antigens is selected because each antigen has a reduced likelihood of being inhibited by central or peripheral tolerance relative to one or more different antigens.

[0113] The methods disclosed herein may also include selecting a subset of antigens (e.g., infectious organism-derived antigens), wherein the subset of antigens is selected because each antigen has a reduced likelihood of being able to induce an autoimmune response to normal tissues of the subject relative to one or more different antigens.

[0114] The methods disclosed herein may also include selecting a subset of antigens (e.g., infectious organism-derived antigens), wherein the subset of antigens is selected because each antigen has a reduced likelihood of being differentially post-translationally modified in an infected cell compared to an APC, optionally wherein the APC is a dendritic cell (DC).

[0115] Unless otherwise indicated, the implementation of the methods herein will adopt conventional protein chemistry, biochemistry, recombinant DNA technology and pharmacological methods that are within the skill of the art. Such techniques are fully described in the literature. See, for example, TECreighton, Proteins: Structures and Molecular Properties (WH Freeman and Company, 1993); A Lehninger, Biochemistry (Worth Publishers, Inc., new in this issue); Sambrook et al., Molecular Cloning: A Laboratory Manual (2nd edition, 1989); Methods In Enzymology (S. Colowick and N. Kaplan, Academic Press, Inc.); Remington's Pharmaceutical Sciences, 18th edition (Easton, Pennsylvania: Mack Publishing Company, 1990); Carey and Sundberg Advanced Organic Chemistry 3rd edition (Plenum Press) Volumes A and B (1992).

[0116] IV. Antigens

[0117] Antigens can include nucleotides or polypeptides. For example, antigens can be RNA sequences encoding polypeptide sequences. Therefore, antigens that can be used in vaccines can include nucleotide sequences or polypeptide sequences. Antigens can be derived from nucleotide sequences or polypeptide sequences of infectious organisms. The polypeptide sequences of infectious organisms include but are not limited to pathogen-derived peptides, virus-derived peptides, bacteria-derived peptides, fungus-derived peptides and / or parasite-derived peptides. Infectious organisms include but are not limited to severe acute respiratory syndrome-related coronaviruses (SARS), severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), Ebola virus, HIV, hepatitis B virus (HBV), influenza, hepatitis C virus (HCV), human papillomavirus (HPV), cytomegalovirus (CMV), chikungunya virus, respiratory syncytial virus (RSV), dengue virus, orthomyxoviridae virus, tuberculosis, pan-coronavirus, herpes simplex virus infection (HSV), influenza, metapneumovirus (MPV) and parainfluenza virus (PIV).

[0118] Disclosed herein are isolated peptides comprising infectious disease organism-specific antigens or epitopes determined by methods disclosed herein, peptides comprising known infectious disease organism-specific antigens or epitopes, and mutant polypeptides or fragments thereof determined by methods disclosed herein. Antigenic peptides can be described in the context of their coding sequences, wherein the antigen includes a nucleotide sequence (e.g., DNA or RNA) encoding a related polypeptide sequence.

[0119] One or more polypeptides encoded by the antigen nucleotide sequence may include at least one of the following: with an affinity to MHC with an IC50 value lower than 1000 nM, for MHC class I peptides of 8-15 in length, 8, 9, 10, 11, 12, 13, 14 or 15 amino acids, in or near the peptide there are sequence motifs that promote proteasome cleavage, and there are sequence motifs that promote TAP transport. For MHC class II peptides of 6-30 in length, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 or 30 amino acids, in or near the peptide there are sequence motifs that promote HLA binding by cleavage of an extracellular or lysosomal protease (e.g., cathepsin) or HLA-DM catalysis.

[0120] One or more antigens may be presented on the surface of the infected cell.

[0121] The one or more antigens can be immunogenic in a subject having an infection, eg, capable of eliciting a T cell response or a B cell response in the subject.

[0122] In the context of vaccine generation for subjects suffering from an infection, one or more antigens that induce an autoimmune response in the subject may be excluded from consideration.

[0123] The size of at least one antigenic peptide molecule can include, but is not limited to, about 5, about 6, about 7, about 8, about 9, about 10, about 11, about 12, about 13, about 14, about 15, about 16, about 17, about 18, about 19, about 20, about 21, about 22, about 23, about 24, about 25, about 26, about 27, about 28, about 29, about 30, about 31, about 32, about 33, about 34, about 35, about 36, about 37, about 38, about 39, about 40, about 41, about 42, about 43, about 44, about 45, about 46, about 47, about 48, about 49, about 50, about 60, about 70, about 80, about 90, about 100, about 110, about 120 or more amino acid molecule residues, and any range derivable therein. In a specific embodiment, the antigenic peptide molecule is equal to or less than 50 amino acids.

[0124] Antigenic peptides and polypeptides may be: 15 residues or less in length for MHC class I, and typically consist of about 8 to about 11 residues, particularly 9 or 10 residues; and 6-30 residues, inclusive, for MHC class II.

[0125] If desired, longer peptides can be designed in several ways. In one case, when the presentation potential of the peptide at the HLA allele is predicted or known, the longer peptide can be composed of one of the following: (1) a single presented peptide extending 2-5 amino acids toward the N-terminus and C-terminus of each corresponding gene product; (2) a concatenation of some or all of the presented peptide with the respective extension sequence. In another case, when sequencing reveals the presence of a longer (>10 residues) epitope sequence in infected cells, the longer peptide is composed of the following: (3) the entire stretch of the epitope sequence present in infected cells - thereby bypassing the need to select the strongest HLA-presenting shorter peptide by computational or in vitro testing. In both cases, the use of longer peptides allows endogenous processing by patient cells and can lead to more efficient antigen presentation and induction of T cell responses. The longer peptide can also be a full-length protein, protein subunit, protein domain, and combinations thereof of a peptide expressed in an infectious organism.

[0126] Antigenic peptides and polypeptides can be presented on HLA proteins. In some aspects, antigenic peptides and polypeptides are presented on HLA proteins with greater affinity than wild-type peptides. In some aspects, antigenic peptides or polypeptides can have an IC50 of at least less than 5000nM, at least less than 1000nM, at least less than 500nM, at least less than 250nM, at least less than 200nM, at least less than 150nM, at least less than 100nM, at least less than 50nM or less.

[0127] In some aspects, the antigenic peptides and polypeptides do not induce an autoimmune response and / or provoke immune tolerance when administered to a subject.

[0128] Compositions comprising at least two or more antigenic peptides are also provided. In some embodiments, the composition contains at least two different peptides. At least two different peptides can be derived from the same polypeptide. Different polypeptides refer to peptides that vary in length, amino acid sequence, or both. These peptides are derived from any polypeptide known or found to be expressed in infectious organisms.

[0129] Antigenic peptides and polypeptides with desired activity or characteristics can be modified to provide certain desired attributes, such as improved pharmacological characteristics, while increasing or at least maintaining substantially all biological activities of unmodified peptides to bind desired MHC molecules and activate appropriate T cells. For example, antigenic peptides and polypeptides can be subjected to various changes, such as conservative or non-conservative substitutions, wherein such changes may provide certain advantages in their use, such as improved MHC binding, stability or presentation. Conservative substitutions mean that a biologically and / or chemically similar amino acid residue is replaced with another amino acid residue, for example, a hydrophobic residue is replaced with another hydrophobic residue, or a polar residue is replaced with another polar residue. Substitutions include combinations such as Gly, Ala; Val, Ile, Leu, Met; Asp, Glu; Asn, Gln; Ser, Thr; Lys, Arg; and Phe, Tyr. The impact of single amino acid substitutions can also be detected using D-amino acids. Such modifications can be made using well-known peptide synthesis procedures, such as those described in Merrifield, Science 232:341-347 (1986), Barany & Merrifield, The Peptides, Gross & Meienhofer, eds. (NY, Academic Press), pp. 1-284 (1979); and Stewart & Young, Solid Phase Peptide Synthesis, (Rockford, Ill., Pierce), 2nd edition (1984).

[0130] Modification of peptides and polypeptides with various amino acid mimetics or non-natural amino acids can be particularly useful for increasing the in vivo stability of the peptides and polypeptides. Stability can be determined in a variety of ways. For example, peptidases and various biological media (such as human plasma and serum) have been used to test stability. See, for example, Verhoef et al., Eur. J. Drug Metab Pharmacokin. 11: 291-302 (1986). The half-life of a peptide can be conveniently determined using a 25% human serum (v / v) assay. The protocol is generally as follows. Combined human serum (AB type, non-heat inactivated) is defatted by centrifugation before use. Next, the serum is diluted to 25% with RPMI tissue culture medium and used to test peptide stability. At predetermined time intervals, a small amount of the reaction solution is taken out and added to a 6% trichloroacetic acid aqueous solution or ethanol. The turbid reaction sample is cooled (4° C.) for 15 minutes and then rotated to granulate the precipitated serum protein. The presence of the peptide is then determined by reverse phase HPLC using stability-specific chromatographic conditions.

[0131] Peptides and polypeptides can be modified to provide the desired attributes except improving serum half-life. For example, by being connected to a sequence containing at least one epitope that can induce T helper cell response, the ability of peptides to induce CTL activity can be enhanced. Immunogenic peptide / T helper cell conjugates can be connected by spacer molecules. Spacers are usually composed of relatively small neutral molecules such as amino acids or amino acid mimetics that are substantially uncharged under physiological conditions. Spacers are usually selected from other neutral spacers of, for example, Ala, Gly or non-polar amino acids or neutral polar amino acids. It should be understood that the optional spacer does not have to be composed of the same residues, and therefore can be a hetero-oligomer or a homo-oligomer. When present, the spacer will usually be at least one or two residues, more generally three to six residues. Alternatively, the peptide can be connected to the T helper peptide without a spacer.

[0132] The antigenic peptide can be linked to the T helper peptide directly or through a spacer at the amino or carboxyl terminus of the peptide. The amino terminus of the antigenic peptide or the T helper peptide can be acylated. Exemplary T helper peptides include tetanus toxoid 830-843, influenza 307-319, and malarial circumsporozoite 382-398 and 378-389.

[0133] Proteins or peptides can be prepared by any technique known to those skilled in the art, including expressing proteins, polypeptides or peptides by standard molecular biology techniques, isolating proteins or peptides from natural sources, or chemically synthesizing proteins or peptides. Previously, nucleotides and protein, polypeptide and peptide sequences corresponding to various genes have been disclosed, and can be found in computerized databases known to those of ordinary skill in the art. One such database is the Genbank and GenPept databases of the National Center for Biotechnology Information located at the website of the National Institutes of Health. The coding region of a known gene can be amplified and / or expressed using techniques disclosed herein or known to those of ordinary skill in the art. Alternatively, various commercial preparations of proteins, polypeptides and peptides are known to those skilled in the art.

[0134] On the other hand, antigens include nucleic acids (e.g., polynucleotides) encoding antigenic peptides or parts thereof. Polynucleotides can be, for example, DNA, cDNA, PNA, CNA, RNA (e.g., mRNA), single-stranded and / or double-stranded, or natural or stable forms of polynucleotides, such as polynucleotides with a thiophosphate backbone, or a combination thereof, and they may or may not contain introns. On the other hand, an expression vector capable of expressing a polypeptide or part thereof is provided. Expression vectors for different cell types are well known in the art and can be selected without excessive experiments. In general, DNA is inserted into an expression vector, such as a plasmid, with appropriate orientation and a correct expression reading frame. If necessary, DNA can be connected to appropriate transcription and translation regulatory control nucleotide sequences that can be recognized by the desired host, but such controls are generally available in expression vectors. The vector is then introduced into the host by standard techniques. Guidance can be found in, for example, Sambrook et al. (1989) Molecular Cloning, A Laboratory Manual, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY.

[0135] IV. Exemplary Therapeutic Agents

[0136] As disclosed herein, selected infectious disease derived antigens (e.g., selected by predicting the likelihood of candidate antigen presentation) can be used to develop therapeutic agents that, if administered to a patient, can induce an immune response to an infectious disease. Example therapeutic agents include vaccine compositions, compositions comprising T cell receptors (TCRs) or chimeric antigen receptors (CARs) bound to one or more selected infectious disease derived antigens, and antibodies that exhibit binding specificity to one or more selected infectious disease derived antigens. More details of example therapeutic agents are described herein.

[0137] IV.A. Vaccine Composition

[0138] Also disclosed herein is an immunogenic composition, such as a vaccine composition, capable of eliciting a specific immune response (e.g., an immune response to an infectious disease). The vaccine composition typically comprises a plurality of infectious disease-derived antigens, such as infectious disease-derived antigens selected using the methods described herein. The vaccine composition may also be referred to as a vaccine.

[0139] The vaccine may contain between 1 and 30 peptides, i.e. 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 or 30 different peptides; 6, 7, 8, 9, 10 11, 12, 13 or 14 different peptides; or 12, 13 or 14 different peptides. The vaccine may contain 1 to 100 or more nucleotide sequences, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75 7, 8, 9, 10, 11, 12, 13, or 14 different nucleotide sequences, or 12, 13, or 14 different nucleotide sequences. The vaccine may contain 1 to 30 antigen sequences, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59 , 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100 or more different antigen sequences, 6, 7, 8, 9, 10, 11, 12, 13, or 14 different antigen sequences, or 12, 13, or 14 different antigen sequences.

[0140] In one embodiment, different peptides and / or polypeptides or nucleotide sequences encoding them are selected so that the peptides and / or polypeptides are able to associate with different MHC molecules, such as different MHC class I molecules. In some aspects, a vaccine composition comprises coding sequences for peptides and / or polypeptides that are able to associate with the most frequently occurring MHC class I molecules. Thus, a vaccine composition may comprise different fragments that are able to associate with at least 2, preferably at least 3, or at least 4, preferred MHC class I molecules.

[0141] The vaccine composition is capable of enhancing a specific cytotoxic T cell response and / or a specific helper T cell response.

[0142] The vaccine composition may further comprise an adjuvant and / or a carrier. Examples of useful adjuvants and carriers are given below. The composition may be associated with a carrier (such as, for example, a protein) or an antigen presenting cell (such as, for example, a dendritic cell (DC) capable of presenting the peptide to a T cell).

[0143] An adjuvant is any substance that is incorporated into a vaccine composition to increase or otherwise alter the immune response to an antigen. A carrier may be a scaffold structure, such as a polypeptide or polysaccharide, to which an antigen can associate. Optionally, the adjuvant is conjugated in a covalent or non-covalent manner.

[0144] The ability of an adjuvant to increase the immune response to an antigen is usually manifested by a significant or substantial increase in immune-mediated reactions, or a reduction in disease symptoms. For example, an increase in humoral immunity is usually manifested as a significant increase in antibody titers produced against the antigen, and an increase in T cell activity is usually manifested as an increase in cell proliferation, or cytotoxicity, or cytokine secretion. An adjuvant can also alter the immune response, for example by changing a primarily humoral or Th response to a primarily cellular or Th response.

[0145] Suitable adjuvants include, but are not limited to, 1018ISS, alum, aluminum salts, Amplivax, AS15, BCG, CP-870,893, CpG7909, CyaA, dSLIM, GM-CSF, IC30, IC31, Imiquimod, ImuFact IMP321, IS Patch, ISS, ISCOMATRIX, JuvImmune, LipoVac, MF59, monophosphoryl lipid A, Montanide IMS1312, Montanide ISA206, Montanide ISA 50V, Montanide ISA-51, OK-432, OM-174, OM-197-MP-EC, ONTAK, PepTel vector system, PLG microparticles, resiquimod, SRL172, virosomes and other virus-like particles, YF-17D, VEGF trap, R848, β-glucan, Pam3Cys, Aquila's QS21 stimulator (Aquila Biotech, Worcester, Mass., USA) (derived from saponins, mycobacterial extracts and synthetic bacterial cell wall mimics) and other proprietary adjuvants such as Ribi's Detox. Quil or Superfos. Adjuvants such as incomplete Freund's adjuvant or GM-CSF are useful. Several dendritic cell-specific immunoadjuvants (e.g., MF59) and their preparation have been described previously (Dupuis M et al., Cell Immunol. 1998; 186(1): 18-27; Allison AC; Dev Biol Stand. 1998; 92: 3-11). Cytokines may also be used. Several cytokines have been directly associated with influencing the migration of dendritic cells to lymphoid tissues (e.g., TNF-α), accelerating the maturation of dendritic cells into potent antigen-presenting cells for T lymphocytes (e.g., GM-CSF, IL-1, and IL-4) (U.S. Pat. No. 5,849,589, specifically incorporated herein by reference in its entirety), and acting as immunoadjuvants (e.g., IL-12) (Gabrilovich DI et al., J Immunother Emphasis Tumor Immunol. 1996(6): 414-418).

[0146] CpG immunostimulatory oligonucleotides have also been reported to enhance the effects of adjuvants in a vaccine setting.Other TLR binding molecules, such as RNA binding to TLR 7, TLR 8 and / or TLR 9, may also be used.

[0147] Other examples of useful adjuvants include, but are not limited to, chemically modified CpG (e.g., CpR, Idera), poly(I:C) (e.g., poly i:CI2U), non-CpG bacterial DNA or RNA, and immunologically active small molecules and antibodies such as cyclophosphamide, sunitinib, bevacizumab, celebrex, NCX-4016, sildenafil, tadalafil, vardenafil, sorafinib, XL-999, CP-547632, pazopanib, ZD2171, AZD2171, ipilimumab, tremilimumab, and SC58175, which may act therapeutically and / or as an adjuvant. The amounts and concentrations of adjuvants and additives can be readily determined by a skilled artisan without undue experimentation. Additional adjuvants include colony stimulating factors, such as granulocyte macrophage colony stimulating factor (GM-CSF, sargramostim).

[0148] The vaccine composition may contain more than one different adjuvant. In addition, the therapeutic composition may contain any adjuvant substance, including any one or a combination of the above substances. It is also contemplated that the vaccine and adjuvant may be administered together or separately in any appropriate order.

[0149] The carrier (or excipient) may exist independently of the adjuvant. The function of the carrier may be, for example, to increase the molecular weight, in particular of the mutant, in order to increase activity or immunogenicity, to confer stability, to increase biological activity or to increase serum half-life. In addition, the carrier may help present the peptide to T cells. The carrier may be any suitable carrier known to those skilled in the art, such as a protein or an antigen presenting cell. The carrier protein may be, but is not limited to, keyhole limpet hemocyanin, serum proteins such as transferrin, bovine serum albumin, human serum albumin, thyroglobulin or ovalbumin, immunoglobulins or hormones such as insulin or palmitic acid. For human immunization, the carrier is generally a physiologically acceptable carrier that is acceptable to the human body and safe. However, tetanus toxoid and / or diphtheria toxoid are suitable carriers. Alternatively, the carrier may be a dextran, such as agarose gel.

[0150] Cytotoxic T cells (CTLs) recognize antigens in the form of peptides bound to MHC molecules rather than the complete foreign antigens themselves. The MHC molecules themselves are located on the cell surface of antigen presenting cells. Therefore, if there is a trimeric complex of peptide antigens, MHC molecules and APCs, activation of CTLs is possible. Accordingly, if not only peptides are used to activate CTLs, but also APCs with each MHC molecule are additionally added, the immune response can be enhanced. Therefore, in some embodiments, the vaccine composition additionally contains at least one antigen presenting cell.

[0151] Infectious disease-derived antigens may also be included in vaccine platforms based on viral vectors, such as vaccinia, fowlpox, self-replicating alphaviruses, Maraba virus, adenovirus (see, e.g., Tatsis et al., Adenoviruses, Molecular Therapy (2004) 10, 616-629), or lentiviruses, including but not limited to second, third, or mixed second / third generation lentiviruses and any generation of recombinant lentiviruses designed to target specific cell types or receptors (see, e.g., Hu et al., Immunization Delivered by Lentiviral Vectors for Cancer and Infectious Diseases, Immunol Rev. (2011) 239(1):45-61, Sakuma et al., Lentiviral vectors: basic to translational, Biochem J. (2012) 443(3):603-18, Cooper et al., Rescue of splicing-mediated intronloss maximizes expression in lentiviral vectors containing the humanubiquitin C promoter, Nucl. Acids Res. (2015) 43 (1): 682-690, Zufferey et al., Self-Inactivating Lentivirus Vector for Safe and Efficient In Vivo Gene Delivery, J. Virol. (1998) 72 (12): 9873-9880). According to the packaging capacity of the above-mentioned virus-based vaccine platform, the method can deliver one or more nucleotide sequences encoding one or more antigenic peptides. According to the description herein, a variety of other vaccine vectors that can be used for therapeutic administration or antigen immunization will be clear to those skilled in the art, such as Salmonella typhi vectors, etc.

[0152] IV.A.1. Considerations in vaccine design and production

[0153] Truncal peptides, i.e., those peptides presented by all or most tumor subclones, will be preferentially included in the vaccine. Optionally, if there are no truncal peptides predicted to be presented with high probability and immunogenic, or if the number of truncal peptides predicted to be presented with high probability and immunogenic is small enough that additional non-truncal peptides can be included in the vaccine, the peptides can be further prioritized by estimating the number of tumor subclones and identifying and selecting peptides so as to maximize the number of tumor subclones covered by the vaccine.

[0154] Additional candidate antigens can still be used for vaccine inclusion that vaccine technology can support. In addition, there may still be uncertainty in various aspects of antigen analysis, and there may be trade-offs between the different characteristics of candidate vaccine antigens. Therefore, instead of the predetermined filters in each step of the selection process, an integrated multidimensional model can be considered, which places the candidate antigens in a space with at least the following axes and uses an integrated approach to optimize the selection.

[0155] 1. Autoimmunity or tolerance risk (germline risk) (lower autoimmunity risk is usually preferred)

[0156] 2. The probability of sequencing artifacts (lower artifact probability is usually preferred)

[0157] 3. Probability of immunogenicity (higher probability of immunogenicity is usually preferred)

[0158] 4. Probability of submission (higher probability of submission is usually preferred)

[0159] 5. Gene expression (higher expression is usually preferred)

[0160] 6. HLA gene coverage (the greater the number of HLA molecules involved in the collective presentation of antigens, the lower the probability that the tumor will escape immune attack through downregulation or mutation of HLA molecules)

[0161] In various embodiments, vaccine design involves the following steps: 1) determining a collection of epitopes, 2) aligning the epitopes to a reference proteome to determine an initial footprint, 3) ranking the remaining epitopes based on coverage and cassette size cost, 4) including the top ranked epitopes, and 5) repeating steps (3) and (4) until a maximum cassette design is reached or all epitopes are included.

[0162] With reference to step (1), it is determined that the set of epitopes may include the set of short amino acid sequences. In various embodiments, the length of the short amino acid sequence is 8 to 12 amino acids. In various embodiments, the length of the short amino acid sequence is 9 to 11 amino acids. In various embodiments, the length of the short amino acid sequence is 8 amino acids, 9 amino acids, 10 amino acids, 11 amino acids or 12 amino acids. In various embodiments, the set of epitopes includes epitopes that may be presented by MHC alleles using methods disclosed herein (e.g., using presentation models disclosed herein). In various embodiments, the set of epitopes includes epitopes that may be presented by MHC alleles that are publicly demonstrated. Example epitopes that may be publicly demonstrated as possible presentations can be found in the Immune Epitope Database (IEDB) and / or Los Alamos National Laboratory (Los Alamos National Laboratory) HIV CD8 T cell epitope database. For a more detailed description of the Immune Epitope Database (IEDB), see Vita R, et al., The Immune Epitope Database (IEDB): 2018 Update. Nucleic Acids Res. October 24, 2018, each of which is incorporated herein by reference in its entirety. For a more detailed description of the Los Alamos National Laboratory HIV CD8 T cell epitope database, see Llano, A., et al., The 2019 Optimal HIV CTL epitopes update: growing variations inepitope length and HLA restriction. HIV Molecular Immunology 2019, 3-27, which is incorporated herein by reference in its entirety.

[0163] In various embodiments, the collection of epitopes includes both: 1) epitopes determined as likely to be presented by MHC alleles using the methods disclosed herein (e.g., using the presentation models disclosed herein) and 2) epitopes that have been publicly demonstrated to be likely to be presented by MHC alleles. In specific embodiments, the collection of validated epitopes includes epitopes determined to be likely to be presented by class I MHC alleles.

[0164] Step (2) involves aligning one or more epitopes in the collection of epitopes with a reference proteome. Figure 1B, which shows an example cassette design method according to an embodiment. Here, one or more epitopes are aligned with a reference proteome to generate an initial footprint. In various embodiments, one or more epitopes are validated epitopes (e.g., epitopes that have been publicly demonstrated to be likely presented by MHC alleles). In various embodiments, two or more parameters controlling the size of the initial footprint are used to limit the initial footprint. For example, the parameters may include a minimum epitope length (min 长度 ), the minimum number of overlapping epitopes required (min 重叠 ), and / or have a minimum number of overlapping epitopes (min 重叠 ) minimum epitope length (min 长度 ) ratio (min 比例 In various embodiments, the minimum epitope length (min 长度 ) is at least 2 amino acids. In various embodiments, the minimum epitope length is at least 3 amino acids, at least 4 amino acids, at least 5 amino acids, at least 6 amino acids, at least 7 amino acids, at least 8 amino acids, at least 9 amino acids, or at least 10 amino acids. In various embodiments, the minimum number of overlapping epitopes required (min 重叠 ) is at least 2 overlapping epitopes. In various embodiments, the minimum number of overlapping epitopes required (min 重叠 ) are at least 3 overlapping epitopes, at least 4 overlapping epitopes, at least 5 overlapping epitopes, at least 6 overlapping epitopes, at least 7 overlapping epitopes, at least 8 overlapping epitopes, at least 9 overlapping epitopes, or at least 10 overlapping epitopes.

[0165] Step (3) involves sorting the remaining epitopes according to coverage and box size cost. Specifically, coverage is expressed as c, which refers to the increased population coverage provided by the epitope if the epitope is contained in the box. In various embodiments, the coverage provided by a given epitope is calculated based on the frequency of the haplotype covering at least one allele associated with the epitope in the reference population. Example reference populations can include any ancestry population: African Americans (AFA), Hispanics (HIS), Asian-Pacific Islanders (API), and Europeans (EUR). Example reference population data can be found in the US National Marrow Donor Program (US National Marrow Donor Program) (e.g., Bioinformatics Be The ). refer to Figure 1C, which shows an example epitope sequence (AQTKILPR) and an example allele. Here, coverage is calculated based on two alleles (e.g., A*01:01 and B*08:01). The total coverage represents the sum of haplotypes containing one or both of the two alleles (e.g., A*01:01 and B*08:01). The cost of the box size is represented as f, which refers to the total increase in box size due to the addition of the epitope. In various embodiments, the epitopes are sorted according to coverage and box size cost so that the best sorted epitopes provide the best coverage improvement at the lowest box size cost.

[0166] Step (4) involves incorporating the top ranked epitopes into the cassette design. Figure 1B As shown in "Extension - Iteration 1" in Figure 2, additional epitopes (such as the top ranked epitopes) are added to extend the initial footprint.

[0167] Step (5) involves further repeating the ranking step (3) among the remaining epitopes in the collection, and further repeating step (4) to include the top ranked epitopes in the cassette design. Figure 1B In the "Extension-2nd Iteration" in , additional epitopes, such as the top ranked epitopes determined during this iteration, may also be included to further extend the footprint. Steps (3) and (4) may continue to be repeated until a maximum cassette design is reached or until all epitopes are included in the cassette.

[0168] IV.T cell receptor (TCR) and / or chimeric antigen receptor (CAR)

[0169] In addition, the present invention also discloses a T cell receptor (TCR) designed to bind to an antigen predicted to be presented on the cell surface. TCR can be separated and purified. In most T cells, TCR is a heterodimeric polypeptide having an α chain and a β chain, which are encoded by TRA and TRB, respectively. The α chain generally comprises an α variable region encoded by TRAV, an α connecting region encoded by TRAJ, and an α constant region encoded by TRAC. The β chain generally comprises a β variable region encoded by TRBV, a β diversity region encoded by TRBD, a β connecting region encoded by TRBJ, and a β constant region encoded by TRBC. The TCR-α chain is generated by VJ recombination of αV and J fragments, and the β chain receptor is generated by V(D)J recombination of βV, D and J fragments. Additional TCR diversity is derived from connection diversity. At each junction, several bases may be missing and other bases (called N and P nucleotides) may be added. In a few T cells, TCR includes γ chain and δ chain. TCRγ chain is generated by VJ recombination, while TCRδ chain is generated by V(D)J recombination (Kenneth Murphy, Paul Travers and Mark Walport, Janeway's Immunology 7th edition, Garland Science, 2007, which is incorporated herein by reference in its entirety). The antigen binding site of TCR generally contains six complementary determining regions (CDRs). The α chain contributes three CDRs, namely alpha ("α") CDR1, αCDR2 and αCDR3. The β chain also contributes three CDRs: beta ("β") CDR1, βCDR2 and βCDR3. In general, αCDR3 and βCDR3 are the regions most affected by V(D)J recombination and explain most of the variations in the TCR library.

[0170] TCR can be designed to specifically recognize antigens disclosed herein, such as infectious disease-derived antigens predicted to be presented on the cell surface. In various embodiments, TCR can also be membrane-bound, for example, on cells such as T cells or natural killer (NK) cells. Therefore, TCR can be used in situations corresponding to soluble antibodies and / or membrane-bound CARs.

[0171] Any of the TCRs disclosed herein may comprise an alpha variable ("V") segment, an alpha joining ("J") segment, optionally an alpha constant region, a beta variable ("V") segment, optionally a beta diversity ("D") segment, a beta joining ("J") segment, and optionally a beta constant region.

[0172] In some embodiments, the TCR or CAR is a recombinant TCR or CAR. The recombinant TCR or CAR may include any TCR determined herein, but includes one or more modifications. Exemplary modifications, such as amino acid substitutions, are described herein. The amino acid substitutions described herein can be made with reference to the IMGT nomenclature and amino acid numbering on www.imgt.org.

[0173] Recombinant TCR or CAR can be human TCR or CAR, including a full human sequence, such as a natural human sequence. Recombinant TCR or CAR may retain its natural human variable domain sequence, but contain modifications to the α constant region, β constant region, or α and β constant regions. Such modifications to the TCR constant region can improve TCR assembly and expression of TCR gene therapy by, for example, driving preferential pairing of exogenous TCR chains.

[0174] In some embodiments, the α and β constant regions are modified by replacing the mouse constant region sequences with the entire human constant region sequences. Such "murine" TCRs and methods for making them are described in Cancer Res. 2006 Sep 1; 66(17): 8878-86, which is hereby incorporated by reference in its entirety.

[0175] In some embodiments, the α and β constant regions are modified by making one or more amino acid substitutions in the human TCRα constant (TRAC) region, the TCRβ constant (TRBC) region, or the TRAC and TRAB regions, wherein the amino acid substitutions exchange specific human residues for mouse residues (human→mouse amino acid exchanges). The one or more amino acid substitutions in the TRAC region may include a Ser substitution at residue 90, an Asp substitution at residue 91, a Val substitution at residue 92, a Pro substitution at residue 93, or any combination thereof. The one or more amino acid substitutions in the human TRBC region may include a Lys substitution at residue 18, an Ala substitution at residue 22, an Ile substitution at residue 133, a His substitution at residue 139, or any combination thereof. Such targeted amino acid substitutions are described in J Immunol 2010 Jun 1, 184(11)6223-6231, which is hereby incorporated by reference in its entirety.

[0176] In some embodiments, human TRAC contains an Asp substitution at residue 210, and human TRBC contains a Lys substitution at residue 134. Such substitutions may promote the formation of salt bridges between the α and β chains and the formation of disulfide bonds between the TCR chains. These targeted substitutions are described in J Immunol 2010 Jun 1, 184 (11) 6232-6241, which is hereby incorporated by reference in its entirety.

[0177] In some embodiments, the human TRAC and human TRBC regions are modified to contain introduced cysteines that can improve preferential pairing of foreign TCR chains by forming additional disulfide bonds. For example, human TRAC may contain a Cys substitution at residue 48, while human TRBC may contain a Cys substitution at residue 57, as described in Cancer Res. 2007 Apr 15; 67(8): 3898-903 and Blood. 2007 Mar 15; 109(6): 2331-8, which are hereby incorporated by reference in their entirety.

[0178] The recombinant TCR or CAR may contain additional modifications to the α and β chains.

[0179] In some embodiments, the α and β chains are modified by linking their extracellular domains to a complete human CD3ζ (CD3-zeta) molecule. Such modifications are described in J Immunol June 1, 2008, 180(11)7736-7746; Gene Ther. 2000 August; 7(16):1369-77; and The Open Gene Therapy Journal, 2011, 4:11-22, which are hereby incorporated by reference in their entirety.

[0180] In some embodiments, the alpha chain is modified by introducing hydrophobic amino acid substitutions in the transmembrane region of the alpha chain as described in J Immunol 2012 Jun 1, 188(11) 5538-5546; which is hereby incorporated by reference in its entirety.

[0181] The alpha or beta chain can be modified by altering any of the N-glycosylation sites in the amino acid sequence as described in J Exp Med. 2009 Feb 16;206(2):463-475; hereby incorporated by reference in its entirety.

[0182] The α chain and the β chain may each comprise a dimerization domain, such as a heterologous dimerization domain. Such heterologous domains may be leucine zippers, 5H3 domains, or hydrophobic proline-rich inverted domains, or other similar morphologies as known in the art. In one example, the α and β chains may be modified by introducing 30-mer fragments into the carboxyl termini of the α and β extracellular domains, wherein the fragments selectively associate to form a stable leucine zipper. Such modifications are described in PNAS 1994 November 22. 91 (24) 11408-11412; https: / / doi.org / 10.1073 / pnas.91.24.11408; the document is hereby incorporated by reference in its entirety.

[0183] The TCRs identified herein may be modified to include mutations that result in increased affinity or half-life, such as those described in WO2012 / 013913, which is hereby incorporated by reference in its entirety.

[0184] Recombinant TCR or CAR can be a single-chain TCR (scTCR). Such scTCR may include an α chain variable region sequence fused to the N-terminal of the TCR α chain constant region extracellular sequence, a TCR β chain variable region fused to the N-terminal of the TCR β chain constant region extracellular sequence, and a joint sequence connecting the C-terminal of the α fragment and the N-terminal of the β fragment (or vice versa). In some embodiments, the constant region extracellular sequences of the α and β fragments of scTCR are connected by disulfide bonds. In some embodiments, the length of the joint sequence and the position of the disulfide bond make the variable region sequences of the α and β fragments substantially oriented to each other as in the natural α and β T cell receptors. Exemplary scTCRs are described in U.S. Patent No. 7,569,664, which is hereby incorporated by reference as a whole.

[0185] In some cases, the variable regions of the scTCR can be covalently linked by a short peptide linker, such as described in Gene Therapy, Vol. 7, pp. 1369-1377 (2000). The short peptide linker can be a serine-rich linker or a glycine-rich linker. For example, the linker can be (Gly4Ser)3, as described in Cancer Gene Therapy (2004) 11, 487–496, which is incorporated by reference in its entirety.

[0186] The recombinant TCR or antigen binding fragment thereof can be expressed as a fusion protein. For example, the TCR or antigen binding fragment thereof can be fused to a toxin. Such fusion proteins are described in Cancer Res. 2002 Mar 15;62(6):1757-60. The TCR or antigen binding fragment thereof can be fused to an antibody Fc region. Such fusion proteins are described in J Immunol 2017 May 1, 198(1 Suppl)120.9.

[0187] The antigen recognition domain of a receptor (such as a TCR or a CAR) can be connected to one or more intracellular signaling components, such as a signaling component activated by a simulation of an antigen receptor complex (such as a TCR complex), and / or signaling via another cell surface receptor. For example, TCR or CAR can be connected to one or more transmembrane and / or intracellular signaling domains. In some embodiments, the transmembrane domain is fused with the extracellular domain. In one embodiment, a transmembrane domain that is naturally associated with one of the domains in a receptor (e.g., a CAR) is used. In some cases, a transmembrane domain is selected or modified by amino acid substitution to avoid such domains from binding to the transmembrane domain of the same or different surface membrane proteins, thereby minimizing the interaction with other members of the receptor complex.

[0188] In some embodiments, the membrane-spanning domain is derived from natural or synthetic sources. In the case of natural sources, in some aspects the domain is derived from any membrane-bound protein or membrane-spanning protein. The membrane-spanning region includes α, β or ζ chains derived from (i.e., at least comprising the following membrane-spanning region) T cell receptors, CD28, CD3ε, CD45, CD4, CD5, CDS, CD9, CD16, CD22, CD33, CD37, CD64, CD80, CD86, CD 134, the membrane-spanning region of CD137 and / or CD154. Or, in some embodiments, the membrane-spanning domain is synthetic. In some aspects, the synthetic membrane-spanning domain mainly includes hydrophobic residues, such as leucine and valine. In some aspects, the triplet of phenylalanine, tryptophan and valine will be present at each end of the synthetic membrane-spanning domain. In some embodiments, connection is completed by a joint, a spacer and / or a membrane-spanning domain.

[0189] Among the intracellular signaling domains are those that mimic or approximate signals through native antigen receptors, through combinations of such receptors with co-stimulatory receptors, and / or through co-stimulatory receptors alone. In some embodiments, there is a short oligo- or polypeptide linker, such as a linker between 2 and 10 amino acids in length, such as a linker containing glycine and serine, e.g., a glycine-serine doublet, and the linker forms a connection between the transmembrane domain of the receptor and the cytoplasmic signaling domain.

[0190] Receptors, such as CAR, may include at least one intracellular signaling component. In some embodiments, the receptor includes an intracellular component of the TCR complex, such as a TCR CD3 chain that mediates T cell activation and cytotoxicity, such as a CD3 ζ chain. For example, HLA-peptide binding ABP (e.g., TCR or CAR) is connected to one or more cell signaling modules. In some embodiments, the cell signaling module includes a CD3 transmembrane domain, a CD3 intracellular signaling domain, and / or other CD transmembrane domains. In some embodiments, a receptor, such as a TCR or CAR, also includes a portion of one or more additional molecules (such as Fc receptor γ, CD8, CD4, CD25, or CD16). For example, in some aspects, TCR or CAR includes a chimeric molecule between CD3 ζ or Fc receptor γ and CD8, CD4, CD25, or CD16.

[0191] In some embodiments, after linking TCR or CAR, the cytoplasmic domain or intracellular signaling domain of the receptor activates at least one normal effector function or response of an immune cell (e.g., a T cell engineered to express a receptor). For example, in some cases, the receptor induces the function of a T cell, such as cytolytic activity or T helper cell activity, such as secretion of cytokines or other factors. In some embodiments, for example, if a truncated portion of an intracellular signaling domain of an antigen receptor component or a co-stimulatory molecule conducts an effector function signal, it is used to replace a complete immunostimulatory chain. In some embodiments, one or more intracellular signaling domains include a cytoplasmic sequence of a T cell receptor (TCR), and in some aspects also include a cytoplasmic sequence of a co-receptor that synergizes with the receptor in a natural environment to initiate signaling after antigen receptor engagement, and / or any derivatives or variants of such molecules, and / or any synthetic sequence with the same functional capabilities.

[0192] In the case of natural TCR, full activation generally requires not only signal transduction by TCR, but also co-stimulatory signals. Therefore, in some embodiments, in order to promote full activation, the receptor also includes a component for generating a secondary or co-stimulatory signal. In other embodiments, the receptor does not include a component for generating a co-stimulatory signal. In some aspects, additional receptors are expressed in the same cell and provide a component for generating a secondary or co-stimulatory signal.

[0193] T cell activation is described in some aspects as being mediated by two different classes of cytoplasmic signaling sequences: those that trigger antigen-dependent primary activation through the TCR (first cytoplasmic signaling sequences) and those that act in an antigen-independent manner to provide a second or costimulatory signal (second cytoplasmic signaling sequences). In some aspects, the receptor includes one or two such signaling components.

[0194] In some aspects, the receptor includes a primary cytoplasmic signaling sequence that regulates the primary activation of the TCR complex. The primary cytoplasmic signaling sequence that acts in a stimulatory manner may contain a signaling motif, which is called an activation motif or ITAM based on immunoreceptor tyrosine. Examples of ITAMs containing primary cytoplasmic signaling sequences include those derived from TCR or CD3ζ, FcRγ, FcRβ, CD3γ, CD3δ, CD3ε, CDS, CD22, CD79a, CD79b, and CD66d. In some embodiments, the cytoplasmic signaling molecule in CAR contains a cytoplasmic signaling domain, a portion thereof, or a sequence derived from CD3ζ.

[0195] In some embodiments, the receptor includes a signaling domain and / or a transmembrane portion of a co-stimulatory receptor (such as CD28, 4-1BB, OX40, DAP10, and ICOS). In some aspects, the same receptor includes both an activating component and a co-stimulatory component.

[0196] In some embodiments, the activation domain is contained in a receptor, and the co-stimulatory component is provided by another receptor that recognizes another antigen. In some embodiments, the receptor includes activation or stimulatory receptors and co-stimulatory receptors, all expressed on the same cell (see WO2014 / 055668). In some aspects, HLA-peptide targeting receptors are stimulatory or activating receptors; in other aspects, it is a co-stimulatory receptor. In some embodiments, the cell also includes inhibitory receptors (e.g., iCAR, see Fedorov et al., Sci. Transl. Medicine, 5 (215) (December 2013), such as receptors for antigens other than HLA-peptides, thereby weakening or inhibiting the activation signal delivered by HLA-peptide targeting receptors by the binding of inhibitory receptors to their ligands, for example, to reduce off-target effects.

[0197] In certain embodiments, the intracellular signaling domain comprises a CD28 transmembrane and signaling domain connected to a CD3 (e.g., CD3 ζ) intracellular domain. In some embodiments, the intracellular signaling domain comprises a chimeric CD28 and CD137 (4-1BB, TNFRSF9) costimulatory domain connected to a CD3 ζ intracellular domain.

[0198] In some embodiments, the receptor comprises one or more, e.g., two or more co-stimulatory domains and activation domains, e.g., primary activation domains, in the cytoplasmic portion. Exemplary receptors include intracellular components of CD3ζ, CD28, and 4-1BB.

[0199] In some embodiments, CAR (or other antigen receptors such as TCR) also includes a marker, such as a cell surface marker, which can be used to confirm the transduction or engineering of cells to express receptors, such as truncated forms of cell surface receptors, such as truncated EGFR (tEGFR). In some aspects, markers include all or part (e.g., truncated forms) of CD34, nerve growth factor receptor (NGFR) or epidermal growth factor receptor (e.g., tEGFR). In some embodiments, the nucleic acid encoding the marker is operably connected to a polynucleotide encoding a joint sequence (such as a cleavable joint sequence or a ribosome skipping sequence such as T2A). See WO2014031687. In some embodiments, the constructs of CAR and EGFRt separated by T2A ribosomal switches are introduced to express two proteins from the same construct, so that EGFRt can be used as a marker for detecting cells expressing such a construct. In some embodiments, markers and optional joint sequences can be any markers and joint sequences disclosed in the published patent application number WO2014031687. For example, the marker can be a truncated EGFR (tEGFR), optionally linked to a linker sequence, such as a T2A ribosomal skipping sequence.

[0200] In some embodiments, a marker is a molecule, such as a cell surface protein, that does not naturally occur on a T cell or does not naturally occur on the surface of a T cell, or a portion thereof.

[0201] In some embodiments, the molecule is a non-self molecule, such as a non-self protein, ie, a molecule that is not recognized as "self" by the immune system of the host into which the cell is adoptively transferred.

[0202] In some embodiments, the marker does not provide a therapeutic function and / or does not produce other effects, except for being used as a marker for genetic engineering, for example, for selecting successfully engineered cells. In other embodiments, the marker can be a therapeutic molecule or a molecule that otherwise exerts some desired effect, such as a ligand of a cell encountered in vivo, such as a co-stimulatory or immune checkpoint molecule that enhances and / or suppresses the response of the cell after adoptive transfer and encountering a ligand.

[0203] TCR or CAR may comprise one or modified synthetic amino acids to replace one or more naturally occurring amino acids. Exemplary modified amino acids include, but are not limited to, aminocyclohexanecarboxylic acid, norleucine, α-aminodecanoic acid, homoserine, S-acetylaminomethylcysteine, trans-3-hydroxyproline and trans-4-hydroxyproline, 4-aminophenylalanine, 4-nitrophenylalanine, 4-chlorophenylalanine, 4-carboxyphenylalanine, (3-phenylserine (3-hydroxyphenylalanine, phenylglycine, α-naphthylalanine, cyclohexylalanine, cyclohexylglycine, dihydroindole-2-yl)-1-hydroxy ... 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalanine, 4-nitrophenylalan -carboxylic acid, 1,2,3,4-tetrahydroisoquinoline-3-carboxylic acid, aminomalonic acid, aminomalonic acid monoamide, N'-benzyl-N'-methyl-lysine, N',N'-dibenzyl-lysine, 6-hydroxylysine, ornithine, α-aminocyclopentanecarboxylic acid, α-aminocyclohexanecarboxylic acid, α-aminocycloheptanecarboxylic acid, α-(2-amino-2-norbornane)-carboxylic acid, α,γ-diaminobutyric acid, α,γ-diaminopropionic acid, homophenylalanine and α-tert-butylglycine.

[0204] In some cases, CAR is referred to as first generation, second generation and / or third generation CAR. In some aspects, first generation CAR is a CAR that only provides CD3 chain induction signal when antigen binds; in some aspects, second generation CAR is a CAR that provides such signal and costimulatory signal, such as a CAR comprising intracellular signaling domains from costimulatory receptors such as CD28 or CD137; in some aspects, third generation CAR is a CAR comprising multiple costimulatory domains of different costimulatory receptors in some aspects.

[0205] In some embodiments, the chimeric antigen receptor comprises an extracellular portion containing a TCR or fragment as described herein. In some aspects, the chimeric antigen receptor comprises an extracellular portion containing a TCR or fragment as described herein and an intracellular signaling domain. In some embodiments, the intracellular domain contains an ITAM. In some aspects, the intracellular signaling domain comprises the signaling domain of the ζ chain of the CD3ζ (CD3) chain. In some embodiments, the chimeric antigen receptor comprises a transmembrane domain connecting the extracellular domain and the intracellular signaling domain.

[0206] In some aspects, the transmembrane domain contains the transmembrane portion of CD28. The extracellular domain and the transmembrane can be directly or indirectly connected. In some embodiments, the extracellular domain and the transmembrane are connected by a spacer, such as any spacer described herein. In some embodiments, the chimeric antigen receptor contains the intracellular domain of the T cell costimulatory molecule, such as between the transmembrane domain and the intracellular signaling domain. In some aspects, the T cell costimulatory molecule is CD28 or 41BB.

[0207] In some embodiments, CAR contains TCR, such as TCR fragments, transmembrane domains that are or contain the transmembrane portion of CD28 or its functional variant, and intracellular signaling domains containing the signaling portion of CD28 or its functional variant and the signaling portion of CD3ζ or its functional variant. In some embodiments, CAR contains TCR, such as TCR fragments, transmembrane domains that are or contain the transmembrane portion of CD28 or its functional variant, and intracellular signaling domains containing the signaling portion of 4-1BB or its functional variant and the signaling portion of CD3ζ or its functional variant. In some such embodiments, the receptor also includes a spacer containing a portion of an Ig molecule (such as a human Ig molecule), such as an Ig hinge, such as an IgG4 hinge, such as a spacer with only a hinge.

[0208] In some embodiments, the transmembrane domain of a receptor, such as a TCR or CAR, is a transmembrane domain of human CD28 or a variant thereof, such as the 27 amino acid transmembrane domain of human CD28 (Accession No.: P10747.1).

[0209] In some embodiments, the chimeric antigen receptor contains the intracellular domain of a T cell co-stimulatory molecule. In some aspects, the T cell co-stimulatory molecule is CD28 or 41BB.

[0210] In some embodiments, the intracellular signaling domain comprises an intracellular co-stimulatory signaling domain of human CD28 or a functional variant or portion thereof, such as a 41 amino acid domain thereof and / or such a domain having a LL to GG substitution at position 186-187 of a native CD28 protein. In some embodiments, the intracellular domain comprises an intracellular co-stimulatory signaling domain of 41BB or a functional variant or portion thereof, such as a 42 amino acid cytoplasmic domain of human 4-1BB (Accession No. Q07011.1) or a functional variant or portion thereof.

[0211] In some embodiments, the intracellular signaling domain comprises a human CD3 zeta stimulatory signaling domain or a functional variant thereof, such as the 112AA cytoplasmic domain of isoform 3 of human CD3 zeta (Accession No.: P20963.2) or a CD3 zeta signaling domain as described in U.S. Pat. No. 7,446,190 or U.S. Pat. No. 8,911,993.

[0212] In some aspects, the spacer contains only the hinge region of IgG, such as a hinge containing only IgG4 or IgG1. In other embodiments, the spacer is an Ig hinge connected to the CH2 and / or CH3 domains, such as an IgG4 hinge. In some embodiments, the spacer is an Ig hinge connected to the CH2 and CH3 domains, such as an IgG4 hinge. In some embodiments, the spacer is an Ig hinge connected to the CH3 domain, such as an IgG4 hinge. In some embodiments, the spacer is or comprises a glycine-serine-rich sequence or other flexible linker, such as a known flexible linker.

[0213] For example, in some embodiments, the CAR includes a TCR or a fragment thereof (such as any one of the HLA-peptide-specific TCRs), a spacer (such as any one of the spacers containing an Ig hinge), a CD28 transmembrane domain, a CD28 intracellular signaling domain, and a CD3 zeta signaling domain. In some embodiments, the CAR includes a TCR or a fragment thereof (such as any one of the HLA-peptide-specific TCRs), a spacer (such as any one of the spacers containing an Ig hinge), a CD28 transmembrane domain, a CD28 intracellular signaling domain, and a CD3 zeta signaling domain.

[0214] IV.C. Methods of Engineering Cells with TCR and / or CAR

[0215] Also provided are methods, nucleic acids, compositions and kits for expressing receptors comprising TCRs, CARs, etc. and for producing genetically engineered cells expressing such TCRs, CARs, etc. Genetic engineering generally involves the introduction of nucleic acids encoding recombinant or engineered components into cells, such as by retroviral transduction, transfection or transformation.

[0216] In some embodiments, gene transfer is achieved by first stimulating the cells, such as by combining them with a stimulus that induces a response (such as proliferation, survival and / or activation), as measured, for example, by expression of cytokines or activation markers, and then transducing the activated cells and expanding them in culture to numbers sufficient for clinical use.

[0217] In some cases, the overexpression of stimulatory factors (e.g., lymphokines or cytokines) may be toxic to the subject. Therefore, in some cases, engineered cells include fragments that cause cells to be susceptible to negative selection in vivo, such as when applied in adoptive immunotherapy. For example, in some aspects, cells are engineered so that they can be eliminated due to changes in the patient's in vivo conditions that receive their application. Negative optional phenotypes may be due to the insertion of genes that confer sensitivity to the administered agent (e.g., compound). Negative optional genes include herpes simplex virus type I thymidine kinase (HSV-I TK) genes (Wigler et al., Cell II: 223, 1977) that confer ganciclovir sensitivity; cell hypoxanthine phosphoribosyltransferase (HPRT) genes, cell adenine phosphoribosyltransferase (APRT) genes, bacterial cytosine deaminase (Mullen et al., Proc. Natl. Acad. Sci. USA. 89: 33 (1992)).

[0218] In some aspects, cells are further engineered to promote the expression of cytokines or other factors. Various methods of introducing genetically engineered components (such as antigen receptors (e.g., TCR)) are well known and can be used together with the provided methods and compositions. Exemplary methods include those methods for transferring nucleic acids encoding receptors, including by viruses (e.g., retroviral or lentiviral transduction), transposons, nuclease-mediated gene editing (e.g., CRISPR, TALEN, meganuclease or ZFN editing systems) and electroporation. For example, nuclease-mediated gene editing (particularly for editing T cells) is described in more detail in international applications WO / 2018 / 232356 and PCT / US2018 / 058230, which are incorporated herein by reference for all purposes.

[0219] In some embodiments, the recombinant nucleic acid is transferred into cells using recombinant infectious viral particles such as, for example, vectors derived from simian virus 40 (SV40), adenovirus, adeno-associated virus (AAV). In some embodiments, the recombinant nucleic acid is transferred into T cells using recombinant lentiviral vectors or retroviral vectors such as gamma-retroviral vectors (see, e.g., Koste et al. (2014) Gene Therapy 2014 Apr 3. doi: 10.1038 / gt.2014.25; Carlens et al. (2000) Exp Hematol 28(10): 1137-46; Alonso-Camino et al. (2013) Mol Ther Nucl Acids 2, e93; Park et al., Trends Biotechnol. 2011 Nov. 29(11): 550-557).

[0220] In some embodiments, the retroviral vector has a long terminal repeat sequence (LTR), such as a retroviral vector derived from Moloney murine leukemia virus (MoMLV), myeloproliferative sarcoma virus (MPSV), mouse embryonic stem cell virus (MESV), mouse stem cell virus (MSCV), spleen focus forming virus (SFFV) or adeno-associated virus (AAV). Most retroviral vectors are derived from mouse retroviruses. In some embodiments, retroviruses include those derived from any avian or mammalian cell source. Retroviruses are usually amphotropic, which means that they can infect host cells of multiple species including humans. In one embodiment, the gene to be expressed replaces the retroviral gag, pol and / or env sequences. Many exemplary retroviral systems have been described (e.g., U.S. Pat. Nos. 5,219,740; 6,207,453; 5,219,740; Miller and Rosman (1989) BioTechniques 7:980-990; Miller, AD (1990) Human Gene Therapy 1:5-14; Scarpa et al., (1991) Virology 180:849-852; Burns et al., (1993) Proc. Natl. Acad. Sci. USA 90:8033-8037; and Boris-Lawrie and Temin (1993) Cur. Opin. Genet. Develop. 3:102-109).

[0221] Methods of lentiviral transduction are known. Exemplary methods are described in, for example, Wang et al. (2012) J. Immunother. 35(9):689-701; Cooper et al. (2003) Blood. 101:1637-1644; Verhoeyen et al. (2009) Methods Mol Biol. 506:97-114; and Cavalieri et al. (2003) Blood. 102(2):497-505.

[0222] In some embodiments, the recombinant nucleic acid is transferred into T cells by electroporation (see, e.g., Chicaybam et al. (2013) PLoS ONE 8(3):e60298; Van Tedeloo et al. (2000) Gene Therapy 7(16):1431-1437; and Roth et al. (2018) Nature 559:405–409). In some embodiments, the recombinant nucleic acid is transferred into T cells by transposition (see, e.g., Manuri et al. (2010) Hum Gene Ther 21(4):427-437; Sharma et al. (2013) Molec Ther Nucl Acids 2, e74; and Huang et al. (2009) Methods Mol Biol 506:115-126). Other methods for introducing and expressing genetic material in immune cells include calcium phosphate transfection (e.g., as described in Current Protocols in Molecular Biology, John Wiley & Sons, New York. NY), protoplast fusion; cationic liposome-mediated transfection; microprojectile bombardment promoted by tungsten particles (Johnston, Nature, 346:776-777 (1990)); and strontium phosphate DNA coprecipitation (Brash et al., Mol. Cell Biol., 7:2031-2034 (1987)).

[0223] Other methods and vectors for transfer of nucleic acids encoding recombinant products are those described in, for example, International Patent Application Publication No.: WO2014055668 and US Patent No. 7,446,190.

[0224] In the additional nucleic acid, for example, the genes for introduction are those for the following: genes for improving therapeutic efficacy (such as by promoting the viability and / or function of the transferred cells); genes for providing genetic markers to select and / or evaluate cells (such as for evaluating survival or localization in vivo); genes for improving safety (e.g., by making cells susceptible to negative selection in vivo), such as Lupton SD et al., Mol. and Cell Biol., 11: 6 (1991); and Riddell et al., Human Gene Therapy 3: 319-338 (1992) as described; also see Lupton et al., publication PCT / US91 / 08442 and PCT / US94 / 05601, describing the use of bifunctional selectable fusion genes derived from fusion dominant positive selectable markers and negative selectable markers. See, for example, Riddell et al., U.S. Pat. No. 6,040,177, columns 14-17.

[0225] IV.D. Antibodies or Antigen-binding Fragments

[0226] Disclosed herein are antibodies or antigen-binding fragments designed to bind to an antigen predicted to be presented on the cell surface. In some embodiments, the antibodies or antigen-binding fragments provided herein comprise a light chain. In some aspects, the light chain is a kappa light chain. In some aspects, the light chain is a lambda light chain.

[0227] In some embodiments, the antibodies or antigen-binding fragments provided herein comprise a heavy chain. In some aspects, the heavy chain is IgA. In some aspects, the heavy chain is IgD. In some aspects, the heavy chain is IgE. In some aspects, the heavy chain is IgG. In some aspects, the heavy chain is IgM. In some aspects, the heavy chain is IgG1. In some aspects, the heavy chain is IgG2. In some aspects, the heavy chain is IgG3. In some aspects, the heavy chain is IgG4. In some aspects, the heavy chain is IgA1. In some aspects, the heavy chain is IgA2.

[0228] In some embodiments, the antibodies or antigen-binding fragments provided herein comprise antibody fragments. In some embodiments, the antibodies or antigen-binding fragments provided herein consist of antibody fragments. In some embodiments, the antibodies or antigen-binding fragments provided herein consist essentially of antibody fragments. In some aspects, the antibody fragment is an Fv fragment. In some aspects, the antibody fragment is a Fab fragment. In some aspects, the antibody fragment is a F(ab')2 fragment. In some aspects, the antibody fragment is a Fab' fragment. In some aspects, the antibody fragment is a scFv (sFv) fragment. In some aspects, the antibody fragment is a scFv-Fc fragment. In some aspects, the antibody fragment is a fragment of a single domain ABP.

[0229] In some embodiments, the antibody fragments provided herein retain the ability to bind to a target, such as an infectious disease-derived antigen predicted to be presented by one or more HLA alleles on the surface of a cell, as measured by one or more assays or biological effects described herein. In some embodiments, the antibody fragments provided herein retain the ability to prevent an infectious disease-derived antigen from interacting with one or more of its ligands, as described herein.

[0230] The antibody fragments provided herein can be prepared by any suitable method, including the illustrative methods described herein or those known in the art. Suitable methods include recombinant techniques and proteolytic digestion of whole antibodies or antigen-binding fragments.

[0231] In some embodiments, the antibodies provided herein are monoclonal antibodies. For example, monoclonal antibodies can be obtained using a hybridoma method or using a library based on phage or yeast. The DNA encoding the monoclonal antibody can be easily separated and sequenced using conventional procedures. In some embodiments, the antibodies provided herein are polyclonal antibodies. In some embodiments, the antibodies provided herein comprise chimeric ABPs. In some embodiments, the antibodies provided herein consist of chimeric antibodies. In some embodiments, the antibodies provided herein consist essentially of chimeric antibodies. Chimeric antibodies can be prepared by any method known in the art. In some embodiments, chimeric antibodies are prepared by combining non-human variable regions (e.g., variable regions derived from mice, rats, hamsters, rabbits, or non-human primates such as monkeys) with human constant regions using recombinant technology.

[0232] In some embodiments, the antibodies provided herein comprise humanized antibodies. In some embodiments, the antibodies provided herein consist of humanized antibodies. In some embodiments, the antibodies provided herein consist essentially of humanized antibodies. Humanized antibodies can be generated by replacing most or all structural parts of non-human monoclonal antibodies with corresponding human antibody sequences.

[0233] In some embodiments, the antibodies provided herein comprise human antibodies. In some embodiments, the antibodies provided herein consist of human antibodies. In some embodiments, the antibodies provided herein consist essentially of human antibodies. Human antibodies can be generated by various techniques known in the art, such as by using transgenic animals (e.g., humanized mice), can be derived from phage display libraries, can be generated by in vitro activated B cells, or can be derived from yeast-based libraries.

[0234] In some embodiments, the antibodies provided herein comprise an alternative scaffold. In some embodiments, the antibodies provided herein consist of an alternative scaffold. In some embodiments, the antibodies provided herein consist essentially of an alternative scaffold. Any suitable alternative scaffold may be used. In some aspects, the alternative scaffold is selected from Adnectin TM 、iMab、 EETI-II / AGRP, Kunitz domain, thioredoxin peptide aptamer, DARPins, Affilins, Tetranectins, Fynomers, and Avimers. The alternative scaffolds provided herein can be prepared by any suitable method, including the illustrative methods described herein or those methods known in the art.

[0235] Also disclosed herein are isolated humanized, human or chimeric antibodies that compete with the antibodies disclosed herein for binding to the HLA-antigen complex.

[0236] In certain aspects, the antibody may comprise a human Fc region comprising at least one modification that reduces binding to a human Fc receptor. It is known that when an antibody is expressed in a cell, the antibody is modified after translation. Examples of post-translational modifications include cleavage of the lysine at the C-terminus of the heavy chain by carboxypeptidase; modification of the glutamine or glutamic acid at the N-terminus of the heavy chain and light chain to pyroglutamic acid by pyroglutamylation; glycosylation; oxidation; deamidation; and glycosylation, and such post-translational modifications are known to occur in various ABPs (see Journal of Pharmaceutical Sciences, 2008, Vol. 97, pp. 2426-2447, which is incorporated by reference in its entirety). In some embodiments, the antibody is a post-translationally modified antibody or antigen-binding fragment thereof. Examples of antibodies or antigen-binding fragments that have been post-translationally modified include antibodies or antigen-binding fragments that have been pyroglutamylated at the N-terminus of the heavy chain variable region and / or have been deleted for lysine at the C-terminus of the heavy chain. It is known in the art that such post-translational modifications due to N-terminal pyroglutamylation and C-terminal lysine deletion do not have any influence on the activity of the antibody or its fragment (Analytical Biochemistry, 2006, Vol. 348, pp. 24-39, which is incorporated by reference in its entirety).

[0237] In some embodiments, the antibodies provided herein are multispecific antibodies.

[0238] In some embodiments, the multispecific antibodies provided herein bind more than one antigen. In some embodiments, the multispecific antibodies bind 2 antigens (e.g., bispecific antibodies). In some embodiments, the multispecific antibodies bind 3 antigens. In some embodiments, the multispecific antibodies bind 4 antigens. In some embodiments, the multispecific antibodies bind 5 antigens.

[0239] In some embodiments, the multispecific ABP comprises an antigen binding domain (ABD) that specifically binds to an HLA-peptide target disclosed herein and an additional ABD that binds to an additional target antigen. Many multispecific antibody constructs are known in the art, and the antibodies provided herein can be provided in the form of any suitable multispecific construct. The multispecific antibodies provided herein can be prepared by any suitable method, including the illustrative methods described herein or those methods known in the art.

[0240] In certain embodiments, the antibodies provided herein include an Fc region. The Fc region may be wild type or a variant thereof. In certain embodiments, the antibodies provided herein include an Fc region having one or more amino acid substitutions, insertions or deletions compared to a naturally occurring Fc region. In some aspects, such substitutions, insertions or deletions produce antibodies with altered stability, glycosylation or other characteristics. In some aspects, such substitutions, insertions or deletions produce glycosylated antibodies.

[0241] In some embodiments, the Fc region is a variant Fc region. A "variant Fc region" or "engineered Fc region" comprises an amino acid sequence that is distinguished from a native sequence Fc region amino acid sequence by at least one amino acid modification (preferably one or more amino acid substitutions). Preferably, the variant Fc region has at least one amino acid substitution compared to the native sequence Fc region or the Fc region of a parent polypeptide, for example, there are about 1 to about 10 amino acid substitutions in the native sequence Fc region or the Fc region of a parent polypeptide, preferably about 1 to about 5 amino acid substitutions. The variant Fc region herein will preferably have at least about 80% homology with the native sequence Fc region and / or with the Fc region of a parent polypeptide, most preferably at least about 90% homology therewith, and more preferably at least about 95% homology therewith.

[0242] The term "Fc region-containing antibody" refers to an antibody comprising an Fc region. The C-terminal lysine (residue 447 according to the EU numbering system) of the Fc region can be removed, for example, during purification of the antibody or by recombinant engineering of the nucleic acid encoding the antibody. Therefore, the antibody with an Fc region according to the present invention may include an antibody with or without K447.

[0243] In some aspects, the Fc region of the antibodies provided herein is modified to produce antibodies with altered affinity for Fc receptors, or more immunologically inert antibodies. In some embodiments, the antibody variants provided herein have some but not all effector functions. For example, when the half-life of the antibody is important in vivo, but when certain effector functions (e.g., complement activation and ADCC) are unnecessary or harmful, such antibodies may be useful.

[0244] In some embodiments, the antibodies provided herein comprise one or more alterations that improve or reduce CIq binding and / or CDC.

[0245] In some embodiments, the antibodies provided herein comprise one or more alterations that increase half-life. In some embodiments, the antibodies comprise one or more non-Fc modifications that extend half-life.

[0246] In some embodiments, the multispecific antibody comprises one or more Fc modifications that promote heteromultimerization. In some embodiments, the Fc modifications comprise a collection of mutations that render homodimerization electrostatically unfavorable but heterodimerization favorable.

[0247] In some embodiments, Fc modifications include modifications in the CH3 sequence that affect the ability of the CH3 domain to bind an affinity agent (eg, Protein A).

[0248] V. Therapeutic Agents and Methods of Making

[0249] Also provided are methods of inducing an infectious organism-specific immune response in a subject, vaccinating against an infectious organism, treating and or ameliorating symptoms of infection associated with an infectious organism in a subject by administering one or more antigens (such as a plurality of antigens identified using the methods disclosed herein) to the subject.

[0250] In some aspects, the subject has been diagnosed with an infection or is at risk of an infection. The subject can be a human, dog, cat, horse, or any animal for which a specific immune response to an infectious agent is desired.

[0251] The antigen may be administered in an amount sufficient to induce a CTL response. The antigen may be administered in an amount sufficient to induce a T cell response. The antigen may be administered in an amount sufficient to induce a B cell response.

[0252] The antigens may be administered alone or in combination with other therapeutic agents.Any suitable therapeutic treatment for the particular infectious disease may be administered.

[0253] The optimal amount and optimal dosing regimen of each antigen included in the vaccine composition can be determined. For example, the antigen or its variant can be prepared for intravenous (iv) injection, subcutaneous (sc) injection, intradermal (id) injection, intraperitoneal (ip) injection, intramuscular (im) injection. The methods of injection include sc, id, ip, im and iv. The methods of DNA or RNA injection include id, im, sc, ip and iv. Other methods of administration of the vaccine composition are known to those skilled in the art.

[0254] Vaccines can be compiled so that the selection, quantity and / or amount of antigens present in the composition are tissue, infectious disease and / or patient specific. For example, the exact selection of peptides can be guided by the expression pattern of the parent protein in a given tissue, or by the mutation status of the patient. The selection can depend on the specific type of infectious disease, the type of organism that causes the infectious disease (e.g., pathogen, virus, bacteria, fungus or parasite), the state of the disease, the early treatment regimen, the immune status of the patient, and the HLA haplotype of the patient. In addition, vaccines can contain individualized components according to the personal needs of a specific patient. Examples include changing the selection of antigens according to the expression of antigens in a specific patient or adjusting secondary treatment after the first round or regimen of treatment.

[0255] Patients suitable for administering antigen vaccines can be determined by using various diagnostic methods (e.g., patient selection methods further described below). Patient selection can involve determining a mutation or expression pattern of one or more genes. In some cases, patient selection involves determining the haplotype of the patient. Various patient selection methods can be performed in parallel, for example, sequencing diagnosis can simultaneously determine the mutation and haplotype of the patient. Various patient selection methods can be performed in sequence, for example, a diagnostic test determines a mutation, a separate diagnostic test determines the haplotype of the patient, and wherein each test can be the same (e.g., two high-throughput sequencing) or different (e.g., a high-throughput sequencing and another Sanger sequencing) diagnostic method.

[0256] For compositions used as vaccines against infectious diseases, antigens with similar normal self peptides that are expressed in large amounts in normal tissues may be avoided or present in small amounts in the compositions described herein. On the other hand, if it is known that a patient's infected cells express a large amount of a certain antigen, the corresponding pharmaceutical composition for treating the infection may be present in large amounts and / or may include more than one antigen specific for that particular antigen or pathway to that antigen.

[0257] Compositions comprising antigens can be applied to individuals who have already suffered from infectious diseases. In therapeutic applications, compositions are applied to patients in an amount sufficient to cause effective CTL responses to infectious biological antigens and to cure or at least partially stagnate symptoms and / or complications. The amount sufficient to achieve this goal is defined as a "therapeutically effective dose". The effective amount for this purpose will depend on, for example, the composition, the mode of administration, the stage and severity of the disease being treated, the patient's weight and overall health, and the judgment of the prescribing physician. It should be remembered that compositions can generally be used for serious infectious disease states, i.e. life-threatening or potentially life-threatening situations. In such cases, in view of the minimization of foreign matter and the relatively non-toxic nature of antigens, it is possible and felt necessary for the treating physician to administer significantly excessive amounts of these compositions.

[0258] For therapeutic use, administration can begin when infection is detected or before infection is detected. Booster doses are then administered until at least symptoms are substantially resolved and for a period of time thereafter or until immunity is deemed to be provided (e.g., generation of memory B cells or T cell populations, or antigen-specific B cells or antibodies).

[0259] Pharmaceutical compositions (e.g., vaccine compositions) for therapeutic treatment are intended for parenteral, topical, nasal, oral or topical administration. Pharmaceutical compositions can be administered parenterally, such as intravenously, subcutaneously, intradermally or intramuscularly. The composition can be administered to the site of infection to induce a local immune response to infection. Disclosed herein are compositions for parenteral administration, comprising a solution of an antigen, and the vaccine composition is dissolved or suspended in an acceptable carrier (e.g., an aqueous carrier). A variety of aqueous carriers can be used, such as water, buffered water, 0.9% saline, 0.3% glycine, hyaluronic acid, etc. These compositions can be sterilized by conventional, well-known sterilization techniques, or they can be sterilized by filtration. The resulting aqueous solution can be packaged for use as is or in a lyophilized form, and the lyophilized preparation is combined with a sterile solution before administration. The composition may contain pharmaceutically acceptable auxiliary substances as required to approximate physiological conditions, such as pH adjusters and buffers, tonicity adjusters, wetting agents, etc., for example, sodium acetate, sodium lactate, sodium chloride, potassium chloride, calcium chloride, sorbitan monolaurate, triethanolamine oleate, etc.

[0260] Antigens can also be administered by liposomes, which target specific cell tissues, such as lymphoid tissue. Liposomes can also be used to extend the half-life. Liposomes include emulsions, foams, micelles, insoluble monolayers, liquid crystals, phospholipid dispersions, lamellar layers, etc. In these preparations, the antigen to be delivered is incorporated as a part of the liposome, either alone or in combination with molecules (such as monoclonal antibodies that are bound to CD45 antigens) that are bound to receptors that are ubiquitous in lymphoid cells, or in combination with other therapeutic or immunogenic compositions. Therefore, liposomes filled with the desired antigen can be directed to the site of lymphoid cells, and then the liposomes deliver the selected treatment / immunogenic composition there. Liposomes can be formed by standard vesicle-forming lipids, which typically include neutral or negatively charged phospholipids and sterols, such as cholesterol. The selection of lipids is typically guided by considering, for example, liposome size, acid instability, and the stability of liposomes in the bloodstream. There are a variety of methods for preparing liposomes as described, for example, in Szoka et al., Ann. Rev. Biophys. Bioeng. 9; 467 (1980), U.S. Patents 4,235,871, 4,501,728, 4,501,728, 4,837,028 and 5,019,369.

[0261] For targeting immune cells, the ligand incorporated into the liposomes may include, for example, antibodies or fragments thereof that are specific for cell surface determinants of the desired immune system cells. The liposomal suspension may be administered intravenously, locally, etc., at doses that vary, inter alia, depending on the mode of administration, the peptide being delivered, and the stage of the disease being treated.

[0262] For treatment or immunization purposes, nucleic acids encoding peptides and optionally one or more peptides described herein can also be administered to patients. Many methods are conveniently used to deliver these nucleic acids to patients. For example, nucleic acids can be directly delivered as "naked DNA". For example, this method is described in Wolff et al., Science 247:1465-1468 (1990) and U.S. Patents 5,580,859 and 5,589,466. Ballistic delivery can also be used to administer nucleic acids, as described in, for example, U.S. Patent No. 5,204,253. Particles consisting only of DNA can be administered. Alternatively, DNA can be adhered to particles (such as gold particles). Methods for delivering nucleic acid sequences can include viral vectors, mRNA vectors and DNA vectors, with or without electroporation.

[0263] Nucleic acids can also be delivered by complexing with cationic compounds (such as cationic lipids). Lipid-mediated gene delivery methods have been described in, for example, 9618372WOAWO 96 / 18372; 9324640WOAWO 93 / 24640; Mannino and Gould-Fogerite, BioTechniques6(7):682-691 (1988); U.S. Patent No. 5,279,833 Rose U.S. Patent No. 5,279,833; 9106309WOAWO 91 / 06309; and Felgner et al., Proc.Natl.Acad.Sci.USA 84:7413-7414 (1987).

[0264] Antigens may also be included in vaccine platforms based on viral vectors, such as vaccinia, fowlpox, self-replicating alphaviruses, Maraba virus, adenovirus (see, e.g., Tatsis et al., Adenoviruses, Molecular Therapy (2004) 10, 616-629), or lentiviruses, including but not limited to second, third, or mixed second / third generation lentiviruses and any generation of recombinant lentiviruses designed to target specific cell types or receptors (see, e.g., Hu et al., Immunization Delivered by Lentiviral Vectors for Cancer and Infectious Diseases, Immunol Rev. (2011) 239(1):45-61, Sakuma et al., Lentiviral vectors: basic to translational, Biochem J. (2012) 443(3):603-18, Cooper et al., Rescue of splicing-mediated intronloss maximizes expression in lentiviral vectors containing the humanubiquitin C promoter, Nucl. Acids Res. (2015) 43 (1): 682-690, Zufferey et al., Self-Inactivating Lentivirus Vector for Safe and Efficient In Vivo Gene Delivery, J. Virol. (1998) 72 (12): 9873-9880). According to the packaging capacity of the above-mentioned virus vector-based vaccine platform, this method can deliver one or more nucleotide sequences encoding one or more antigenic peptides.The sequence may be flanked by non-mutated sequences, may be separated by a linker, or may be preceded by one or more sequences that target subcellular compartments (see, e.g., Gros et al., Prospective identification of neoantigen-specific lymphocytes in the peripheral blood of melanoma patients, Nat Med. (2016) 22(4): 433-8, Stronen et al., Targeting of cancer neoantigens with donor-derived T cell receptor repertoires, Science. (2016) 352(6291): 1337-41, Lu et al., Efficient identification of mutated cancer antigens recognized by T cells associated with durable tumor regressions, Clin Cancer Res. (2014) 20(13): 3401-10). When introduced into a host, the infected cells express the antigen, thereby eliciting a host immune (e.g., CTL) response to the infectious disease-derived peptide. Vaccinia vectors and methods for immunization protocols are described, e.g., in U.S. Pat. No. 4,722,848. Another vector is BCG (bacillus Calmette-Guerin). BCG vectors are described in Stover et al. (Nature 351: 456-460 (1991)). Based on the description herein, a variety of other vaccine vectors that can be used for therapeutic administration or antigen immunization will be clear to those skilled in the art, such as Salmonella typhi vectors, etc.

[0265] The method of administering nucleic acids is to use a minigene construct encoding one or more epitopes. In order to generate DNA sequences encoding selected CTL epitopes for expression in human cells (minigenes), the amino acid sequences of these epitopes are reverse translated. The human codon usage table is used to guide the codon selection for each amino acid. These epitope encoding DNA sequences are directly adjacent to each other to create a continuous polypeptide sequence. In order to optimize expression and / or immunogenicity, additional elements can be incorporated into the minigene design. Examples of amino acid sequences that can be reverse translated and included in the minigene sequence include: helper T lymphocytes, epitopes, leader (signal) sequences, and endoplasmic reticulum retention signals. In addition, the MHC presentation of CTL epitopes can be improved by incorporating synthetic (e.g., polyalanine) or naturally occurring side sequences adjacent to the CTL epitopes. The minigene sequence can be converted to DNA by assembling oligonucleotides encoding the positive and negative strands of the minigene. Using well-known techniques, overlapping oligonucleotides (30-100 bases long) are synthesized, phosphorylated, purified, and annealed under appropriate conditions. The ends of the oligonucleotides can be connected using T4 DNA ligase. This synthetic minigene encoding the CTL epitope polypeptide can then be cloned into a desired expression vector.

[0266] Purified plasmid DNA for injection can be prepared using a variety of formulations. The simplest method is to redissolve lyophilized DNA in sterile phosphate-buffered saline (PBS). Various methods have been described, and new technologies can become available. As mentioned above, nucleic acids can be easily prepared with cationic lipids. In addition, glycolipids, fusogenic liposomes, peptides and compounds collectively referred to as protective, interactive, non-condensing (PINC) compounds can also be compounded with purified plasmid DNA to affect variables such as stability, intramuscular dispersion or transportation to specific organs or cell types.

[0267] Also disclosed is a method of making an infectious disease vaccine, comprising performing the steps of the method disclosed herein; and producing an infectious disease vaccine comprising a plurality of antigens or a subset of a plurality of antigens.

[0268] Antigens disclosed herein can be manufactured using methods known in the art. For example, methods for producing antigens disclosed herein or vectors (e.g., vectors comprising at least one sequence encoding one or more antigens) can include culturing host cells under conditions suitable for expressing antigens or vectors, wherein the host cells comprise at least one polynucleotide encoding the antigen or vector, and purifying the antigen or vector. Standard purification methods include chromatographic techniques, electrophoresis, immunology, precipitation, dialysis, filtration, concentration, and chromatofocusing techniques.

[0269] The host cell may include a Chinese hamster ovary (CHO) cell, a NS0 cell, a yeast or a HEK293 cell. The host cell may be transformed with one or more polynucleotides comprising at least one nucleic acid sequence encoding an antigen or a vector disclosed herein, optionally wherein the isolated polynucleotide additionally comprises a promoter sequence operably linked to the at least one nucleic acid sequence encoding an antigen or a vector. In certain embodiments, the isolated polynucleotide may be a cDNA.

[0270] VII. Submission Determination System

[0271] VII.A. System Overview

[0272] 1 is an overview of an environment for determining the likelihood of presentation of a peptide in a patient, according to one embodiment. Environment 100 provides context to introduce a presentation determination system 160, which itself includes a presentation information storage 165.

[0273] The presentation determination system 160 is one or more of the following: Figure 7 A computer model embodied in the computing system discussed receives a peptide sequence (e.g., an infectious disease-derived peptide sequence) associated with a set of MHC alleles and determines the likelihood that the infectious disease-derived peptide sequence is presented by one or more of the set of associated MHC alleles. This can be used in a variety of situations. One specific use case of the presentation determination system 160 is that it is capable of receiving a nucleotide sequence of a candidate antigen (e.g., a candidate antigen sequence 114) associated with a set of MHC alleles expressed by a patient 110, and determining the likelihood that the candidate antigen is presented by one or more associated MHC alleles of the patient 110 and / or induces an immunogenic response in the immune system of the patient 110. Those candidate antigens determined by the system 160 to have a high likelihood can be selected for development of a treatment 118 (e.g., inclusion in a vaccine, development of a TCR specific for a selected antigen, and / or development of an antibody exhibiting binding affinity for a selected antigen). Thus, if the therapeutic is administered, an anti-infectious disease immune response from the immune system of the patient 110 can be elicited.

[0274] The presentation determination system 160 determines the presentation probability through one or more presentation models (also referred to herein as multi-part presentation models). Specifically, the presentation model generates the probability of whether a given peptide sequence will be presented for a set of relevant MHC alleles, and is generated based on the presentation information stored in the memory 165. For example, the presentation model can generate the probability that the peptide sequence "YVYVADVAAK" is presented on the cell surface of the sample for the set of alleles HLA-A*02:01, HLA-B*07:02, HLA-B*08:03, HLA-C*01:04, HLA-A*06:03, HLA-B*01:04. The presentation information 165 contains information about whether the peptide binds to different types of MHC alleles so that those peptides are presented by the MHC alleles, which is determined in the model according to the position of the amino acids in the peptide sequence. The presentation model can predict, based on the presentation information 165, whether an unrecognized peptide sequence will be presented with a set of relevant MHC alleles.

[0275] VII.B. Submission of Information

[0276] Figure 2A and 2B A method of obtaining presentation information according to an embodiment is shown. In various embodiments, presentation information 165 includes two general information categories: allele interaction information and allele non-interaction information. Allele interaction information includes information that affects presentation of peptide sequences that depend on the type of MHC allele. Allele non-interaction information includes information that affects presentation of peptide sequences that are unrelated to the type of MHC allele.

[0277] VII.B.1. Allelic interaction information

[0278] Allele interaction information may include known determined peptide sequences presented by one or more determined MHC molecules from humans, mice, etc. It is noteworthy that this may or may not include data obtained from infectious disease samples. The presented peptide sequences can be determined from cells expressing a single MHC allele. In various embodiments, the presented peptide sequences are collected from monoallelic cell lines that are engineered to express predetermined MHC alleles and are subsequently exposed to synthetic proteins. The peptides presented on the MHC alleles are separated by techniques such as acid elution and determined by mass spectrometry. Figure 2A An example of this is shown, where an example peptide YEMFNDKS presented on a predetermined MHC allele HLA-A*01:01 was isolated and identified by mass spectrometry. Since the peptides were identified by cells engineered to express a single predetermined MHC protein, the direct association between the presented peptide and the MHC protein to which it binds is clearly known.

[0279] The peptide sequences presented can also be collected from cells expressing multiple MHC alleles. In humans, one cell expresses 6 different types of MHC molecules. Such peptide sequences presented can be determined from a multiple allele cell line engineered to express multiple predetermined MHC alleles. Such peptide sequences presented can also be determined from tissue samples, which can be from normal tissue samples, or tissue samples exposed to one of the pathogens, viruses, bacteria, fungi or parasites that can cause infectious diseases. In this case, MHC molecules can be immunoprecipitated from normal or infectious disease tissues. Peptides presented on multiple MHC alleles can be similarly separated by techniques such as acid elution and determined by mass spectrometry. Figure 2B An example of this is shown, where six example peptides YEMFNDKSF, HROEIFSHDFJ, FJIEJFOESS, NEIOREIREI, JFKSIFEMMSJDSSU and KNFLENFIESOFI presented on the identified MHC alleles HLA-A*01:01, HLA-A*02:01, HLA-B*07:02, HLA-B*08:01, HLA-C*01:03 and HLA-C*01:04 were isolated and identified by mass spectrometry. In contrast to monoallelic cell lines, the direct association between the presented peptide and the MHC protein to which it binds may be unknown, as the bound peptide is isolated from the MHC molecule before being identified.

[0280] Allele interaction information can also include mass spectrometry ion current, which depends on the concentration of the peptide-MHC molecule complex and the ionization efficiency of the peptide. The ionization efficiency varies in a sequence-dependent manner between peptides. In general, the ionization efficiency varies between peptides in about two orders of magnitude, while the concentration of the peptide-MHC complex varies in a larger range than it.

[0281] Allele interaction information can also include a measurement or prediction of binding affinity between a given MHC allele and a given peptide. One or more affinity models can generate such predictions. For example, presentation information 165 can include a prediction of a binding affinity of 1000 nM between peptide YEMFNDKSF and allele HLA-A*01:01. Peptides with IC50>1000 nm are rarely presented by MHC, and lower IC50 values ​​will increase the probability of presentation.

[0282] Allelic interaction information may also include measurements or predictions of MHC complex stability. One or more stability models may generate such predictions. For example, Figure 2B In the example shown, the presented information 165 may include a stability prediction for the molecule HLA-A*01:01 with a half-life of 1 h.

[0283] Allelic interaction information may also include measured or predicted rates of peptide-MHC complex formation reactions. Complexes formed at higher rates are more likely to be presented on the cell surface at high concentrations.

[0284] Allelic interaction information can also include the sequence and length of the peptide. MHC class I molecules generally prefer to present peptides between 8 and 15 peptides in length. 60-80% of the presented peptides are 9 in length.

[0285] Allelic interaction information can also include the presence of kinase sequence motifs on the antigen encoding peptide, and the absence or presence of specific post-translational modifications on the antigen encoding peptide. The presence of kinase motifs can affect the probability of post-translational modifications, which may enhance or interfere with MHC binding.

[0286] Allelic interaction information may also include proteins involved in post-translational modification processes, such as expression levels or activity levels of kinases (as measured or predicted by RNA seq, mass spectrometry or other methods).

[0287] Allelic interaction information may also include the probability of presentation of peptides with similar sequences in cells from other individuals expressing a particular MHC allele, which can be assessed by mass spectrometry proteomics or other means.

[0288] Allelic interaction information can also include the expression level of a particular MHC allele in the individual in question (e.g., as measured by RNA-seq or mass spectrometry). Peptides that bind most strongly to MHC alleles expressed at high levels are more likely to be presented than peptides that bind most strongly to MHC alleles expressed at low levels.

[0289] Allelic interaction information may also include the peptide sequence-independent probability of the overall antigen being encoded by a particular MHC allele presented in other individuals expressing the particular MHC allele.

[0290] Allele interaction information can also include the overall peptide sequence-independent probability of presentation by MHC alleles in the same family of molecules (e.g., HLA-A, HLA-B, HLA-C, HLA-DQ, HLA-DR, HLA-DP) in other individuals. For example, HLA-C molecules are generally expressed at lower levels than HLA-A or HLA-B molecules, and therefore, the probability of a peptide being presented by HLA-C is a priori less than the probability of a peptide being presented by HLA-A or HLA-B 11.

[0291] Allelic interaction information may also include protein sequences for specific MHC alleles.

[0292] Any of the MHC allele non-interaction information listed in the next section can also be modeled as MHC allele interaction information.

[0293] VII.B.2. Allele Non-Interaction Information

[0294] Allele-independent information may include C-terminal sequences flanking the antigen-encoding peptide within its source protein sequence. The C-terminal flanking sequence may affect the proteasomal processing of the peptide. However, the C-terminal flanking sequence is cleaved from the peptide by the proteasome before the peptide is transported to the endoplasmic reticulum and encounters the MHC allele on the cell surface. Therefore, the MHC molecule does not receive information about the C-terminal flanking sequence, and thus, the influence of the C-terminal flanking sequence cannot vary depending on the MHC allele type. For example, see again Figure 2B In the example shown in , the presentation information 165 may include the C-terminal flanking sequence FOEIFNDKSLDKFJI of the presenting peptide FJIEJFOESS determined from the source protein of the peptide.

[0295] Allele non-interaction information can also include mRNA quantitative measurement values. For example, mRNA quantitative data can be obtained from the same sample providing mass spectrometry training data. In one embodiment, mRNA quantitative measurement values ​​are determined by software tool RSEM. The detailed implementation of RSEM software tool can be found in Bo Li and Colin N.Dewey.RSEM:accurate transcript quantification from RNA-Seq data with or without areference genome.BMC Bioinformatics, 12:323, August 2011. In one embodiment, mRNA quantitative is measured in units of fragment number (FPKM) per kilobase transcript in every million positioning reads.

[0296] Allele-independent interaction information may also include the N-terminal sequence flanking the peptide in its source protein sequence.

[0297] In a specific embodiment, the allele-non-interaction information includes the C-terminal sequence flanking the peptide in its source protein sequence and the N-terminal sequence flanking the peptide in its source protein sequence.

[0298] Allelic non-interaction information can also include the presence of a protease cleavage motif in the peptide. Peptides containing a protease cleavage motif are less likely to be presented because they are more easily degraded by proteases and are therefore less stable within the cell.

[0299] Allele non-interaction information can also include the turnover rate of the source protein as measured in the appropriate cell type. A faster turnover rate (i.e., a shorter half-life) increases the probability of presentation; however, if measured in a dissimilar cell type, the predictive power of this feature is low.

[0300] Allele-non-interaction information may also include the length of the source protein.

[0301] Allele non-interaction information can also include the expression level of proteasome, immunoproteasome, thymic proteasome or other proteases. Different proteasomes have different cleavage site preferences. The cleavage preference of each type of proteasome will be given greater weight according to its expression level.

[0302] Allele non-interaction information can also include the expression of the source gene of the peptide (e.g., measured by RNA-seq or mass spectrometry). Peptides from genes with higher expression levels are more likely to be presented. Peptides from genes with undetectable expression levels can be excluded from consideration.

[0303] Allelic non-interaction information may also include the probability that the source mRNA encoding the antigen peptide undergoes nonsense-mediated decay as predicted by a nonsense-mediated decay model, such as the model from Rivas et al., Science 2015.

[0304] Allele-independent interaction information can also include typical tissue-specific expression of the source gene of the peptide during various stages of the cell cycle. A gene that is expressed at low levels overall (as measured by RNA sequencing or mass spectrometry proteomics) but is known to be expressed at high levels during a specific stage of the cell cycle may produce more presented peptides than a gene that is stably expressed at very low levels.

[0305] Allele non-interaction information can also include, for example, a comprehensive catalog of source protein features provided in uniProt or PDB http: / / www.rcsb.org / pdb / home / home.do. These features can include, in particular: secondary and tertiary structures, subcellular localization 11, gene ontology (GO) terms of proteins. Specifically, this information can contain annotations that work at the protein level, such as 5'UTR length, and annotations that work at the specific residue level, such as the helical motif between residues 300 and 310. These features can also include turn motifs, sheet motifs, and disordered residues.

[0306] Allelic non-interaction information may also include features describing the properties of the source protein domain containing the peptide, such as: secondary or tertiary structure (eg, alpha helix vs. beta sheet); alternative splicing.

[0307] Allelic non-interaction information may also include features describing the presence or absence of presentation hotspots at the position of the peptide in the peptide source protein.

[0308] Allele-independent interaction information may also include the probability of presentation of a peptide derived from the source protein of the peptide in question in other individuals (after adjusting for the expression level of the source protein in these individuals and the influence of different HLA types of those individuals).

[0309] Allelic non-interaction information can also include the probability that a peptide will not be detected or will be over-represented by mass spectrometry due to technical bias.

[0310] Allele-independent interaction information may also include the probability of a peptide binding to TAP or the measured or predicted binding affinity of a peptide to TAP. Peptides that are more likely to bind to TAP, or peptides that bind to TAP with higher affinity, are more likely to be presented.

[0311] Allele non-interaction information can also include the known function of the HLA allele, such as reflected by the HLA allele suffix. For example, the N suffix in the allele name HLA-A*24:09N indicates a null allele that is not expressed and therefore unlikely to present an epitope; a description of the complete HLA allele suffix nomenclature is available at https: / / www.ebi.ac.uk / ipd / imgt / hla / nomenclature / suffixes.html.

[0312] VII.C. Submission Determination System

[0313] Figure 3A 1 is a high-level block diagram illustrating computer logic components of a presentation determination system 160 according to one embodiment. In this example embodiment, the presentation determination system 160 includes a data management module 312, an encoding module 314, a training module 316, and a prediction module 320. The presentation determination system 160 also includes a training data memory 170A and a presentation model memory 175. Some embodiments of the model management system 160 have different modules than those described herein. Similarly, functionality may be allocated between modules in a manner different from that described herein.

[0314] VII.C.1. Data Management Module

[0315] The data management module 312 generates a set of training data according to the submitted information 165. Each set of training data contains multiple data instances, where each data instance i contains an independent variable z i The independent variable includes at least a presented or unpresented peptide sequence p i , one or more i Related related MHC alleles ai , and the dependent variable y representing the information presented to the determination system 160 of interest in the prediction of the new value of the independent variable i .

[0316] In a specific implementation described throughout the rest of this specification, the dependent variable y i is the indicator peptide p i Whether one or more relevant MHC alleles a i However, it should be understood that in other implementations, the dependent variable y i It can be said that the presentation determination system 160 intentionally determines the independent variable z i Any other type of information for making predictions. For example, in another implementation, the dependent variable y i It may also be a numerical value indicative of the mass spectrometer ion current determined for the data instance.

[0317] The peptide sequence p of data instance i i It is k i A sequence of amino acids, where k i can vary within a range between data instances i. For example, the range may be 8-15 for MHC class I, or 9-30 for MHC class II. In one particular implementation of system 160, all peptide sequences p in the training data set are i All may have the same length, for example, 9. The number of amino acids in the peptide sequence may vary depending on the type of MHC allele (e.g., MHC alleles in humans, etc.). i Indicates which MHC alleles are associated with the corresponding peptide sequence p i Relevant exists.

[0318] The data management module 312 may also include additional allele interaction variables, such as peptide sequences p contained in the training data 170. i and related MHC alleles a i Combined binding affinity b i and stability i For example, the training data 170 may contain peptide p i With a i The predicted binding affinity between each relevant MHC molecule indicated b i As another example, the training data 170 may contain i Stability prediction values ​​for each MHC allele indicated i .

[0319] In certain embodiments, the data management module 312 includes 1) indicating the peptide p iWhether one or more relevant MHC alleles a i present (as determined by mass spectrometry), and 2) a tag that is associated with the peptide sequence p i and related MHC alleles a i Combined additional binding affinity b i This enables the use of training data derived from both: 1) binding affinity data between peptide sequences and HLA alleles, and 2) elution peptide data from mass spectra representing presentation of the peptide sequences and HLA alleles to train a presentation model (e.g., a multipartite presentation model). Here, it may be preferred to use both types of data to train a multipartite presentation model, especially in situations where the amount of elution peptide data for a peptide sequence generated by mass spectrometry is limited.

[0320] In various embodiments, the peptide sequences (e.g., training peptide sequences) used to train the multi-part presentation model are infectious disease derived peptides. Therefore, the multi-part presentation model can be trained to accurately predict whether the infectious disease derived peptides are likely to be presented by one or more HLA alleles. In various embodiments, the peptide sequences (e.g., training peptide sequences) used to train the multi-part presentation model are human peptides. In various embodiments, the training peptide sequences used to train the multi-part presentation model include infectious disease derived peptides as well as human peptides. In such embodiments, if the number of infectious disease derived peptide sequences is limited, human peptides can represent additional training peptide sequences for training the multi-part presentation model.

[0321] In a specific embodiment, the training data for training the multi-part presentation model include the binding affinity data between the infectious disease derived peptide sequence and the HLA allele. In a specific embodiment, the training data for training the multi-part presentation model include the eluted human peptide data (e.g., human immunopeptidomics) from mass spectrometry, representing the presentation of human peptide sequences and HLA alleles. In a specific embodiment, the training data for training the multi-part presentation model include 1) the binding affinity data between the infectious disease derived peptide sequence and the HLA allele and 2) the eluted human peptide data (e.g., human immunopeptidomics) from mass spectrometry, representing the presentation of human peptide sequences and HLA alleles. When the eluted peptide data of the infectious disease derived peptide sequence is limited, this situation is beneficial. Therefore, the eluted human peptide data are used to supplement the training of the multi-part presentation model.

[0322] In various embodiments, the data management module 312 may also include allele non-interaction variables w i , such as with the peptide sequence p i Combined C-terminal flanking sequences and mRNA quantitative measurements.

[0323] The data management module 312 may also determine peptide sequences that are not presented by the MHC alleles to generate the training data 170. For example, this involves determining a "longer" sequence of the source protein that includes the presented peptide sequence prior to presentation. When the presentation information contains an engineered cell line, the data management module 312 determines a series of peptide sequences in the synthetic protein to which these cells are exposed that are not presented on the MHC alleles of the cells. When the presentation information contains a tissue sample, the data management module 312 determines the source protein from which the presented peptide sequence is derived, and determines a series of peptide sequences in the source protein that are not presented on the MHC alleles of the tissue sample cells.

[0324] In various embodiments, the data management module 312 can also artificially generate peptides with random amino acid sequences and identify the generated sequences as peptides that are not presented on the MHC allele. This can be achieved by randomly generating peptide sequences, so that the data management module 312 can easily generate a large amount of synthetic data of peptides that are not presented on the MHC allele. Since a small percentage of peptide sequences are actually presented by the MHC allele, the synthetically generated peptide sequences are likely not presented by the MHC allele even if they are included in the protein processed by the cell.

[0325] In various embodiments, the data management module 312 artificially generates peptides to balance the training data set, thereby minimizing the bias caused by the unbalanced training data set. For example, the training data set may have many peptides (M1) presented by the MHC allele and many peptides (M2) not presented by the MHC allele. In various embodiments, if the number of peptides not presented by the MHC allele is less than the number of peptides presented by the MHC allele in the training data set (e.g., M2 < M1), the data management module 312 artificially generates peptides not presented by the MHC allele to balance the training data set. In various embodiments, if the ratio of peptides not presented by the MHC allele to peptides presented by the MHC allele (M2 < M1) is less than a threshold value (N), the data management module 312 artificially generates peptides not presented by the MHC allele. The threshold value N can be any one of 1, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, or 100. In a specific embodiment, the threshold value N is 50. Therefore, if the ratio of peptides not presented by the MHC allele to peptides presented by the MHC allele (M2 < M1) is less than 50, the data management module 312 artificially generates peptides not presented by the MHC allele to further balance the training data set. In such embodiments, the balanced data set ensures that there are significantly more peptides not presented by the MHC allele in the training data set, so that the presentation model is fully trained to identify peptides not presented by the MHC allele.

[0326] In various embodiments, the training data set includes binding affinity data between infectious disease-derived peptide sequences and HLA alleles. Therefore, in such embodiments, if the ratio (M2<M1) of peptides not presented by MHC alleles (as shown by the binding affinity value) to peptides presented by MHC alleles (as shown by the binding affinity value) is less than a threshold value (N), the data management module 312 artificially generates peptides not presented by MHC alleles. In various embodiments, the training data set includes eluted human peptide data from mass spectrometry (e.g., human immunopeptidomics), indicating the presentation of human peptide sequences and HLA alleles. Therefore, in such embodiments, if the ratio (M2<M1) of peptides not presented by MHC alleles to peptides presented by MHC alleles is less than a threshold value (N), the data management module 312 artificially generates peptides not presented by MHC alleles.

[0327] In various embodiments, the training data set includes both: 1) binding affinity data between infectious disease derived peptide sequences and HLA alleles and 2) eluted human peptide data from mass spectrometry (e.g., human immunopeptidomics), representing the display of human peptide sequences and HLA alleles. Therefore, in such embodiments, the data management module 312 can artificially generate peptides in the binding affinity data and / or the eluted human peptide data. For example, if the ratio (M2<M1) of peptides not presented by MHC alleles (as shown by the binding affinity values) to peptides presented by MHC alleles (as shown by the binding affinity values) is less than a threshold value (N), the data management module 312 artificially generates peptides not presented by MHC alleles (as shown by the binding affinity values) to further balance the training data set.

[0328] As another example, if the ratio of peptides not presented by the MHC allele (as shown in the eluted human peptide data) to peptides presented by the MHC allele (as shown in the eluted human peptide data) (M2 < M1) is less than a threshold value (N), the data management module 312 artificially generates peptides not presented by the MHC allele (in the eluted human peptide data) to further balance the training data set. The threshold value N can be any one of 1, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, or 100. In a specific embodiment, the threshold value N is 50.

[0329] In various embodiments, the data management module 312 artificially generates peptides of different lengths. In various embodiments, the data management module 312 artificially generates peptides by extracting from a length distribution that matches the peptide length distribution of the source data set. In various embodiments, the data management module 312 artificially generates peptides by sampling HLA genotype frequencies that are identical or similar to the HLA genotype distribution of the source data set. The source data set may refer to the data set from which the training data set was originally obtained. After the peptide sequences are artificially generated, the generated peptides are combined with the original training data set. For example, the data management module 312 appends the generated peptides to the original training data set before shuffling the training data set to ensure that the training data is randomly sampled when used to train the presentation model.

[0330] Figure 3B An example set of training data 170 according to one embodiment is shown. Specifically, the first three data instances in the training data 170 indicate peptide presentation information from a monoallelic cell line, involving allele HLA-C*01:03 and three peptide sequences QCEIOWARE, FIEUHFWI, and FEWRHRJTRUJR. The fourth data instance in the training data 170 indicates peptide information from a polyallelic cell line, involving alleles HLA-B*07:02, HLA-C*01:03, HLA-A*01:01, and peptide sequence QIEJOEIJE. The first data instance indicates that the peptide sequence QCEIOWARE is not presented by the allele HLA-C*01:03. As discussed in the previous two paragraphs, the peptide sequence can be randomly generated by the data management module 312, or determined from the source protein of the presented peptide. The training data 170 also includes a binding affinity prediction of 1000 nM and a stability prediction of 1 h for peptide sequence-allele pairs. The training data 170 also includes allele-independent interaction variables, such as the C-terminal flanking sequence of the peptide FJELFISBOSJFIE, and 10 2 The fourth data example shows that the peptide sequence QIEJOEIJE is presented by one of the alleles HLA-B*07:02, HLA-C*01:03, or HLA-A*01:01. The training data 170 also includes the binding affinity prediction and stability prediction for each allele, as well as the C-flanking sequence of the peptide and the mRNA quantitative measurement of the peptide.

[0331] VII.C.2. Encoding module

[0332] In various embodiments, the encoding module 314 encodes the information contained in the training data 170 into a representation (e.g., a numerical representation) that can be used to generate one or more presentation models. As an example, the encoding module 314 encodes the peptide sequence of the candidate infectious disease-derived peptide into a representation for analysis by one or more presentation models. As another example, the encoding module 314 encodes the peptide sequence of the HLA allele into a representation for generating a presentation probability. In various embodiments, the encoding module 314 encodes the information contained in the training data 170 into one or more numerical vectors, which are input into one or more presentation models.

[0333] In one embodiment, the encoding module 314 performs one-hot encoding on a sequence on a predetermined 20-letter amino acid alphabet (e.g., a peptide sequence, a C-terminal flanking sequence, an N-terminal flanking sequence, or a peptide sequence of an MHC allele). i The peptide sequence p of amino acids i Represented as a 20·k i A row vector of elements, where p i 20·(j-1)+1 ,p i 20·(j-1)+2 ,…,p i 20·j The value of the single element in the alphabet corresponding to the amino acid at the jth position of the peptide sequence is 1. Otherwise, the values ​​of the remaining elements are 0. For example, for a given alphabet {A, C, D, E, F, G, H, I, K, L, M, N, P, Q, R, S, T, V, W, Y}, the peptide sequence EAF of 3 amino acids for data instance i can be represented by a row vector of 60 elements: i =[0 0 0 1 0 00 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]. C-terminal flanking sequence c i , and the protein sequences of the MHC alleles d h And other sequence data in the presentation information can be encoded similarly as described above.

[0334] When the training data 170 contains amino acid sequences of different lengths, the encoding module 314 can also expand the predetermined alphabet by adding PAD characters to encode the peptides into vectors of equal length. For example, this can be achieved by left-padding the peptide sequence with PAD characters until the length of the peptide sequence reaches the longest peptide sequence in the training data 170. Therefore, when the peptide sequence with the longest length has k maxWhen there are 20 amino acids, the encoding module 314 numerically represents each sequence as having (20+1)k max As an example, for the extended alphabet {PAD, A, C, D, E, F, G, H, I, K, L, M, N, P, Q, R, S, T, V, W, Y} and k max =5, the same example peptide sequence EAF of 3 amino acids can be represented by a row vector of 105 elements: i =[1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]. i Or other sequence data can be encoded in a similar manner as described above. Thus, the peptide sequence p i or c i Each independent variable or column in represents the presence of a specific amino acid at a specific position in the sequence.

[0335] Although the above method of encoding sequence data is described with reference to a sequence having an amino acid sequence, the method can be similarly extended to other types of sequence data, such as DNA or RNA sequence data, etc.

[0336] The encoding module 314 also adds one or more MHC alleles a of the data instance i i is encoded as a row vector of m elements, where each element h=1, 2, ..., m corresponds to a uniquely determined MHC allele. The value of the element corresponding to the MHC allele determined for data instance i is 1. Otherwise, the values ​​of the remaining elements are 0. For example, the alleles HLA-B*07:02 and HLA-C*01:03 of data instance i corresponding to a multi-allele cell line in m=4 uniquely determined MHC allele types {HLA-A*01:01, HLA-C*01:08, HLA-B*07:02, HLA-C*01:03} can be represented by a row vector of 4 elements: a i =[0 0 1 1], where a3 i =1 and a4 i= 1. Although this example is described herein with 4 identified MHC allele types, the number of MHC allele types may actually be in the hundreds or thousands. As discussed previously, each data instance i typically contains up to 6 different allele types associated with the peptide sequence p. i Related MHC allele types.

[0337] The encoding module 314 also adds the label y of each data instance i i Encoded as a binary variable with values ​​from the set {0,1}, where the value 1 indicates that the peptide x i By the relevant MHC allele a i , while a value of 0 indicates that peptide x i Not affected by any related MHC allele a i When the dependent variable y i When representing mass spectrometer ion currents, the encoding module 314 may additionally use various functions, such as scaling the values ​​using a logarithmic function with a range of [-∞,∞] for ion current values ​​between [0,∞].

[0338] The encoding module 314 can convert the peptide p i A pair of alleles interacting with the associated MHC allele h h i is represented as a row vector in which the numerical representations of the allele interaction variables are concatenated one after another. For example, the encoding module 314 can encode x h i Expressed as equal to [p i ]、[p i b h i ]、[p i s h i ] or [p i b h i s h i ], where b h i Peptide p i and binding affinity predictions for the relevant MHC allele h, and similarly for s h i Alternatively, one or more combinations of allele interaction variables can be stored separately (e.g., as separate vectors or matrices).

[0339] In one example, the encoding module 314 generates an allele interaction variable x by incorporating a measured or predicted value of binding affinity into the allele interaction variable x. h iIn the figure, the binding affinity information is shown.

[0340] In one example, the encoding module 314 generates a binding stability variable x by incorporating a measured or predicted value of binding stability into the allele interaction variable x. h i In the figure, the binding affinity information is shown.

[0341] In one example, the encoding module 314 generates an association rate by incorporating the measured or predicted value of the association rate into the allele interaction variable x h i The binding rate information is represented in .

[0342] Further examples of training data that the encoding module 314 can encode are described in WO2017106638 and WO2019168984, each of which is incorporated herein by reference in its entirety.

[0343] VIII. Training Modules

[0344] The training module 316 constructs one or more presentation models, such as one or more multi-part presentation models, which generate the probability of whether a peptide sequence (e.g., an infectious disease-derived peptide sequence) is presented by an MHC allele associated with the peptide sequence. Specifically, given a peptide sequence p k and MHC allele a k A collection and / or with the peptide sequence p k Related MHC allele sequences d k , the multi-part presentation model produces an estimate u k , indicating the peptide sequence p k By one or more related MHC alleles a k Possibility of presentation.

[0345] Reference now Figure 4A , which shows the process of implementing a multi-part presentation model according to one embodiment. Figure 4A As shown, the multi-part presentation model 430 receives as input one or more peptide sequences 410 (e.g., infectious disease derived peptide sequences), HLA allele sequences 420 from a patient (e.g., HLA allele sequences expressed by the patient), and / or determined HLA alleles 425 of the patient. Figure 4A The HLA allele sequence from patient 420 and the determined HLA allele 425 of the patient are shown, but in various embodiments, only the HLA allele sequence from patient 420 is required. For example, given an HLA allele sequence that is sufficiently unique and attributable to only one HLA allele, the determined HLA allele 425 can be derived from the HLA allele sequence.

[0346] Typically, the multi-part presentation model 430 has been previously trained using training data. In certain embodiments, the training data used to train the multi-part presentation model includes 1) binding affinity data between infectious disease-derived peptide sequences and HLA alleles and 2) eluted human peptide data from mass spectrometry (e.g., human immunopeptidomics) representing presentation of human peptide sequences and HLA alleles. This is beneficial when there is limited eluted peptide data for infectious disease-derived peptide sequences.

[0347] Typically, the multi-part presentation model 430 analyzes at least the peptide sequences 410 and HLA allele sequences from the patient 420 and determines a set of presentation probabilities 440. Here, each presentation probability may represent the likelihood that a particular peptide sequence is presented by an HLA allele expressed by the patient. Based on the presentation probabilities 440, antigens are determined to generate selected antigens 450. For example, a threshold number of peptide sequences having the highest presentation probability 440 values ​​are included as selected antigens, which can be used to develop therapeutic agents, as described in further detail herein.

[0348] VIII.A. Overview of the Multi-Part Presentation Model

[0349] The training module 316 constructs one or more multi-part rendering models based on a training data set stored in memory 170 generated from the rendering information stored in 165 .

[0350] Reference now Figure 4B, which shows the network architecture of the multi-part presentation model in more detail according to one embodiment.The multi-part presentation model includes a first part and a second part, wherein the first part includes at least one pan-allele model 460 part (also referred to as a pan-specific model part), and the second part includes multiple allele-specific models 470.In general, the pan-allele model 460 and the allele-specific model 470 analyze different information as input, but both output independent allele presentation possibilities.At a high level, the pan-allele model analyzes peptide sequences and HLA sequences to generate independent allele presentation possibilities.Therefore, the pan-allele model architecture represents alleles as HLA sequences rather than classification values, thereby allowing similar alleles to share information, but blurring the distinction between alleles.On the contrary, the allele-specific model analyzes peptide sequences without analyzing HLA sequences.In various embodiments, the allele-specific model can analyze additional information, such as allele non-interaction information, examples of which include C-side or N-side sequences. Here, the allele-specific model treats each allele independently of all other alleles, so each instance of the allele-specific model learns from all alleles in the training data set, but similar HLA alleles are learned independently of each other. By utilizing a multi-part presentation model 430 that includes both a pan-allele model 460 and an allele-specific model 470, the multi-part presentation model 430 is able to predict independent allele presentation probabilities with improved accuracy.

[0351] Referring first to the pan-allele model 460 (also referred to as the pan-specific model), in various embodiments, the pan-allele model 460 can receive as input a representation of a peptide sequence 410, which is Figure 4B 480. In various embodiments, the peptide sequence 410 is from an infectious disease derived peptide. In various embodiments, the peptide sequence 410 is from a human peptide. As described herein, the peptide representation 480 can be a representation of the peptide sequence 410 generated by the encoding module 314. Therefore, the pan-allele model 460 is trained to appropriately analyze the peptide representation 480. In some embodiments, the pan-allele model 460 directly receives the peptide sequence 410 as input. Therefore, the peptide sequence 410 does not need to be transformed into a peptide representation before being input into the pan-allele model 460.

[0352] In various embodiments, pan-allele model 460 may receive as input a representation of the HLA sequence from patient 420, which is then Figure 4B485. As described herein, HLA sequence representation 485 can be a representation of the HLA sequence from patient 420 generated by encoding module 314. Thus, pan-allele model 460 is trained to appropriately analyze HLA sequence representation 485. In some embodiments, pan-allele model 460 directly receives the HLA sequence from patient 420 as input. Thus, the HLA sequence of patient 420 does not need to be transformed into a representation before being input into pan-allele model 460.

[0353] In various embodiments, there may be multiple instances of pan-allele model 460 in multi-part presentation model 430. For example, Figure 4B As shown, there may be M instances of pan-allele model 460. Each instance of pan-allele model 460 can be trained with random shuffling of training data, so each instance sees different groups of combinations in different training data to learn different patterns. In various embodiments, M is any one of one, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen, fifteen, sixteen, seventeen, eighteen, nineteen or twenty instances of pan-allele model 460. In a specific embodiment, M represents 10 instances of pan-allele model 460.

[0354] Referring to the allele-specific model 470, in various embodiments, the allele-specific model 470 may receive as input a representation of the peptide sequence 410, which is Figure 4B 480. In various embodiments, Figure 4B As shown, the peptide representation in the input allele-specific model 470 can be the same as the same peptide representation in the input pan-allele model 460. In other embodiments, different peptide representations are generated so that the pan-allele model 460 and the allele-specific model 470 analyze different peptide representations. In some embodiments, the allele-specific model 470 directly receives the peptide sequence 410 as input. Therefore, the peptide sequence 410 does not need to be converted into a peptide representation before being input into the allele-specific model 470.

[0355] Despite Figure 4B Not shown, but in various embodiments, the allele-specific model may also receive additional allele-non-interaction information as input, examples of which include C-flanking or N-flanking sequences of the peptide sequence 410.

[0356] In various embodiments, there may be multiple instances of allele-specific models 470 in a multi-part presentation model 430. For example, Figure 4BAs shown, there may be N instances of allele-specific model 470. In various embodiments, N is any one of one, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen, fifteen, sixteen, seventeen, eighteen, nineteen, or twenty instances of allele-specific model 470. In a specific embodiment, N represents 10 instances of allele-specific model 470.

[0357] like Figure 4B As shown, the output of the allele-specific model 470 is combined with the HLA allele representation 490, which is a representation of the determined HLA alleles 425 expressed by the patient. The combination of the output of the allele-specific model 470 and the HLA allele representation 490 can be an independent allele presentation probability. In a specific embodiment, the HLA allele representation is a one-hot encoding of the determined HLA alleles 425. Therefore, the combination of the output of the allele-specific model 470 and the HLA allele representation 490 can be an independent allele presentation probability for a specific HLA allele expressed by the patient.

[0358] In various embodiments, the independent allele probabilities determined by the pan-allele model 460 and the independent allele probabilities determined by the multiple allele-specific models 470 are transformed, such as Figure 4B S-shaped transformation shown. Other scaling transformations may be implemented. In various embodiments, the transformation of independent allele probabilities is optional and need not be performed.

[0359] The independent allele probabilities determined by the pan-allele model 460 and the independent allele probabilities determined by multiple allele-specific models 470 are combined to generate the final presentation probability 440. In various embodiments, the combination is a statistical combination of the independent allele probabilities determined by the pan-allele model 460 and the independent allele probabilities determined by multiple allele-specific models 470. Example statistical combinations include mean value, median, mode, variance and standard deviation. In specific embodiments, the final presentation probability 440 represents the average value of the independent allele probabilities determined by the pan-allele model 460 and the independent allele probabilities determined by multiple allele-specific models 470. In various embodiments, the final presentation probability 440 represents the weighted average value of the independent allele probabilities determined by the pan-allele model 460 and the independent allele probabilities determined by multiple allele-specific models 470. This enables the final presentation probability 440 to give a greater weight to the contribution of a model type relative to the model type.

[0360] In various embodiments, regardless of the specific type of multi-part presentation model, all multi-part presentation models capture the dependencies between the independent and dependent variables in the training data 170 such that the loss function is minimized. Specifically, the loss function (y i∈S 、u i∈S ; θ) represents the dependent variable y of one or more data instances S in the training data 170 i∈S The value of is related to the estimated likelihood u of the data instance S generated by the presentation model i∈S In a specific implementation described throughout the rest of this specification, the loss function (y i∈S 、u i∈S ; θ) is the negative log-likelihood function given by formula (1a), as follows:

[0361]

[0362] However, in practice, another loss function may be used. For example, when predicting mass spectrometer ion currents, the loss function is the mean square loss given by Equation 1b, as follows:

[0363]

[0364] The multi-part presentation model can be a parametric model, where one or more parameters θ mathematically specify the dependency between the independent and dependent variables. i∈S ,u i∈S The various parameters of the parameterized type presentation model that minimizes θ) are determined by a gradient-based numerical optimization algorithm, such as a batch gradient algorithm, a stochastic gradient algorithm, etc. Alternatively, the multi-part presentation model can be a non-parametric model, in which the model structure is determined by the training data 170 rather than being strictly based on a fixed set of parameters.

[0365] In various embodiments, the multi-part presentation model achieves a particular performance metric, such as a precision-recall area under the curve (PR-AUC) performance metric. In various embodiments, the multi-part presentation model achieves at least 0.05, at least 0.06, at least 0.07, at least 0.08, at least 0.09, at least 0.1, 0.11, at least 0.12, at least 0.13, at least 0.14, at least 0.15, at least 0.16, at least 0.17, at least 0.18, at least 0.19, at least 0.20, at least 0.21, at least 0.22, at least 0.23, at least 0.24, at least 0.25, at least 0.26, at least 0.27, at least 0.28, at least 0.29, at least 0.30, at least 0.31, at least 0.32, at least 0.33, at least 0.34, at least 0.35, at least 0.36, at least 0.37, at least 0.38, at least 0.39, at least 0.40, at least 0.41, at least 0.42, at least 0.43, at least 0.44, at least 0.45, at least 0.46, at least 0.47, at least 0.48, at least 0.49, at least 0.50, at least 0.51 , at least 0.52, at least 0.53, at least 0.54, at least 0.55, at least 0.56, at least 0.57, at least 0.58, at least 0.59, at least 0.60, at least 0.61, at least 0.62, at least 0.63, at least 0.64, at least 0.65, at least 0.66, at least 0.67, at least 0.68, at least 0.69, at least 0.70, at least 0.71, at least 0.72, at least 0.73, at least 0.74, at least 0.75, at least 0.76 In various embodiments, the multi-part presentation model achieves a PR-AUC of at least 0.05. In various embodiments, the multi-part presentation model achieves a PR-AUC of at least 0.10. In various embodiments, the multi-part presentation model achieves a PR-AUC of at least 0.15. In various embodiments, the multi-part presentation model achieves a PR-AUC of at least 0.20. In various embodiments, the multi-part presentation model achieves a PR-AUC of at least 0.25. In various embodiments, the multi-part presentation model achieves a PR-AUC of at least 0.30. In various embodiments, the performance of the multi-part presentation model is measured based on presentation of viral epitopes by X alleles (e.g., class I or class II HLA alleles).In various embodiments, X alleles include 1 allele, 2 alleles, 3 alleles, 4 alleles, 5 alleles, 6 alleles, 7 alleles, 8 alleles, 9 alleles, 10 alleles, 11 alleles, 12 alleles, 13 alleles, 14 alleles, 15 alleles, 16 alleles, 17 alleles, 18 alleles, 19 alleles, 20 alleles, 21 alleles, 22 alleles, 23 alleles, 24 alleles, 25 alleles, 26 alleles, 27 alleles, 28 alleles, 29 alleles, or 30 alleles. In specific embodiments, X alleles include 5 alleles. In specific embodiments, X alleles include 10 alleles. In a specific embodiment, X alleles include 15 alleles. In a specific embodiment, X alleles include 20 alleles. In a specific embodiment, X alleles include 25 alleles. In a specific embodiment, X alleles include 30 alleles.

[0366] VIII.B. Independent Allele Model

[0367] See now Figure 5A , it shows the implementation of the allele-specific model part of the multi-part presentation model according to an embodiment.In various embodiments, allele-specific model 470 can be a neural network.In specific embodiments, allele-specific model 470 is a multilayer perceptron (MLP).Therefore, allele-specific model 470 can be composed of multiple layers 520.As described in further detail herein, neural network can also include nodes in layer 520 and connections between nodes with relevant parameters.In various embodiments, allele-specific model 470 includes two or more layers 520.In various embodiments, allele-specific model 470 includes three or more, four or more, five or more, six or more, seven or more, eight or more, nine or more, or ten or more layers 520.

[0368] The initial layer (hereinafter referred to as the input layer) of the allele-specific model 470 receives values ​​of peptide representations 480 derived from peptide sequences 410. In various embodiments, the peptide sequences 410 are from infectious disease derived peptides. In various embodiments, the peptide sequences 410 are from human peptides. The values ​​of the peptide representations 480 are propagated through layer 520 to generate output values. Here, the output values ​​are combined with HLA allele representations 490, which are generated based on the determined HLA alleles expressed by the patient (e.g., HLA allele 1 to HLA allele 6). Figure 5A As shown, the combination of the output of the allele-specific model 470 and the HLA allele representation 490 results in the likelihood 550 value of the independent allele. Here, the combination can be a dot product between the output of the allele-specific model 470 and the HLA allele representation 490. Therefore, the independent allele likelihood 550 value represents a personalized value for the patient based on the expressed HLA alleles. For example, each allele likelihood value 550 may include six likelihood values ​​corresponding to six expressed HLA alleles, while other likelihood values ​​corresponding to non-expressed HLA alleles are equal to zero or a value close to zero.

[0369] The training module 316 can construct the allele-specific model as an independent allele model to predict the presentation likelihood of a peptide on an allele-specific basis. In this case, the training module 316 can train the presentation model based on data instances S in the training data 170 generated by cells expressing a single MHC allele.

[0370] In one implementation, for the independent allele model, the training module 316 trains the peptide p for a particular allele h by: k The estimated probability of presentation u k To model:

[0371]

[0372] where x h k Peptide p k and the allele interaction variable encoded by the corresponding MHC allele h, f(·) is any function and is referred to as the transformation function throughout this article for convenience of description. In addition, g h (·) is any function, referred to herein as the dependency function for ease of description, and is based on the parameter set θ determined for the MHC allele h h Generate allele interaction variables x h k The dependency score of each MHC allele can be determined by minimizing the loss function with respect to θh h The parameter set θ h , where i is each instance in the subset S of training data 170 generated from cells expressing a single MHC allele h.

[0373] Dependency function g h (x h k θ h) represents the dependency score of MHC allele h, indicating whether MHC allele h will interact with at least one of the alleles based on the allele interaction feature x. h k , especially based on peptide p k The amino acid position of the peptide sequence to present the corresponding antigen. For example, if MHC allele h may present peptide p k , then the dependency score of MHC allele h may have a higher value, while if it is less likely to be presented, the dependency score may have a lower value. The transformation function f(·) transforms the input, and more specifically, in this case is g h (x h k θ h ) is transformed into appropriate values ​​to indicate the peptide p k The probability of being presented by an MHC allele.

[0374] In one specific implementation referred to throughout the rest of this specification, f(·) is a function having a range in [0,1] for a suitable domain range. In one example, f(·) is an expit function, given by the following formula:

[0375]

[0376] As another example, when the value of the domain z is equal to or greater than 0, f(·) may also be a hyperbolic tangent function, given by the following formula:

[0377] f(z)=tanh(z) (4).

[0378] Alternatively, when predicting a mass spectrometer ion current having a value outside the range [0, 1], f(·) may be any function, such as an identity function, an exponential function, a logarithmic function, and the like.

[0379] Therefore, the dependence function g of MHC allele h can be expressed as h (·) applied to the peptide sequence p k The encoded form of the peptide sequence p is used to generate the corresponding dependency score. k The independent allele probability presented by MHC allele h. The dependency score can be transformed by the transformation function f(·) to generate the peptide sequence p k Independent allele probability presented by MHC allele h.

[0380] VIII.B.1 Dependence functions of allele interaction variables

[0381] In a specific implementation described throughout the specification, the dependency function g h(·) is an affine function given by the following formula:

[0382]

[0383] The function converts x h k Each allele interaction variable in is associated with a set of parameters θ determined for the relevant MHC allele h. h is a linear combination of the corresponding parameters in .

[0384] In another specific implementation mentioned throughout the specification, the dependency function g h (·) is the network function, given by the following formula:

[0385]

[0386] A network model NN with a series of nodes arranged in one or more layers h (·) indicates that a node can be connected to other nodes through connections, and these connections are each in the parameter set θ h The value at a particular node can be expressed as the sum of the values ​​of the nodes connected to that particular node, weighted by the relevant parameters of the activation function map associated with that particular node. In contrast to affine functions, network models are advantageous because their presentation models can incorporate nonlinearities and handle data with amino acid sequences of different lengths. Specifically, through nonlinear modeling, network models can capture the interactions between amino acids at different positions in a peptide sequence and how the interactions affect peptide presentation.

[0387] Generally speaking, the network model NN h (·) can be constructed as a feedforward network, such as an artificial neural network (ANN), a convolutional neural network (CNN), a deep neural network (DNN), and / or a recurrent network, such as a long short-term memory network (LSTM), a bidirectional recurrent network, a deep bidirectional recurrent network, etc.

[0388] In one example referred to throughout the rest of this specification, each MHC allele in h=1, 2, ..., m is associated with a separate network model, and NN h (·) represents the output of the network model associated with the MHC allele h.

[0389] Figure 5B An example network model associated with an MHC allele (e.g., any MHC allele h=3) is shown. Here, the example network model may include a total of four layers, and may represent a reference Figure 5A Layer 520 of the allele-specific model 470 is described. Figure 5BAs shown in , the network model NN3(·) for MHC allele h=3 includes three input nodes at layer l=1, four nodes at layer l=2, two nodes at layer l=3, and one output node at layer l=4. The network model NN3(·) is associated with ten parameter sets θ3(1), θ3(2), ...θ3(10). The network model NN3(·) receives three allele interaction variables x3 for MHC allele h=3 k (1), x3 k (2) and x3 k (3) takes as input a single data instance of the encoded polypeptide sequence data and any other training data used and outputs a value NN3(x3 k )The network function may also include one or more network models, each network model taking a different allele interaction variable as input.

[0390] In another example, the determined MHC alleles h=1, 2, ..., m are associated with a single network model NN H (·) related, and NN h (·) represents one or more outputs of a single network model associated with MHC allele h. In such an example, the parameter set θ h can correspond to the parameter set of a single network model, so the parameter set θ h Shared by all MHC alleles.

[0391] Figure 5C An example network model NN showing sharing of MHC alleles h=1, 2, ..., m H (·).like Figure 5C As shown, the network model NN H (·) includes m output nodes, each of which corresponds to an MHC allele. The network model NN3(·) receives the allele interaction variable x3 for MHC allele h=3 k And output m values, including the value NN3(x3 corresponding to MHC allele h=3 k ).

[0392] In yet another example, the dependency function g h (·) can be expressed as:

[0393]

[0394] where g' h (x h k ;θ' h ) is a parameter set θ' haffine function, network function, etc., where the parameter set of the allele interaction variable of the MHC allele has a deviation parameter θ h 0 , represents the baseline probability of MHC allele h presentation.

[0395] In another implementation, the bias parameter θ can be shared according to the gene family of MHC allele h. h 0 That is, the bias parameter θ of the MHC allele h h 0 can be equal to θ 基因(h) 0 , where gene(h) is the gene family of MHC allele h. For example, the class I MHC alleles HLA-A*02:01, HLA-A*02:02, and HLA-A*02:03 might be assigned to the “HLA-A” gene family, and the deviation parameter θ for each of these MHC alleles is h 0 As another example, the class II MHC alleles HLA-DRB1:10:01, HLA-DRB1:11:01, and HLA-DRB3:01:01 may be assigned to the “HLA-DRB” gene family, and the deviation parameter θ for each of these MHC alleles is h 0 Can be shared.

[0396] Going back to formula (2), as an example, using the affine dependency function g h (·), among m = 4 different identified MHC alleles, peptide p k The probability of presentation by MHC allele h=3 can be generated by the following formula:

[0397]

[0398] where x3 k is the allele interaction variable determined for MHC allele h=3, and θ3 is the parameter set determined for MHC allele h=3 by minimizing the loss function.

[0399] As another example, using a separate network transformation function g h (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC allele h=3 can be generated as follows:

[0400]

[0401] where x3 kis the allele interaction variable determined for MHC allele h=3, and θ3 is the parameter set determined for the network model NN3(·) associated with MHC allele h=3.

[0402] Figure 5D The generation of peptide p associated with MHC allele h=3 using the example network model NN3(·) is shown. k The possibility of submission. Figure 5D As shown in , the network model NN3(·) receives the allele interaction variable x3 for the MHC allele h=3. k And generate output NN3(x3 k The output is mapped by the function f(·) to generate the estimated presentation probability uk

[0403] VIII.B.2. Independent Alleles and Allele-Non-Interacting Variables

[0404] In one implementation, the training module 316 combines allele non-interacting variables and trains peptide p by: k The estimated probability of presentation u k To model:

[0405]

[0406] where w k Peptide p k The non-interacting variable of the coded alleles, g w (·) is based on the parameter set θ determined for the allele non-interacting variable w The allele non-interaction variable w k Specifically, the parameter set θ for each MHC allele h h and the parameter set θ for the allele non-interacting variable w The value of θ can be obtained by h and θ w , where i is each instance in the subset S of training data 170 generated from cells expressing a single MHC allele.

[0407] Dependency function g w (w k θ w ) represents the dependence score of the allele-independent interaction variable, indicating the influence of the peptide p based on the allele-independent interaction variable. k Is it presented by one or more MHC alleles. For example, if peptide p k Peptide p kThe dependence score of the allele-independent variant is likely to be high if the peptide p k Compared with known peptide k If the C-terminal flanking sequences negatively affect the presentation of the allele, the dependency score of the non-interacting allele may have a lower value.

[0408] According to formula (7), the function g of MHC allele h can be h (·) applied to the peptide sequence p k The encoded form of the peptide sequence p is used to generate the corresponding dependency score of the allele interaction variable k The probability of an independent allele being presented by an MHC allele h. Function of the non-interacting variable g for alleles w (·) is also applied to the encoded form of the allele-independent interacting variable to generate a dependency score for the allele-independent interacting variable. The two scores are combined and the combined score is transformed by the transformation function f(·) to generate the peptide sequence p k Independent allele probability presented by MHC allele h.

[0409] Alternatively, the training module 316 can be implemented by transforming the allele non-interaction variable w k The allele interaction variable x added to formula (2) h k , to convert the allele non-interaction variable w k Included in the predicted value. Therefore, the presentation probability can be given by the following formula:

[0410]

[0411] VIII.B.3 Dependence functions of non-interacting variables of alleles

[0412] Dependence function g of the interaction variable with the allele h (·) Similarly, the dependence function g of the allele non-interacting variable w (·) can be an affine function or a network function, where a separate network model is associated with the allele non-interacting variable w k Related.

[0413] Specifically, the dependency function g w (·) is an affine function given by the following formula:

[0414] g w (w k θ w )=w k ·θ w .

[0415] w k The allele non-interacting variables and parameter set θ in w is a linear combination of the corresponding parameters in .

[0416] Dependency function g w (·) can also be a network function, given by the following formula:

[0417] g w (w k θ w )=NN w (w k θ w ).

[0418] With parameter set θ w The network model NN with relevant parameters in w (·) indicates that the network function may also include one or more network models, each of which uses a different allele non-interacting variable as input.

[0419] In another example, the dependency function g of the allele non-interacting variable w (·) can be given by the following formula:

[0420]

[0421] where g' w (w k ;θ' w ) is an affine function with allele non-interacting parameter set θ' w The network function, m k Peptide p k The mRNA quantitative measurement value, h(·) is the function of transforming the quantitative measurement value, θ w m It is a parameter in the allele non-interaction variable parameter set that is combined with the mRNA quantitative measurement value to generate the dependence score of the mRNA quantitative measurement value. In a specific embodiment mentioned throughout the remainder of this specification, h(·) is a logarithmic function, but in practice, h(·) can be any of a variety of different functions.

[0422] In yet another example, the dependency function g of the allele non-interacting variable w (·) can be given by the following formula:

[0423]

[0424] where g' w (w k;θ' w ) is an affine function with allele non-interacting parameter set θ' w Network functions, etc. k is the peptide p in the human proteome k The indicator vectors of proteins and isoforms, and θ w o is the parameter set of the allele non-interacting variables combined with the indicator vector. In one variation, when o k and parameter set θ w o When the dimension of is significantly higher, we can add parameter regularization terms to the loss function when determining the parameter values, such as where ||·|| represents L1 norm, L2 norm, combination, etc. The optimal value of the hyperparameter λ can be determined by appropriate methods.

[0425] In yet another example, the dependency function g of the allele non-interacting variable w (·) can be given by the following formula:

[0426]

[0427] where g' w (w k ;θ' w ) is an affine function with allele non-interacting parameter set θ' w The network functions, etc. (gene(p k =1)) is the indicator function, if the peptide p k from the source gene l, then the indicator function is equal to 1, as described above for the allele non-interacting variables, and θ w l is a parameter indicating the "antigenicity" of the source gene l. In one variation, when L is significantly higher and thus the parameter θ w l＝1、2、…、L When the number of is significantly high, a parameter regularization term can be added to the loss function when determining the parameter value, such as where ||·|| represents L1 norm, L2 norm, combination, etc. The optimal value of the hyperparameter λ can be determined by appropriate methods.

[0428] In yet another example, the dependency function g of the allele non-interacting variable w (·) can be given by the following formula:

[0429]

[0430] where g' w (wk ;θ' w ) is an affine function with allele non-interacting parameter set θ' w The network functions, etc. (gene(p k )=l,organization(p k )=m) is the indicator function, if the peptide p k From source gene l and if peptide p k from tissue type m then the indicator function equals 1, as described above for allele non-interacting variables, and θ w l is a parameter indicating the antigenicity of the combination of source gene 1 and tissue type m. Specifically, the antigenicity of gene 1 of tissue type m may represent the residual propensity of cells of tissue type m to present peptides from gene 1 after controlling for RNA expression and peptide sequence context.

[0431] In one variation, when L or M is significantly higher and thus the parameter θ w lm＝1、2、…、LM When the number of is significantly high, a parameter regularization term can be added to the loss function when determining the parameter value, such as Where ||·|| represents L1 norm, L2 norm, combination, etc. The optimal value of the hyperparameter λ can be determined by appropriate methods. In another variation, the parameter regularization term can be added to the loss function when determining the parameter value so that the parameters of the same source gene do not differ significantly between tissue types. For example, a penalty factor such as the following:

[0432]

[0433] in is the average antigenicity of the source gene l in different tissue types, which may penalize the standard deviation of antigenicity between different tissue types in the loss function.

[0434] In yet another example, the dependency function g of the allele non-interacting variable w (·) can be given by the following formula:

[0435]

[0436] where g' w (w k ;θ' w ) is an affine function with allele non-interacting parameter set θ' w The network functions, etc. (gene(p k =1)) is the indicator function, if the peptide p kfrom the source gene l, then the indicator function is equal to 1, as described above for the allele non-interacting variables, and θ w l is a parameter indicating the "antigenicity" of the source gene l, and (loc(p k =m)) is the indicator function, if the peptide p k from a proteome position m, the indicator function is equal to 1, and is a parameter indicating the extent to which a proteomic position m is a presentation "hotspot". In one embodiment, a proteomic position may include n adjacent peptide blocks from the same protein, where n is a hyperparameter of the model determined by an appropriate method such as grid search cross validation.

[0437] In practice, additional terms from any of equations (9), (10), (11), (12a), and (12b) can be combined to generate the dependency function g of the allele non-interacting variable w (·). For example, the term h(·) in formula (9) indicating the quantitative measurement of mRNA and the term in formulas (11), (12a) and (12b) indicating the antigenicity of the source gene can be summed with any other affine or network function to generate a dependency function of the allele non-interacting variable.

[0438] Going back to formula (7), as an example, using the affine transformation function g h (·),g w (·), among m = 4 different identified MHC alleles, peptide p k The probability of presentation by MHC allele h=3 can be generated by the following formula:

[0439]

[0440] where w k It is for peptide p k Determined allele non-interaction variable, θ w is the set of parameters determined for allele non-interacting variables.

[0441] As another example, using the network transformation function g h (·), g w (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC allele h=3 can be generated as follows:

[0442]

[0443] where w kIt is for peptide p k Determined allele interaction variable, θ w is the set of parameters determined for allele non-interacting variables.

[0444] Figure 5E The example network models NN3(·) and NN are shown. w (·) Generate peptide p associated with MHC allele h=3 k The possibility of submission. Figure 5E As shown in , the network model NN3(·) receives the allele interaction variable x3 for the MHC allele h=3. k And generate output NN3(x3 k ). Network model NN w (·) Receiving peptide p k The allele non-interaction variable w k And generate output NN w (w k ). The outputs are combined and mapped by the function f(·) to generate the estimated presentation probability u k .

[0445] VIII.C. Multi-allele model

[0446] The training module 316 can also construct a presentation model to predict the presentation likelihood of a peptide in a polymorphic environment in which two or more MHC alleles are present. In this case, the training module 316 can train the presentation model based on data instances S in the training data 170 generated by cells expressing a single MHC allele, cells expressing multiple MHC alleles, or a combination thereof.

[0447] VIII.C.1. Example 1: Maximum value for the independent allele model

[0448] In one implementation, the training module 316 trains peptides p associated with a set of multiple MHC alleles H. k The estimated probability of presentation u k Modeled as the presentation probability u determined for each MHC allele h in the set H based on cells expressing a single allele k h∈H is a function of , as described above in conjunction with formulas (2)-(10). Specifically, the presentation probability u k It can be u k h∈H In one implementation, as shown in formulas (11), (12a) and (12b), the function is a maximum function and presents the possibility u k It can be determined as the maximum value of the presentation probability of each MHC allele h in the set H.

[0449]

[0450] VIII.C.2. Example 2.1: Summation Function Model

[0451] In one implementation, the training module 316 trains the peptide p by: k The estimated probability of presentation u k To model:

[0452]

[0453] The element a h k For the peptide sequence p k The associated multiple MHC alleles H are 1, and x h k Peptide p k and the allele interaction variables encoded by the corresponding MHC allele. This can be done by h The loss function is minimized to determine the parameter set θ for each MHC allele h h , where i is each instance in the subset S of the training data 170 generated by cells expressing a single MHC allele and / or cells expressing multiple MHC alleles. The dependency function g h It can be the dependency function g introduced above. h any form of.

[0454] According to formula (13), the dependency function g h (·) Peptide sequence p applied to each of the MHC alleles H k The encoded form of the peptide sequence p is used to generate the corresponding score of the allele interaction variable k The probability of presentation by one or more MHC alleles h. The scores for each MHC allele h are combined and transformed by the transformation function f(·) to generate the peptide sequence p k The probability of being presented by a set of MHC alleles H.

[0455] The presentation model of formula (13) differs from the independent allele model of formula (2) in that each peptide p k The number of related alleles can be greater than 1. In other words, for a peptide sequence p k Multiple MHC alleles associated with H, a h k There can be more than one element with value 1.

[0456] As an example, using the affine transformation function gh (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC alleles h=2, h=3 can be generated as follows:

[0457]

[0458] where x2 k , x3 k is the allele interaction variable determined for MHC alleles h=2 and h=3, and θ2 and θ3 are parameter sets determined for MHC alleles h=2 and h=3.

[0459] As another example, using the network transformation function g h (·), g w (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC alleles h=2, h=3 can be generated as follows:

[0460]

[0461] Among them, NN2(·) and NN3(·) are network models determined for MHC alleles h=2 and h=3, and θ2θ3 are parameter sets determined for MHC alleles h=2 and h=3.

[0462] Fig. 5F The peptides p associated with MHC alleles h=2 and h=3 are generated using the example network models NN2(·) and NN3(·). k The possibility of submission. Fig. 5F As shown, the network model NN2(·) receives the allele interaction variable x2 for MHC allele h=2 k And generate output NN2(x2 k ), while the network model NN3(·) receives the allele interaction variable x3 for MHC allele h=3 k And generate output NN3(x3 k ). The outputs are combined and mapped by the function f(·) to generate the estimated presentation probability u k .

[0463] VIII.C.3. Example 2.2: Sum Function Model with Allele Non-Interacting Variables

[0464] In one implementation, the training module 316 combines allele non-interacting variables and trains peptide p by: k The estimated probability of presentation u k To model:

[0465]

[0466] where w k Peptide p k Specifically, the parameter set θ for each MHC allele h is h and the parameter set θ for the allele non-interacting variable w The value of θ can be obtained by h and θ w The dependency function g is determined by minimizing the loss function of , where i is each instance in the subset S of the training data 170 generated by cells expressing a single MHC allele and / or cells expressing multiple MHC alleles. w It can be the dependency function g introduced above. w any form of.

[0467] Therefore, according to formula (14), the function g can be h (·) Peptide sequence p applied to each of the MHC alleles H k The encoded form of the peptide sequence p is generated to generate the corresponding dependency score of the allelic interaction variable for each MHC allele h k The probability of presentation by one or more MHC alleles H. Function of the allele non-interacting variable g w (·) is also applied to the encoded form of the allele-independent interacting variables to generate dependency scores for the allele-independent interacting variables. The scores are combined and transformed by the transformation function f(·) to produce the peptide sequence p k Presentation probability presented by MHC allele H.

[0468] In the presentation model of formula (14), each peptide p k The number of related alleles can be greater than 1. In other words, for a peptide sequence p k Multiple MHC alleles associated with H, a h k There can be more than one element with value 1.

[0469] As an example, using the affine transformation function g h (·), g w (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC alleles h=2, h=3 can be generated as follows:

[0470]

[0471] where w kIt is for peptide p k Determined allele non-interaction variable, θ w is the set of parameters determined for allele non-interacting variables.

[0472] As another example, using the network transformation function g h (·), g w (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC alleles h=2, h=3 can be generated as follows:

[0473]

[0474] where w k It is for peptide p k Determined allele interaction variable, θ w is the set of parameters determined for allele non-interacting variables.

[0475] Figure 5G The example network models NN2(·), NN3(·) and NN w (·) Generate peptide p associated with MHC alleles h=2 and h=3 k The possibility of submission. Figure 5G As shown in , the network model NN2(·) receives the allele interaction variable x2 for the MHC allele h=2. k And generate output NN2(x2 k ). The network model NN3(·) receives the allele interaction variable x3 for the MHC allele h=3. k And generate output NN3(x3 k ). Network model NN w (·) Receiving peptide p k The allele non-interaction variable w k And generate output NN w (w k ). The outputs are combined and mapped by the function f(·) to generate the estimated presentation probability u k .

[0476] Alternatively, the training module 316 can be implemented by transforming the allele non-interaction variable w k The allele interaction variable x added to equation (15) h k , to convert the allele non-interaction variable w k Included in the predicted value. Therefore, the presentation probability can be given by the following formula:

[0477]

[0478] VIII.C.4. Example 3.1: Model using implicit independent allele likelihood

[0479] In another implementation, the training module 316 trains the peptide p k The estimated probability of presentation u k To model:

[0480]

[0481] The element a h k For the peptide sequence p k The associated multiple MHC alleles h∈H are 1, u' k h is the implicit independent allele presentation probability of MHC allele h, and the vector v is a vector in which element v h Corresponding to a h k ·u' k h , s(·) is a function that maps elements of v, and r(·) is a clipping function that clips the values ​​of the input to within a given range. As described in more detail below, s(·) can be a summation function or a second-order function, but it should be understood that in other embodiments, s(·) can be any function, such as a maximum function. The value of the parameter set θ for the implicit independent allele likelihood can be determined by minimizing a loss function with respect to θ, where i is each instance in a subset S of the training data 170 generated by cells expressing a single MHC allele and / or cells expressing multiple MHC alleles.

[0482] The presentation probability in the presentation model of formula (16) is modeled as the implicit independent allele presentation probability u' k h function, each possibility corresponds to a peptide p k The implicit independent allele probability differs from the independent allele presentation probability in that the parameters of the implicit independent allele probability are known from the polymorphic environment, where the direct association between the presented peptide and the corresponding MHC allele is unknown except in the monomorphic environment. Therefore, in the polymorphic environment, the presentation model can estimate not only the peptide p k Will it be presented by the set of MHC alleles H as a whole, and can also provide individual possibilities u' k h ∈ H , indicating which MHC allele h is most likely to present peptide p kThis has the advantage that the presentation model does not require training data on cells expressing a single MHC allele to generate implicit likelihoods.

[0483] In one specific implementation referred to throughout the rest of this specification, r(·) is a function with a range of [0,1]. For example, r(·) may be a clipping function:

[0484] r(z)=min(max(z,0),1)

[0485] The minimum value between z and 1 is selected as the presentation probability u k In another implementation, when the value of the domain z is equal to or greater than 0, r(·) is a hyperbolic tangent function given by the following formula:

[0486] r(z)=tanh(z).

[0487] VIII.C.5. Example 3.2: Sum-of-Functions Model

[0488] In one particular implementation, s(·) is a summation function, and the presentation probability is given by summing the implicit independent allele presentation probabilities:

[0489]

[0490] In one implementation, the implicit independent allele presentation probability of MHC allele h is generated by the following formula:

[0491]

[0492] The probability of presentation is estimated by:

[0493]

[0494] According to formula (19), we can h (·) Peptide sequence p applied to each of the MHC alleles H k The encoded form of the peptide sequence p is used to generate the corresponding dependency score of the allele interaction variable k The probability of presentation by one or more MHC alleles H. Each dependency score is first transformed by a function f(·) to generate an implicit independent allele presentation probability u' k h . The independent allele probability u' k h combination, and a clipping function can be applied to the combined likelihood to clip the values ​​to the range [0,1] to generate the peptide sequence p kThe probability of presentation presented by the set of MHC alleles H. Dependency function g h It can be any dependency function g introduced above. h any form of .

[0495] As an example, using the affine transformation function g h (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC alleles h=2, h=3 can be generated as follows:

[0496]

[0497] where x2 k , x3 k is the allele interaction variable determined for MHC alleles h=2 and h=3, and θ2 and θ3 are parameter sets determined for MHC alleles h=2 and h=3.

[0498] As another example, using the network transformation function g h (·), g w (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC alleles h=2, h=3 can be generated as follows:

[0499]

[0500] Among them, NN2(·) and NN3(·) are network models determined for MHC alleles h=2 and h=3, and θ2θ3 are parameter sets determined for MHC alleles h=2 and h=3.

[0501] Figure 5H The peptides p associated with MHC alleles h=2 and h=3 are generated using the example network models NN2(·) and NN3(·). k The possibility of submission. Figure 5H As shown, the network model NN2(·) receives the allele interaction variable x2 for MHC allele h=2 k And generate output NN2(x2 k ), while the network model NN3(·) receives the allele interaction variable x3 for MHC allele h=3 k And generate output NN3(x3 k ). Each output is mapped by a function f(·) and combined to generate an estimated presentation likelihood u k .

[0502] In another implementation, when predicting the logarithm of the mass spectrometer ion current, r(·) is a logarithmic function and f(·) is an exponential function.

[0503] VIII.C.6. Example 3.3: Functional sum model with allele non-interacting variables

[0504] In one implementation, the implicit independent allele presentation probability of MHC allele h is generated by the following formula:

[0505]

[0506] The presentation possibility is generated by:

[0507]

[0508] To incorporate the effects of allele-noninteracting variables on peptide presentation.

[0509] According to formula (21), we can h (·) Peptide sequence p applied to each of the MHC alleles H k The encoded form of the peptide sequence p is generated to generate the corresponding dependency score of the allelic interaction variable for each MHC allele h k The probability of presentation by one or more MHC alleles H. Function of the allele non-interacting variable g w (·) is also applied to the encoded form of the allele non-interacting variables to generate dependency scores for the allele non-interacting variables. The scores for the allele non-interacting variables are combined with each of the dependency scores for the allele interacting variables. Each of the combined scores is transformed by a function f(·) to generate an implicit independent allele presentation likelihood. The implicit likelihoods are combined, and a clipping function can be applied to the output of the combination to clip the values ​​to the range [0,1] to generate the peptide sequence p k The probability of presentation by MHC allele H. Dependency function g w It can be the dependency function g introduced above. w any form of.

[0510] As an example, using the affine transformation function g h (·), g w (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC alleles h=2, h=3 can be generated as follows:

[0511]

[0512] where w kIt is for peptide p k Determined allele non-interaction variable, θ w is the set of parameters determined for allele non-interacting variables.

[0513] As another example, using the network transformation function g h (·), g w (·), among m = 4 different identified MHC alleles, peptide p k The probability of being presented by MHC alleles h=2h=3 can be generated as follows:

[0514]

[0515] where w k It is for peptide p k Determined allele interaction variable, θ w is the set of parameters determined for allele non-interacting variables.

[0516] Fig.5I The example network models NN2(·), NN3(·) and NN w (·) Generate peptide p associated with MHC alleles h=2 and h=3 k The possibility of submission. Fig.5I As shown in , the network model NN2(·) receives the allele interaction variable x2 for the MHC allele h=2. k And generate output NN2(x2 k ). Network model NN w (·) Receiving peptide p k The allele non-interaction variable w k And generate output NN w (w k The outputs are combined and mapped by the function f(·). The network model NN3(·) receives the allele interaction variable x3 for the MHC allele h=3. k And generate output NN3(x3 k ), and again compare the output with the same network model NN w The output NN of (·) w (w k ) and mapped by the function f(·). The two outputs are combined to generate the estimated presentation probability u k .

[0517] In another implementation, the implicit independent allele presentation probability of MHC allele h is generated by the following formula:

[0518]

[0519] The presentation possibility is generated by:

[0520]

[0521] VIII.C.7. Example 4: Second-Order Model

[0522] In one implementation, s(·) is a second-order function, and the peptide p k The estimated probability of presentation u k Given by the following formula:

[0523]

[0524] The element u' k h is the implicit independent allele presentation probability for MHC allele h. The values ​​of the parameter set θ for the implicit independent allele probability can be determined by minimizing a loss function with respect to θ, where i is each instance in a subset S of the training data 170 generated by cells expressing a single MHC allele and / or cells expressing multiple MHC alleles. The implicit independent allele presentation probability can be in any of the forms shown in equations (18), (20), and (22) described above.

[0525] On the one hand, the model of formula (23) can imply the existence of peptide p k The probability that presentation by two MHC alleles will be simultaneous, whereas presentation by two HLA alleles is statistically independent.

[0526] According to formula (23), the peptide sequence p k The presentation probability of a peptide presented by one or more MHC alleles H can be generated by combining the implicit independent allele presentation probabilities and subtracting from the sum the probability of each pair of MHC alleles presenting the peptide simultaneously p k The possibility to generate a peptide sequence p k Presentation probability presented by MHC allele H.

[0527] As an example, using the affine transformation function g h (·), among m = 4 different identified HLA alleles, peptide p k The probability of being presented by HLA alleles h=2h=3 can be generated as follows:

[0528]

[0529] where x2 k , x3 kis the allele interaction variable determined for HLA alleles h=2, h=3, and θ2, θ3 are parameter sets determined for HLA alleles h=2, h=3.

[0530] As another example, using the network transformation function g h (·), g w (·), among m = 4 different identified HLA alleles, peptide p k The probability of being presented by HLA alleles h=2, h=3 can be generated as follows:

[0531]

[0532] Among them, NN2(·) and NN3(·) are network models determined for HLA alleles h=2 and h=3, and θ2θ3 are parameter sets determined for HLA alleles h=2 and h=3.

[0533] VIII.D. Pan-allele model

[0534] In contrast to the independent allele model, the pan-allelic model (e.g. Figure 4B The pan-allele model 460 described in is a model that is capable of predicting the probability of presentation of a peptide on a pan-allele basis. Specifically, unlike an independent allele model that is capable of predicting the probability of a peptide being presented by one or more known MHC alleles that have been previously used to train an independent allele model, a pan-allele model is a presentation model that is capable of predicting the probability of a peptide being presented by any MHC allele, including unknown MHC alleles that the model has not previously encountered during training.

[0535] In short, the pan-allele model is trained by the training module 316. Similar to the training of the independent allele model, the training module 316 can train the pan-allele presentation model based on data instances S in the training data 170 generated by cells expressing a single MHC allele, cells expressing multiple MHC alleles, or a combination thereof. However, the training module 316 does not use a specific MHC allele or MHC allele a. k h Instead of training the pan-allele presentation model using a specific set of MHC allele peptide sequences available in the training data set 170 h Specifically, the training module 316 trains the pan-allele presentation model based on the amino acid positions of the MHC alleles available in the training data 170 .

[0536] After the pan-allele model has been trained, when a peptide sequence and a known or unknown MHC allele peptide sequence are input into the model to determine the probability of presentation of the peptide by a known or unknown MHC allele, the model is able to accurately predict this probability by using the information learned during training with similar MHC allele peptide sequences. For example, a pan-allele model trained using training data 170 that does not contain any occurrences of the A*02:07 allele can still accurately predict peptide presentation of the A*02:07 allele by utilizing information learned during training with similar alleles (e.g., alleles in the A*02 gene family). In this way, a single presentation pan-allele model can predict the likelihood of presentation of a peptide on any MHC allele.

[0537] VIII.D.2. Advantages of the pan-allele model

[0538] The main advantage of the pan-allele presentation model is that the pan-allele presentation model has greater versatility than the independent allele presentation model. As mentioned above, the independent allele model can predict the probability of one or more determined MHC alleles presenting peptides for training the independent allele model. In other words, the independent allele model is associated with a limited set of one or more known MHC alleles.

[0539] Thus, given a sample containing a particular set of one or more MHC alleles, to determine the probability of a peptide being presented by the particular set of MHC alleles, an independent allele model trained using the particular set of MHC alleles is selected. In other words, when relying on an independent allele model to predict the probability of a peptide being presented by an MHC allele, predictions are made only for MHC alleles that have appeared in the training data 170. Because there are many MHC alleles (especially minor variations within the same gene family), a very large number of training samples are required to train an independent allele presentation model so that peptide presentation predictions can be made for all MHC alleles.

[0540] In contrast, the pan-allele model is not limited to making predictions for the specific set of one or more MHC alleles for which it was trained. Instead, during use, the pan-allele model is able to accurately predict the probability of a given peptide being presented by a previously seen and / or previously unseen MHC allele using information learned during training with similar MHC allele peptide sequences. Therefore, the pan-allele model is independent of a specific set of one or more MHC alleles and is able to predict the probability of a peptide being presented by any MHC allele. This versatility of the pan-allele model means that a single model can be used to predict the likelihood of any peptide being presented by any MHC allele. Therefore, the use of the pan-allele model reduces the amount of training data required to maximize individual HLA coverage and population HLA coverage.

[0541] VIII.D.3. Use of the pan-allele model

[0542] In short, when using the pan-allele model to predict the possibility of a peptide being presented by a single MHC allele, a set of inputs is provided to the pan-allele model, as described in detail below, and the pan-allele model generates a single output. On the other hand, when using the pan-allele model to predict the possibility of a peptide being presented by multiple MHC alleles, the pan-allele model is iteratively used for each MHC allele in multiple MHC alleles. Specifically, when using the pan-allele model to predict the possibility of a peptide being presented by multiple MHC alleles, the first set of inputs associated with the first MHC allele in multiple MHC alleles is provided to the pan-allele model, and the pan-allele model generates the first output of the first MHC allele. Then, the second set of inputs associated with the second MHC allele in multiple MHC alleles is provided to the pan-allele model, and the pan-allele model generates the second output of the second MHC allele. This process is iteratively performed for each MHC allele in multiple MHC alleles. Finally, the outputs generated by the pan-allelic model for each of the multiple MHC alleles are combined to generate a single probability that a given peptide is presented by the multiple MHC alleles.

[0543] VIII.D.4. Overview of the Pan-Allele Model

[0544] Reference now Fig. 6A , which illustrates an implementation of the pan-allele model portion 460 of a multi-part presentation model according to one embodiment. As discussed herein, the multi-part presentation model can include multiple instances of the pan-allele model portion 460 (e.g., 10 instances of the pan-allele model portion 460). In various embodiments, the pan-allele model 460 is a neural network. In various embodiments, as Fig. 6AAs shown, the pan-allele model 460 includes two sets of layers (e.g., the first set 630 of layers and the second set 640 of layers). Typically, each of the first set 630 of layers and the second set 640 of layers can be composed of multiple layers, nodes in the layer and the connection between the nodes with relevant parameters. In various embodiments, each of the first set 630 of layers and the second set 640 of layers includes two or more layers. In various embodiments, each of the first set 630 of layers and the second set 640 of layers includes three or more, four or more, five or more, six or more, seven or more, eight or more, nine or more, or ten or more layers. In a specific embodiment, each of the first set 630 of layers and the second set 640 of layers is a multilayer perceptron (MLP). Typically, the first set 630 of layers and the second set 640 of layers perform different functions, as described in further detail herein.

[0545] The first set 630 of layers receives as input an HLA sequence representation 485, which may be a representation of a sequence of an individual HLA allele expressed by a patient. In various embodiments, the sequence of a single HLA allele is encoded (e.g., a one-hot encoding scheme) to generate the HLA sequence representation 485. The first set 630 of layers reduces the dimension of the HLA sequence representation 485 to a new lower dimensional representation. For example, the new low dimensional representation may retain meaningful properties of the HLA sequence representation 485. In various embodiments, in order to reduce the HLA sequence representation 485 in dimension, the first set 630 of layers includes an input layer having a greater number of nodes than the terminal layer.

[0546] like Fig. 6A As shown, the second set of layers 640 receives as input a peptide representation 480 derived from a peptide sequence 410. In various embodiments, the peptide sequence 410 is from an infectious disease derived peptide. In various embodiments, the peptide sequence 410 is from a human peptide. In addition, the second set of layers 640 can receive as input the output of the first set of layers 630. As described above, the output of the first set of layers 630 is a dimensionality-reduced representation of the HLA allele sequence. The second set of layers 640 models the interaction between infectious disease derived peptides and HLA allele sequences to determine whether the HLA sequence is likely to present the peptide sequence. The output of the second set of layers 640 is the likelihood 620 of an independent allele (e.g., the likelihood of each of the six HLA alleles expressed by the patient presenting the peptide sequence 410). As described above with reference to Figure 4B As discussed, the independent allele probabilities of the pan-allele model 460 can be transformed and combined to generate the final presented probability 440.

[0547] In one implementation, the pan-allele model is used to estimate the peptide p for allele h k The possibility of submission u k In some embodiments, the pan-allelic model is represented by the following formula:

[0548]

[0549] where p k represents the peptide sequence, d h represents the peptide sequence of MHC allele h, f(·) is any transformation function, g H (·) is any dependency function. The pan-allele model is based on a shared parameter set θ determined for all MHC alleles H , for the peptide sequence p k and MHC allele peptide sequences d h Generate dependency scores. Shared parameter set θ H The value of is learned during training of the pan-allele model.

[0550] Dependency function g H ([p k d h ];θ H ) represents the dependency score of MHC allele h, indicating whether MHC allele h will be expressed based on at least one peptide sequence p k Amino acid positions and MHC allele peptide sequences h The amino acid position of the peptide p k For example, considering the input MHC allele peptide sequence d h , if MHC allele h is likely to present peptide p k , then the dependency score of MHC allele h may have a higher value, while if it is less likely to be presented, the dependency score may have a lower value. The transformation function f(·) transforms the input, and more specifically, in this case is g H ([p k d h ];θ H ) is transformed into appropriate values ​​to indicate the peptide p k The probability of being presented by MHC allele h.

[0551] In a specific implementation referred to throughout the rest of this specification, f(·) is a function having a range in [0,1] for a suitable domain range. In one example, f(·) is an expit function. As another example, f(·) may also be a hyperbolic tangent function when the value of the domain z is equal to or greater than 0. Alternatively, when predicting mass spectrometer ion currents having values ​​outside the range [0,1], f(·) may be any function, such as an identity function, an exponential function, a logarithmic function, etc.

[0552] Therefore, we can use the dependency function g H (·) applied to the peptide sequence p k The coding form and application to MCH allele peptide sequence d h The encoded form of the peptide sequence p is used to generate the corresponding dependency score. k The probability of being presented by MHC allele h. The dependency score can be transformed by the transformation function f(·) to generate the peptide sequence p k The probability of being presented by MHC allele h.

[0553] VIII.D.5. Dependence Functions of Allelic Interaction Variables

[0554] In a specific implementation described throughout the specification, the dependency function g H (·) is an affine function given by the following formula:

[0555]

[0556] where α is the intercept, Peptide p k The residue at position i, d hj represents the residue at position j of MHC allele h, 1[] represents an indicator variable whose value is 1 if the condition in the brackets is true and 0 otherwise. k The amino acid at position i is amino acid k, then True, otherwise false, if the amino acid at position j of MHC allele h is amino acid l, then d hj = l is true, otherwise it is false, n pep represents the length of the modeled peptide, n MHC represents the number of MHC residues considered in the model, θ H，ijkl is the coefficient describing the contribution of residue k at peptide position i and residue l at MHC allele position j to the likelihood of presentation. This is a linear model in the one-hot encoded peptide sequence and one-hot encoded MHC allele sequence, where interactions between peptide residues and MHC residues exist for all peptide residues and MHC allele residues.

[0557] In another specific implementation mentioned throughout the specification, the dependency function g H (·) is the network function, given by the following formula:

[0558] g H ([p k d h ];θ H )=N N H([p k d h ];θ H ) (26)

[0559] A network model NN with a series of nodes arranged in one or more layers H (·) indicates that a node can be connected to other nodes through connections, and these connections are each in the parameter set θ H The value at a particular node can be expressed as the sum of the values ​​of the nodes connected to that particular node, weighted by the relevant parameters of the activation function map associated with that particular node. In contrast to affine functions, network models are advantageous because their presentation models can incorporate nonlinearities and handle data with amino acid sequences of different lengths. Specifically, through nonlinear modeling, network models can capture the interactions between amino acids at different positions in a peptide sequence, as well as the interactions between amino acids at different positions in an MHC allele peptide sequence, and how these interactions affect peptide presentation.

[0560] Generally speaking, the network model NN H (·) can be constructed as a feedforward network, such as an artificial neural network (ANN), a convolutional neural network (CNN), a deep neural network (DNN), and / or a recurrent network, such as a long short-term memory network (LSTM), a bidirectional recurrent network, a deep bidirectional recurrent network, etc.

[0561] In one example, the single network model NN H (·) can be the peptide sequence p encoded by a given MHC allele h k and protein coding sequence d h A network model that outputs dependency scores. In such an instance, the parameter set θ H can correspond to the parameter set of a single network model, so the parameter set θ H can be shared by all MHC alleles. Therefore, in such an example, NN H (·) can represent any input [p k d h ] single network model NN HAs mentioned above, this network model is advantageous because it can predict the peptide presentation probability of unknown MHC alleles in the training data simply by determining the protein sequence of the MHC allele.

[0562] Figure 6B Example network model NN showing MHC allele sharing H (·).like Figure 6B As shown, the network model NN H (·) Peptide sequence p that receives MHC allele h k and protein sequence d h As input, and outputs the dependency score NN corresponding to the MHC allele h H ([p k d h ]).

[0563] Figure 6C An example network model NN is shown H (·). Here, the example network model can be represented as Fig. 6A The second set 640 of layers of the pan-allele model 460 is shown. Figure 6C As shown, the network model NN H (·) includes four input nodes at layer l=1, five nodes at layer l=2, two nodes at layer l=3, and one output node at layer l=4. In an alternative embodiment, the network model NN H (·) can contain any number of layers, and each layer can contain any number of nodes. Network model NN H (·) with thirteen non-zero parameter sets θ H (1) θ H (2),…,θ H (13) These parameters are used to transform the values ​​propagated from node to node through the network model.

[0564] like Figure 6C As shown, the network model NN H The input values ​​received by the four input nodes at layer l=1 of (·) include encoded polypeptide sequence data and encoded MHC allele peptide sequence data. The encoded polypeptide sequence data contains the amino acid sequence of the peptide, and the encoded MHC allele peptide sequence data contains the amino acid sequence of the MHC allele that may (or may not) present the peptide. In some embodiments, once the network model NN is input through the input node at layer l=1, H (·), the encoded polypeptide sequence will be connected to the network model NN H The MHC allele peptide sequence is encoded in the layer (·). These input values ​​are then passed through the network model NN according to the values ​​of the parameters.H (·) propagation. In some embodiments, the network model NN H The layer of (·) comprises two fully connected dense network layers. In other embodiments, the first layer of the two fully connected dense network layers comprises 64-128 nodes with a rectified linear unit activation function. In even other embodiments, the second layer of the two fully connected dense network layers comprises a single node with a linear output. In such embodiments, the single node can be a network model NN H (·) output node. Finally, the network model NN H (·) The value NN H ([p k d h ]) output. This output represents the dependency score of MHC allele h, indicating whether MHC allele h will present peptide sequence p k The network function may also include one or more network models, each network model taking a different allele interaction variable (eg, peptide sequence) as input.

[0565] In yet another example, the dependency function g H (·) can be expressed as:

[0566]

[0567] where g' H ([p k d h ];θ' H ) is a parameter set θ' H Affine functions, network functions, etc. of allele interaction variables, where the shared parameter set θ H There is a bias parameter θ h 0 , represents the baseline probability of presentation of any MHC allele.

[0568] In another implementation, the bias parameter θ can be shared according to the gene family of MHC allele h. H 0 That is, the bias parameter θ of the MHC allele h H 0 can be equal to θ 基因(h) 0 , where gene(h) is the gene family of MHC allele h. For example, the class I MHC alleles HLA-A*02:01, HLA-A*02:02, and HLA-A*02:03 might be assigned to the “HLA-A” gene family, and the deviation parameter θ for each of these MHC alleles is H 0As another example, the class II MHC alleles HLA-DRB1:10:01, HLA-DRB1:11:01, and HLA-DRB3:01:01 may be assigned to the “HLA-DRB” gene family, and the deviation parameter θ for each of these MHC alleles is H 0 As mentioned above, the gene family may be one of the allele interaction variables associated with the MHC allele h.

[0569] Returning to formula (23), as an example, using the affine dependence function g H (·), peptide p k The probability of presentation by MHC allele h can be generated by the following formula:

[0570]

[0571] where α is the intercept, Peptide p k The residue at position i, d hj represents the residue at position j of MHC allele h, 1[] represents an indicator variable whose value is 1 if the condition in the brackets is true and 0 otherwise. k The amino acid at position i is amino acid k, then True, otherwise false, if the amino acid at position j of MHC allele h is amino acid l, then d hj = l is true, otherwise it is false, n pep represents the length of the modeled peptide, n MHC represents the number of MHC residues considered in the model, θ H，ijkl is the coefficient describing the contribution of residue k at peptide position i and residue l at MHC allele position j to the likelihood of presentation. This is a linear model in the one-hot encoded peptide sequence and one-hot encoded MHC allele sequence, where interactions between peptide residues and MHC residues exist for all peptide residues and MHC allele residues.

[0572] As another example, using the network transformation function g H (·), peptide p k The probability of being presented by MHC allele h can be generated by:

[0573]

[0574] where p k represents the peptide sequence, d h represents the peptide sequence of MHC allele h, θ H is a network model NN associated with all MHC alleles H(·) Determined parameter set.

[0575] Fig.6D shows the use of an example shared network model NN H (·) Generate peptide p associated with MHC allele h k The possibility of submission. Fig.6D As shown, the shared network model NN H (·) receiving peptide sequence p k and MHC allele peptide sequences d h , and generate the output NN H ([p k d h ]). The output is mapped by the function f(·) to generate the estimated presentation probability u k .

[0576] VIII.D.6. Allele Non-Interaction Variables

[0577] As described above, allele-independent interaction variables contain information that affects peptide presentation that is independent of the MHC allele type. For example, allele-independent interaction variables may include protein sequences at the N-terminus and C-terminus of the peptide, the protein family of the presented peptide, the RNA expression level of the peptide source gene, and any additional allele-independent interaction variables.

[0578] In one implementation, training module 316 incorporates allele non-interaction variables into the pan-allele presentation model in a similar manner to that described for independent allele model and complex allele model.For example, in some embodiments, allele non-interaction variables can be used as input to enter dependency function, and the function is separated from the dependency function for allele interaction variables.In such embodiments, the output of two independent dependency functions can be summed, and the sum obtained can be input into the transformation function to generate a presentation prediction.

[0579] VIII.D.7. Multi-allele samples

[0580] As described above, the test sample may contain multiple MHC alleles instead of a single MHC allele. In fact, most samples collected from nature contain more than one MHC allele. For example, each human genome contains six MHC class I loci. Therefore, a sample containing a human genome can contain up to six different MHC class I alleles. Therefore, samples containing multiple MHC alleles instead of a single MHC allele are typical samples of real testing situations.

[0581] In the embodiment that the test sample contains multiple MHC alleles, the pan-allele model can be used to determine the probability that a given peptide in the test sample is presented by multiple MHC alleles. However, as briefly described above, when the pan-allele model is used to predict the possibility that a peptide is presented by multiple MHC alleles, the above-mentioned pan-allele model can be used iteratively for each MHC allele of multiple MHC alleles. In other words, for each MHC allele in multiple MHC alleles, the MHC allele peptide sequence and the peptide sequence are independently input into the dependency function shared by all MHC alleles. According to these inputs, the output corresponding to the MHC allele is generated by the dependency function. This process is iteratively performed for each MHC allele in multiple MHC alleles. Therefore, each MHC allele in multiple MHC alleles is independently related to the output of the dependency function. Then the output related to each MHC allele in multiple MHC alleles is combined.

[0582] The outputs of the dependency function associated with each of the multiple MHC alleles can be combined. The manner in which the multiple outputs of the dependency function are combined can vary. For example, in some embodiments, the outputs of the dependency function iterations can be summed, and the resulting sum can be input into a transformation function to generate a presentation prediction. A formula that captures this embodiment can be written as:

[0583]

[0584] Where T is the total number of unique MHC alleles in a sample containing multiple alleles. In an alternative embodiment, each individual output of the dependency function iteration can be input into a transformation function, and the resulting outputs of the transformation function can be summed to generate a presentation prediction. A formula describing this alternative embodiment can be written as:

[0585]

[0586] Such embodiments and other embodiments are described herein in which multiple outputs of dependency functions are combined to predict the probability of a peptide being presented in a polyallelic environment.

[0587] VIII.D.8. Training of the Pan-Allele Model

[0588] Training the pan-allele model involves optimizing a set of shared parameters θ associated with the dependency function H Specifically, for the parameter θ H Optimization is performed so that the dependency function outputs a dependency score that accurately indicates whether a given MHC allele will present a given peptide sequence.

[0589] To optimize the parameter θH , using training data 170. As described above, the training data 170 used to train the model may include training samples containing cells expressing a single MHC allele, training samples containing cells expressing multiple MHC alleles, or training samples containing cells expressing a combination of a single MHC allele and multiple MHC alleles. Therefore, each data instance i from the training data 170 is input into the pan-allele model, and more specifically, into the dependency function of the pan-allele model. For example, in some embodiments, MHC allele peptide sequences and peptide sequences can be input into the pan-allele model. The pan-allele model will then process these inputs as if the model was being used conventionally. However, unlike during operation of the pan-allele model, known results of peptide presentation are also input into the model during training of the pan-allele model. In other words, the label y i In embodiments where the training sample input to the pan-allele model contains cells expressing multiple MHC alleles, for each of the multiple MHC alleles in the sample, y i Set to 1.

[0590] After each iteration of the pan-allele model using data instance i, the model determines the predicted probability of the MHC allele presenting the peptide and the known label y i Then, to minimize this difference, the pan-allele model modifies the parameter θ H In other words, the pan-allele model transforms H The loss function is minimized to determine the parameter θ H When the pan-allele model reaches a certain level of prediction accuracy, training is complete and the model is ready to use.

[0591] IX. Prediction Module

[0592] The prediction module 320 receives sequence data (e.g., infectious disease-derived peptide sequence data and / or HLA sequence data of a patient) and selects candidate antigens in the sequence data using a multi-part presentation model. Specifically, the sequence data may be a DNA sequence, an RNA sequence, and / or a protein sequence. The prediction module 320 processes the sequence data into a plurality of peptide sequences p having 8-15 amino acids. k For example, the prediction module 320 may process a given sequence "IEFROEIFJEF" into three peptide sequences "IEFROEIFJ", "EFROEIFJE", and "FROEIFJEF" having 9 amino acids.

[0593] The presentation module 320 applies one or more multi-part presentation models to at least the processed peptide sequence to estimate the presentation probability of the peptide sequence (e.g., infectious disease derived peptide sequence). Specifically, the prediction module 320 can select one or more candidate antigen peptide sequences that may be presented on the HLA molecule by applying the multi-part presentation model to the candidate antigen. In one implementation, the presentation module 320 selects candidate antigen sequences with an estimated presentation probability higher than a predetermined threshold. In another implementation, the presentation model selects N candidate antigen sequences with the highest estimated presentation probability.

[0594] In various embodiments, therapeutic agents may be provided to patients to induce an immune response. Example treatments include vaccine compositions comprising one or more selected antigens, engineered cells expressing TCRs or CARs specific for one or more selected antigens, or antibodies (e.g., bispecific antibodies) that exhibit binding specificity to one or more selected antigens.

[0595] XI.VIII.A. Example Computer

[0596] Figure 7 An example computer for implementing the entities shown in Figures 1 and 3A is shown. Computer 700 includes at least one processor 702 coupled to a chipset 704. Chipset 704 includes a memory controller hub 720 and an input / output (I / O) controller hub 722. Memory 706 and graphics adapter 712 are coupled to memory controller hub 720, and display 718 is coupled to graphics adapter 712. Storage device 708, input device 714, and network adapter 716 are coupled to I / O controller hub 722. Other embodiments of computer 700 have different architectures.

[0597] The storage device 708 is a non-transitory computer-readable storage medium, such as a hard drive, a compact disk read-only memory (CD-ROM), a DVD, or a solid-state memory device. The memory 706 holds instructions and data used by the processor 702. The input interface 714 is a touch screen interface, a mouse, a trackball or other type of pointing device, a keyboard, or a combination thereof, and is used to enter data into the computer 700. In some embodiments, the computer 700 can be configured to receive input (e.g., commands) from the input interface 714 through gestures of a user. The graphics adapter 712 displays images and other information on the display 718. The network adapter 716 couples the computer 700 to one or more computer networks.

[0598] The computer 700 is adapted to execute computer program modules to provide the functions described herein. As used herein, the term "module" refers to computer program logic for providing a specified function. Therefore, a module can be implemented in hardware, firmware and / or software. In one embodiment, the program module is stored on a storage device 708, loaded into the memory 706 and executed by the processor 702.

[0599] The type of computer 700 used by the entity of FIG. 1 may vary depending on the implementation and the processing power required by the entity. For example, the rendering determination system 160 may be run in a single computer 700 or in multiple computers 700 that communicate with each other over a network (such as in a server farm). The computer 700 may lack some of the components described above, such as a graphics adapter 712 and a display 718.

[0600] Example

[0601] Example 1: Multi-component presentation model accurately predicts presentation of viral epitopes

[0602] Various multi-part presentation models were constructed and their ability to predict the possible presentation of viral epitopes was evaluated. Specifically, three viruses were selected, including human immunodeficiency virus (HIV), influenza A virus (IAV) and SARS-COV-2. These three viruses are well documented in the literature. T cell validation epitopes are aggregated from multiple sources, including literature, immune epitope databases (IEDB) and virus-specific databases, such as the Los Alamos National Laboratory HIV CD8T cell epitope database. A more detailed description of the immune epitope database (IEDB) is provided in Vita R, et al., The Immune Epitope Database (IEDB): Updated in 2018. Nucleic Acids Res. October 24, 2018, each of which is incorporated herein by reference in its entirety. For a more detailed description of the Los Alamos National Laboratory HIV CD8 T cell epitope database, see Llano, A., et al., The 2019 Optimal HIV CTL epitopes update: growing variations inepitope length and HLA restriction. HIV Molecular Immunology 2019, 3-27, which is incorporated herein by reference in its entirety.

[0603] The first multi-part model, referred to in this paper as the “ViralEDGE” model, is a cluster of 10 pan-specific and 10 allele-specific models trained using human mass spectrometry data and all IEDB infectious disease binding affinity data. The architecture of the ViralEDGE model is as described in this paper Figure 4BDescribed. For each selected virus, the ViralEDGE model was further retrained to remove all epitopes of these three viral taxa. In the binding affinity training and validation data sets, synthetic negatives per positive were generated. Specifically, the retrained ViralEDGE model does not contain any HIV, IAV, and SARS epitopes in its training set, and has n=0 synthetic negatives per positive, referred to herein as the "New Viral EDGE (0 Synth Neg)" model. The retrained ViralEDGE model does not contain any HIV, IAV, and SARS epitopes in its training set, and has n=50 synthetic negatives per positive, referred to herein as the "New Viral EDGE (50 Synth Neg)" model. The retrained ViralEDGE model does not contain any HIV, IAV, and SARS epitopes in its training set, and has n=100 synthetic negatives per positive, referred to herein as the "New Viral EDGE (100 Synth Neg)" model.

[0604] In addition, a previously published model was further constructed, including a presentation model previously trained and deployed for predicting cancer epitope presentation (hereinafter referred to as "EDGE" model). The EDGE model does not include a multi-part presentation model, and human immunopeptidomics is trained. The more details of the EDGE model are described in U.S. Patent No. 10,055,540 (see Figure 13B, showing the prediction of the "MS" model). Another previously published model, called the MHC Flurry model, was further constructed. The MHC Flurry model incorporates binding affinity and mass spectrometry data to train the model. The more details of the MHC Flurry model are described in O'Donnell, TJ, et al., MHC Flurry 2.0: Improved Pan-Allele Prediction of MHC Class I-Presented Peptides by Incorporating Antigen Processing. Cell Systems, 11 (1), 42-48.e7, which is incorporated herein by reference as a whole.

[0605] For each selected evaluation virus, the UniProt reference proteome (The UniProt Consortium. (2014). UniProt: a hub for protein information. Nucleic Acids Research, 43 (D1), D204-212) was used to generate the reference proteome of the virus. In addition, 8-11-mer epitopes were generated for each protein sequence, and any repeated epitopes were removed due to the alternative polyprotein representation. The epitopes were analyzed by multiple models to predict the possible presentation of various class I HLA alleles.

[0606] Each model (e.g., EDGE, ViralEDGE, Novel EDGE (0 synthetic negatives), Novel EDGE (50 synthetic negatives), Novel EDGE (100 synthetic negatives), and MHC Flurry) was deployed to predict each allele. If the epitope-HLA pair was reported as CD8 positive in any of the validation sources collected, the epitope-HLA prediction was treated as correct, and all other predictions were assumed to be incorrect.

[0607] Since each virus has a different number of proteins, validated epitopes, and HLA alleles with validated epitopes to use with them, each benchmark was interpreted overall based on the set of all HLA alleles with at least one validated epitope in the dataset, independent alleles of the top 5 alleles ranked by the number of validated epitopes for that virus, or taking the top 5, 10, 20, or 25 alleles in total to detect any allele-specific bias in prediction or validation. Models were evaluated based on precision-recall curves and AUC values ​​as well as the number of positives in the top 500 predictions ranked by model score.

[0608] Reference now Figures 8A-8C , which shows the performance of various models. Figures 8A-8C The bar graphs for each allele show, from left to right, the performance of the MHC Flurry model, the EDGE model, the Viral EDGE model, the New Viral EDGE (50 synthetic negatives) model, the New Viral EDGE (0 synthetic negatives) model, and the New Viral EDGE (100 synthetic negatives) model.

[0609] Specifically, Fig. 8AThe performance of various models for predicting HIV epitope presentation in the top five HLA class I alleles is shown. The top five HLA class I alleles include HLA-A*02:01, HLA-B*07:02, HLA-A*03:01, HLA-A*11:01, and HLA-B*35:01. In general, across the five HLA class I alleles, the multipartite presentation models (ViralEDGE, Viral EDGE (0 synthetic negatives), Viral EDGE (50 synthetic negatives), Viral EDGE (100 synthetic negatives)) achieved higher precision-recall area under the curve (PR-AUC) than the two previously published models (MHCFlurry and EDGE).

[0610] Figure 8B The performance of various models for predicting influenza A epitope presentation in the top 5 alleles is shown. Here, the top 5 class I HLA alleles include HLA-A*02:01, HLA-A*11:01, HLA-A*03:01, HLA-A*24:02, and HLA-A*68:01. Similar to the results for HIV epitopes, across the 5 class I HLA alleles, the multipartite presentation models (ViralEDGE, Viral EDGE (0 synthetic negatives), Viral EDGE (50 synthetic negatives), Viral EDGE (100 synthetic negatives)) achieved higher precision-recall area under the curve (PR-AUC) compared to the two previously published models (MHCFlurry and EDGE).

[0611] Figure 8C The performance of various models for predicting SARS-CoV-2 epitope presentation in the top 5 alleles is shown. Here, at least according to the HLA-A*01:01, HLA-A*03:01, and HLA-B*40:01 alleles, the multipart presentation models (ViralEDGE, Viral EDGE (0 synthetic negatives), Viral EDGE (50 synthetic negatives), Viral EDGE (100 synthetic negatives)) achieved higher precision-recall area under the curve (PR-AUC) compared to two previously published models (MHCFlurry and EDGE). For two alleles (HLA-A*02:01 and HLA-A*24:02), the multipart presentation models (ViralEDGE, Viral EDGE (0 synthetic negatives), Viral EDGE (50 synthetic negatives), Viral EDGE (100 synthetic negatives)) achieved comparable PR-AUC values ​​compared to two previously published models (MHCFlurry and EDGE).

[0612] Overall, Figures 8A-8CWe show that various multipartite presentation models generally outperform previously published models in predicting presentation of different viral epitopes (e.g., HIV, influenza A, and SARS-CoV-2).

[0613] Further references Fig. 9A , which shows the precision-recall curves of various models for predicting HIV epitope presentation in the top 25 alleles, and Fig. 9B , which shows the precision-recall curves of various models for predicting influenza A epitope presentation in the top 25 alleles. Fig. 9A and 9B As shown, the multi-part presentation models (ViralEDGE, Novel Viral EDGE (0 synthetic negatives), Novel Viral EDGE (50 synthetic negatives), Novel Viral EDGE (100 synthetic negatives)) achieved comparable or improved PR-AUC values ​​compared to the previously published MHC Flurry model.

[0614] While the invention has been particularly shown and described with reference to a preferred embodiment and various alternative embodiments, it will be understood by those skilled in the relevant art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.

[0615] All references, issued patents, and patent applications cited in the body of this specification are incorporated herein by reference in their entirety for all purposes.

Claims

1. A method for determining one or more infectious disease-derived antigens that may be presented by cells of a subject, the method comprising: Obtain peptide sequences of multiple infectious disease-derived antigens; obtaining a sequence of one or more MHC alleles of the subject; inputting the peptide sequences of the plurality of infectious disease-derived antigens and the sequences of one or more MHC alleles of the subject into a multi-part presentation model to generate a set of numerical likelihoods that the plurality of infectious disease-derived antigens are presented by the one or more MHC alleles expressed on the surface of a cell of the subject, wherein the first part of the multi-part presentation model comprises a pan-allele model part which receives as input a peptide sequence of one or more infectious disease-derived antigens and said sequence or representation of one or more MHC alleles of said subject, and wherein the second part of the multi-part presentation model comprises a plurality of allele-specific models, each model receiving as input the peptide sequences or representations thereof of the plurality of infectious disease-derived antigens; and A subset of the plurality of infectious disease-derived antigens is selected based on the set of numerical likelihoods to generate a set of selected antigens.

2. The method of claim 1, wherein the multi-part presentation model comprises a plurality of parameters generated using at least 1) mass spectrometry data and 2) binding affinity data determined from a plurality of samples.

3. The method of claim 1, wherein the multi-part presentation model comprises a plurality of parameters generated using a training data set, the training data set comprising: training peptide sequences, and For one or more of the training peptide sequences, a label derived from the mass spectrometry data indicates whether the training peptide sequence is presented by one or more MHC class I alleles present in the plurality of samples.

4. The method of claim 3, wherein the training peptide sequences are determined by mass spectrometry analysis of separated peptides eluted from the MHC alleles present in the plurality of samples.

5. The method of claim 3 or 4, wherein the multi-part presentation model comprises a plurality of parameters generated using a training data set, the training data set comprising: For one or more of the training peptide sequences, a label derived from the binding affinity data indicates whether the training peptide sequence binds to one or more MHC class I alleles present in the plurality of samples.

6. The method of any one of claims 3-5, wherein the length of the training peptide sequence is in the range of k-mers, where k is between 8-15, inclusive.

7. The method of any one of claims 3-6, wherein the length of the training peptide sequence is in the range of k-mers, where k is between 8-11, inclusive.

8. The method of any one of claims 3-7, wherein the training peptide sequence is an infectious disease-derived training peptide sequence.

9. The method of any one of claims 1-8, wherein the pan-allele model comprises, in part, a neural network.

10. The method of claim 9, wherein a first set of layers of the neural network of the pan-allele model portion performs dimensionality reduction on the sequence of one or more MHC alleles of the subject.

11. The method of claim 9, wherein a second set of layers of the neural network of the pan-allele model portion receives as input the peptide sequence representations of the plurality of infectious disease-derived antigens and a dimensionality-reduced representation of the sequence of one or more MHC alleles of the subject.

12. The method of claim 11, wherein the representations of the peptide sequences of the plurality of infectious disease derived antigens are generated by encoding the peptide sequences via a one-hot encoding scheme.

13. The method of claim 11, wherein the second set of layers of the neural network models interactions between the peptide sequences of the plurality of infectious disease-derived antigens and the sequences of one or more MHC alleles of the subject.

14. The method of any one of claims 1-13, wherein one or more of the allele-specific models comprises a neural network.

15. The method of claim 14, wherein the neural network of the allele-specific network receives as input the peptide sequence representations of the plurality of infectious disease-derived antigens and outputs independent allelic presentation probabilities for alleles.

16. The method of claim 15, wherein the representations of the peptide sequences of the plurality of infectious disease derived antigens are generated by encoding the peptide sequences via a one-hot encoding scheme.

17. The method of any one of claims 1-16, wherein each of the allele-specific models comprises a neural network.

18. The method of any one of claims 1-16, wherein the second part of the multi-part presentation model comprises ten or more allele-specific models.

19. The method of any one of claims 1-18, wherein the numerical likelihood of an antigen is a combination of the output of the pan-allele model portion and the outputs of the multiple allele-specific models.

20. The method of any one of claims 1-18, wherein the set of numerical likelihoods is further determined by a feature comprising at least one of: (a) C-terminal sequences flanking the peptide sequences of the plurality of infectious disease derived antigens; and (b) N-terminal sequences flanking the peptide sequences of the plurality of infectious disease derived antigens.

21. The method of any one of claims 2-20, wherein the plurality of samples comprises at least one of: (a) one or more cell lines engineered to express a single MHC class I allele; (b) one or more cell lines engineered to express multiple MHC class I alleles; (c) one or more human cell lines obtained or derived from a plurality of patients; (d) fresh or frozen samples obtained from multiple patients; and (e) Fresh or frozen tissue samples obtained from multiple patients.

22. The method of any one of claims 1-21, wherein the cells of the subject comprise cells infected with one of a pathogen, a virus, a bacterium, a fungus, or a parasite.

23. The method of any one of claims 1-22, wherein the infectious disease derived antigen is derived from one of a pathogen, a virus, a bacterium, a fungus, or a parasite.

24. The method of any one of claims 1-22, wherein the infectious disease-derived antigen is derived from an infectious disease organism selected from the group consisting of severe acute respiratory syndrome-associated coronavirus (SARS), severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), Ebola virus, HIV, hepatitis B virus (HBV), influenza, hepatitis C virus (HCV), human papillomavirus (HPV), cytomegalovirus (CMV), chikungunya virus, respiratory syncytial virus (RSV), dengue virus, Orthomyxoviridae virus, tuberculosis, pan-coronavirus, herpes simplex virus infection (HSV), influenza, metapneumovirus (MPV) and parainfluenza virus (PIV).

25. A method of treating an infectious disease in a subject, comprising performing any of claims 1-24, and further comprising obtaining a vaccine comprising the collection of selected antigens, and administering the vaccine to the subject.

26. The method of claim 25, wherein the vaccine is administered to the subject prophylactically.

27. The method of claim 25, wherein the vaccine is administered to the subject therapeutically.

28. A method for manufacturing a vaccine, comprising performing one of claims 1 to 24, and further comprising producing or having produced a vaccine comprising said collection of selected antigens.

29. A vaccine comprising a collection of selected antigens selected by performing the method of any one of claims 1-24.

30. A method of treating an infectious disease in a subject, comprising performing the method of any one of claims 1-24, and further comprising obtaining an isolated antigen binding protein that exhibits binding specificity for one or more selected antigens, and administering the isolated antigen binding protein to the subject.

31. A method of treating an infectious disease in a subject, comprising performing the method of any one of claims 1-24, and further comprising obtaining an isolated antigen binding protein that exhibits binding specificity for one or more selected antigens, and administering the isolated antigen binding protein to the subject.

32. The method of claim 31, wherein the antigen binding protein is an antibody or an antigen binding fragment.

33. The method of claim 31, wherein the antigen binding protein is a T cell receptor or a chimeric antigen receptor.

34. A non-transitory computer readable medium comprising instructions that, when executed by a processor, cause the processor to: Obtain peptide sequences of multiple infectious disease-derived antigens; obtaining a sequence of one or more MHC alleles of the subject; inputting the peptide sequences of the plurality of infectious disease-derived antigens and the sequences of one or more MHC alleles of the subject into a multi-part presentation model to generate a set of numerical likelihoods that the plurality of infectious disease-derived antigens are presented by the one or more MHC alleles expressed on the surface of a cell of the subject, wherein the first part of the multi-part presentation model comprises a pan-allele model part which receives as input a peptide sequence of one or more infectious disease-derived antigens and said sequence or representation of one or more MHC alleles of said subject, and wherein the second part of the multi-part presentation model comprises a plurality of allele-specific models, each model receiving as input the peptide sequences or representations thereof of the plurality of infectious disease-derived antigens; and A subset of the plurality of infectious disease-derived antigens is selected based on the set of numerical likelihoods to generate a set of selected antigens.

35. The non-transitory computer readable medium of claim 34, wherein the multi-part presentation model comprises a plurality of parameters generated using at least 1) mass spectrometry data and 2) binding affinity data determined from a plurality of samples.

36. The non-transitory computer-readable medium of claim 34, wherein the multi-part presentation model comprises a plurality of parameters generated using a training data set, the training data set comprising: training peptide sequences, and For one or more of the training peptide sequences, a label derived from the mass spectrometry data indicates whether the training peptide sequence is presented by one or more MHC class I alleles present in the plurality of samples.

37. The non-transitory computer readable medium of claim 36, wherein the training peptide sequences are determined by mass spectrometry analysis of isolated peptides eluted from MHC alleles present in the plurality of samples.

38. The non-transitory computer-readable medium of claim 36 or 37, wherein the multi-part presentation model comprises a plurality of parameters generated using a training data set, the training data set comprising: For one or more of the training peptide sequences, a label derived from the binding affinity data indicates whether the training peptide sequence binds to one or more MHC class I alleles present in the plurality of samples.

39. The non-transitory computer readable medium of any one of claims 36-38, wherein the length of the training peptide sequence is in the range of k-mers, where k is between 8-15, inclusive.

40. The non-transitory computer readable medium of any one of claims 36-39, wherein the length of the training peptide sequence is in the range of k-mers, where k is between 8-11, inclusive.

41. The non-transitory computer readable medium of any one of claims 36-40, wherein the training peptide sequence is an infectious disease-derived training peptide sequence.

42. The non-transitory computer-readable medium of any one of claims 34-41, wherein the pan-allele model comprises, in part, a neural network.

43. The non-transitory computer-readable medium of claim 42, wherein the first set of layers of the neural network of the pan-allele model portion performs dimensionality reduction on the sequence of one or more MHC alleles of the subject.

44. A non-transitory computer-readable medium as described in claim 42, wherein the second set of layers of the neural network of the pan-allele model portion receives as input the peptide sequence representations of the multiple infectious disease-derived antigens and the reduced dimension representations of the sequences of one or more MHC alleles of the subject.

45. The non-transitory computer-readable medium of claim 44, wherein the representations of the peptide sequences of the plurality of infectious disease-derived antigens are generated by encoding the peptide sequences via a one-hot encoding scheme.

46. ​​The non-transitory computer readable medium of claim 44, wherein the second set of layers of the neural network models interactions between the peptide sequences of the plurality of infectious disease-derived antigens and the sequences of one or more MHC alleles of the subject.

47. The non-transitory computer-readable medium of any one of claims 34-46, wherein one or more of the allele-specific models comprises a neural network.

48. The non-transitory computer-readable medium of claim 47, wherein the neural network of the allele-specific network receives as input the peptide sequence representations of the plurality of infectious disease-derived antigens and outputs independent allelic presentation probabilities for alleles.

49. The non-transitory computer-readable medium of claim 48, wherein the representations of the peptide sequences of the plurality of infectious disease-derived antigens are generated by encoding the peptide sequences via a one-hot encoding scheme.

50. The non-transitory computer-readable medium of any one of claims 34-49, wherein each of the allele-specific models comprises a neural network.

51. The non-transitory computer-readable medium of any one of claims 34-50, wherein the second part of the multi-part presentation model comprises ten or more allele-specific models.

52. The non-transitory computer-readable medium of any one of claims 34-51, wherein the numerical likelihood of an antigen is a combination of an output of the pan-allele model portion and outputs of the plurality of allele-specific models.

53. The non-transitory computer-readable medium of any one of claims 34-52, wherein the set of numerical likelihoods is further determined by a feature comprising at least one of: (a) C-terminal sequences flanking the peptide sequences of the plurality of infectious disease derived antigens; and (b) N-terminal sequences flanking the peptide sequences of the plurality of infectious disease derived antigens.

54. The non-transitory computer-readable medium of any one of claims 35-53, wherein the plurality of samples comprises at least one of: (a) one or more cell lines engineered to express a single MHC class I allele; (b) one or more cell lines engineered to express multiple MHC class I alleles; (c) one or more human cell lines obtained or derived from a plurality of patients; (d) fresh or frozen samples obtained from multiple patients; and (e) Fresh or frozen tissue samples obtained from multiple patients.

55. The non-transitory computer-readable medium of any one of claims 34-54, wherein the cells of the subject comprise cells infected with one of a pathogen, a virus, a bacterium, a fungus, or a parasite.

56. The non-transitory computer readable medium of any one of claims 34-54, wherein the infectious disease derived antigen is derived from one of a pathogen, a virus, a bacterium, a fungus, or a parasite.

57. The non-transitory computer-readable medium of any one of claims 34-54, wherein the infectious disease-derived antigen is derived from an infectious disease organism selected from the group consisting of severe acute respiratory syndrome-associated coronavirus (SARS), severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), Ebola virus, HIV, hepatitis B virus (HBV), influenza, hepatitis C virus (HCV), human papillomavirus (HPV), cytomegalovirus (CMV), chikungunya virus, respiratory syncytial virus (RSV), dengue virus, orthomyxoviridae virus, tuberculosis, pan-coronavirus, herpes simplex virus infection (HSV), influenza, metapneumovirus (MPV) and parainfluenza virus (PIV).

Images