How to design a vaccine

Abstract

1. A computer-implemented method for selecting one or more amino acid sequences for inclusion in a neoantigen vaccine from a set of candidate neoantigen amino acid sequences, the method comprising the steps of: obtaining a set of input data related to a patient; simulating a plurality of cancer cells based on the set of input data, where simulating each cancer cell comprises predicting a cell surface presentation of the cancer cell; for each candidate neoantigen amino acid sequence, predicting a likelihood of the candidate neoantigen amino acid sequence to elicit an immune response against the cancer cell based on the predicted cell surface presentation of each cancer cell; and selecting one or more amino acid sequences for inclusion in the vaccine that maximizes the likelihood of the vaccine eliciting an immune response against the cancer cell based on the predicted likelihood of each candidate neoantigen amino acid sequence to elicit an immune response against the cancer cell.

Description

[Technical field]

[0001] The present invention relates to the field of vaccine design and production, including the selection of amino acid sequences for inclusion in a vaccine, and the synthesis of one or more amino acid sequences to produce a vaccine. [Background technology]

[0002] In recent years, there has been an increasing interest in developing therapeutic cancer vaccines, which is a form of cancer immunotherapy that is used to stimulate the patient's immune system to attack and kill cancer cells.Healthy cells do not contain the same DNA alterations that exist in cancer cells, which makes these DNA alterations as well as the associated proteins and peptides that are synthesized and processed in cancer cells, which are potential targets for vaccines.

[0003] Neoantigen vaccines specifically target neoantigens, proteins that form in cancer cells in response to mutations in the tumor's DNA, and aim to enable the human immune system to avoid non-specific or autoimmune responses.

[0004] Inside all human cells, DNA is transcribed into messenger RNA (mRNA), and then mRNA is translated into protein.If the DNA of a cell contains a mutation (i.e., a change in DNA), the mutation can also be transcribed into mRNA, causing a change in the amino acid sequence of the protein synthesized in the cell.These altered proteins are generally not useful to the cell, and therefore are processed in one of two antigen processing pathways, each of which always leads to the cleavage of the protein into peptides.

[0005] Intrinsic processing pathway: Under this pathway, in the same cell in which the protein was synthesized, the proteasome splits the protein into subunits called peptides. These peptides can then be transported to the endoplasmic reticulum (ER) where they can bind to major histocompatibility complex I proteins (MHC-I). After binding to MHC-I proteins, the peptide-MHC-I complexes can then be presented on the surface of the cell. When the peptide-MHC-I complex is presented on the surface of an antigen-presenting cell (APC), a T cell can bind to it by its T cell receptor (TCR), which recognizes the peptide, also called the epitope, in combination with its co-receptor, the cluster of differentiation 8 receptor (CD8+). The T cell induces cell death of the presenting cell. For this reason, CD8+ T cells are also called cytotoxic T cells (CTL).

[0006] Exogenous processing pathway: Under this pathway, proteins containing mutations are absorbed by cells via endocytosis. Similar to what happens in the intrinsic processing pathway, malformed proteins are degraded by proteases into small sequences of amino acids (peptides). The peptides then bind to major histocompatibility complex II proteins (MHC-II), and the peptide-MHC-II complexes are presented on the cell surface of antigen-presenting cells (APCs). T cells with cluster of differentiation 4 receptor proteins (CD4+) bind to the peptide-MHC-II complexes. After this event, CD4+ T cells release substances called cytokines that can activate B cells or CTLs. Thus, CD4+ T cells are also called helper T cells.

[0007] In humans, MHC is also called human leukocyte antigens (HLA). In the human genome, there are MHC-I (also called HLA-I) and MHC-II (also called HLA-II) genes. Each individual has a set of three major HLA-I genes: HLA-, HLA-B, and HLA-C. For each of these genes, an individual has two versions, called alleles, inherited by the father and the mother. Thus, within an individual, there can be up to six different major HLA-I molecules, each of which binds to a different set of epitopes.

[0008] For HLA-II, there are also three major genes: HLA-DR, HLA-DP and HLA-DQ. For each gene, each individual has two alleles inherited by the father and the mother. However, the HLA-II system is more complex than the HLA-I system: the HLA-II molecule is a heterodimeric complex formed by polymorphic genes (alpha and beta chains). This allows each individual to have up to 12 different HLA-II complexes. In addition, the epitopes presented by HLA-II are longer and more variable in length compared to HLA-I.

[0009] The intrinsic and extrinsic processing pathways are discussed in more detail in Alberts, B.; Johnson A., Lewis, J., Raff, M., Roberts, K. & Walter, P., Molecular Biology of the Cell. Garland Science, 2002.

[0010] As described above, the pathway by which epitopes, including neoepitopes / neoantigens, can induce an immune response is complex and involves many steps. Any of these steps (e.g., binding of the epitope to an HLA molecule or presentation of the epitope-HLA on the surface of a cell) may not function. This may make some tumor mutations good candidates for a neoantigen vaccine, while others may not be as promising. For example, some mutations may not be translated into protein, in which case the pathway described above is not activated in the first place. Other mutations that are translated into protein may result in peptides that do not bind well to the HLA complex of a given individual. Furthermore, it may be possible that the neoepitope-MHC complex is presented on the surface of a cell, but is not recognized by T cells.

[0011] Therefore, to develop an effective neoantigen vaccine, it is important to understand which neoantigen candidates are best to include in the vaccine. [Prior art documents] [Non-patent literature]

[0012] [Non-Patent Document 1] Alberts, B., Johnson, A., Lewis, J., Raff, M., Roberts, K. & Walter, P., Molecular Biology of the Cell. Garland Science, 2002 [Non-Patent Document 2] GATK Best Practices (https: / / software.broadinstitute.org / gatk / best-practices / workflow?id=11146 13.10.2021) [Non-Patent Document 3] Cufflinks (http: / / cole-trapnell-lab.github.io / cufflinks / cufflinks / #fpkm-tracking-format 14.10.2021) [Non-Patent Document 4] NetMHC (Gapped sequence alignment using artificial neural networks: application to the MHC class I system; Andreatta M, Nielsen M, Bioinformatics 32.4 (2016): pp. 511~517) [Non-Patent Document 5] MHCflurry (MHCflurry 2.0: Improved Pan-Allele Prediction of MHC Class I-Presented Peptides by Incorporating Antigen Processing, Timothy J.O'Donnell, Alex Rubinsteyn, Uri Laserson, Cell systems 11.1 (2020): pp. 42-48) [Non-Patent Document 6] Antigen receptor repertoire profiling from RNA-seq data; Dmitriy A Bolotin, Stanislav Poslavsky, Alexey N Davydov, Felix E Frenkel et al., Nature Biotechnology 35.10 (2017): pp. 908-911 Summary of the Invention [Problem to be solved by the invention]

[0013] Aspects of the present invention provide methods and systems for selecting a set of candidate neoantigen elements for inclusion in a vaccine that maximizes the likelihood of the vaccine eliciting an immune response against cancer cells in a patient. [Means for solving the problem]

[0014] According to a first aspect of the present invention, there is provided a computer-implemented method for selecting one or more amino acid sequences for inclusion in a neoantigen vaccine from a set of candidate neoantigen amino acid sequences, the method comprising: obtaining a set of input data related to a patient, simulating a plurality of cancer cells based on the set of input data, where simulating each cancer cell comprises predicting cell surface presentation of the cancer cell, for each candidate neoantigen amino acid sequence, predicting a likelihood of the candidate neoantigen amino acid sequence to elicit an immune response against the cancer cell based on the predicted cell surface presentation of each cancer cell, and selecting one or more amino acid sequences for inclusion in the vaccine that maximize the likelihood of the vaccine to elicit an immune response against the cancer cell based on the predicted likelihood of each candidate neoantigen amino acid sequence to elicit an immune response against the cancer cell.

[0015] The first aspect of the present invention allows optimizing the composition of a therapeutic cancer vaccine. In contrast to conventional approaches, the present invention does this by simulating the population of cancer cells in a patient and then predicting the likelihood of a vaccine to induce an immune response against those cancer cells. By maximizing this likelihood, a set of vaccine elements (amino acid sequences) can then be selected to optimize the composition of the vaccine.

[0016] An additional advantage of the present invention is that it allows for an estimation of the amount of cancer cells killed as well as an assessment of the immune response likely to be induced by a vaccine, allowing the breadth of vaccine effectiveness to be estimated in a way that is not possible with traditional approaches to selecting vaccine compositions.

[0017] Those skilled in the art will understand that maximizing the likelihood of vaccine to induce immune response against cancer cells involves using one of several optimization methods, each of which can lead to different maximum values ​​being reached.Similarly, there can be different measures of the likelihood of vaccine to induce immune response against cancer cells.Therefore, this process can be carried out in different ways, which can lead to different amino acid sequences being selected for inclusion in vaccine.

[0018] Advantageously, the step of simulating the plurality of cancer cells involves modeling one or more of the biochemical processes occurring in the cancer cells. To this end, the set of input data advantageously comprises one or more of: a representation of the patient's HLA-I alleles, gene expression information, a set of identified genetic variants, a binding affinity index for each tuple of candidate neoantigen amino acid sequences and HLA-I alleles, and a presentation index for each tuple of candidate neoantigen amino acid sequences and HLA-I alleles.

[0019] The biochemical processes that may be simulated preferably include one or more of the steps of the endogenous processing pathway: transcription, translation, intracellular processing, HLA binding, and cell surface presentation.

[0020] To simulate the transcriptional steps of an endogenous processing pathway, simulating the plurality of cancer cells preferably includes predicting the presence or absence of each of the identified genetic variants in each of the plurality of cancer cells based on the statistical distribution of the identified genetic variants.

[0021] To simulate the translation steps of an endogenous processing pathway, simulating the plurality of cancer cells preferably includes estimating the abundance of one or more proteins synthesized in each cancer cell based on gene expression information and genetic variants predicted to be present in each cancer cell.

[0022] To simulate the intracellular processing steps of an endogenous processing pathway, simulating a plurality of cancer cells includes estimating the abundance of one or more peptides processed in each cancer cell based on the estimated abundance of one or more proteins synthesized in the cancer cell and the likelihood of each of the one or more proteins being divided into one or more peptides.

[0023] To simulate the HLA binding step of the endogenous processing pathway, simulating the plurality of cancer cells preferably comprises simulating the binding of one or more peptides to HLA molecules to estimate the likelihood of one or more peptide-HLA complexes being present in each cancer cell, and simulating the binding of peptides to HLA molecules is based on the abundance of said one or more peptides and a binding affinity index for each tuple of candidate neoantigen amino acid sequence and HLA-I allele.

[0024] To simulate the cell surface presentation step of the endogenous processing pathway, simulating the plurality of cancer cells preferably includes predicting the cell surface presentation of each cancer cell based on the likelihood of one or more peptide-HLA complexes being present in each cancer cell and the presentation index for each tuple of candidate neoantigen amino acid sequence and HLA-I allele.

[0025] Simulating a population of cancer cells (also called a cancer digital twin) through stochastic simulation of endogenous processing pathways allows combining statistical predictions (including machine learning predictions) with mechanistic models of cancer cells. In other words, the properties of data-driven approaches are combined with those of model-driven approaches, whereby information extracted from data is combined with knowledge of the biochemistry of cancer cells. This combined approach allows for a more accurate simulation of cancer cells.

[0026] The process of simulating the binding of peptides to HLA molecules allows for improved simulation of cancer cells in particular. A given neoantigen and a given HLA-I molecule can bind with a given affinity, with pairs with stronger affinity having a greater probability of binding. Pairs with lesser affinity can also bind, but with a smaller probability. However, there is always a limited amount of neoantigens and HLA-I molecules present in a cancer cell. Thus, the binding of neoantigens to HLA-I molecules is competitive. By taking into account the estimated abundance of peptides in cancer cells and the affinity of said peptides to HLA-I molecules, this competitive process can be modeled to estimate the likelihood of one or more peptide-HLA complexes present in each cancer cell.

[0027] This process significantly affects the cell surface presentation of cancer cells and, as a consequence, the immune response to those cells, and therefore simulating this process allows for a more accurate simulation of cancer cells.

[0028] The step of predicting the immune response advantageously comprises estimating the likelihood of the patient's immune system comprising T cells with receptors that bind to the surface of cancer cells. Although the immune response can be predicted directly by predicting the TCR-peptide-HLA binding, it is preferred to estimate the likelihood of neoantigen-presenting cancer cells to be killed by T cells, i.e. the possibility of the presence of T cells that are able to reach the tumor and recognize the neoantigen. This can be calculated by taking into account the information of tumor infiltrating lymphocytes (TIL), i.e. T cells present in the tumor sequencing data. This data consists of TCR information such as V, D and J alleles, CDR3 sequences, TIL marker genes and corresponding cancer cell markers.

[0029] To this end, where the likelihood of the patient's immune system comprising T cells with receptors that bind to the surface of cancer cells must be estimated, the input data preferably further comprises TCR repertoire and associated gene expression data, which can then be used to determine the T cells present in the patient's immune system and to estimate the likelihood of any of the T cells having receptors that bind to the surface of cancer cells.

[0030] As mentioned above, there are various optimization processes that can be used to maximize the likelihood of a vaccine that induces an immune response against cancer cells. In one such process, the step of selecting one or more amino acid sequences for inclusion in the vaccine involves applying a mathematical optimization algorithm to minimize the likelihood of the vaccine not eliciting an immune response against cancer cells.

[0031] Those skilled in the art will understand that maximizing the likelihood that a phenomenon occurs is equivalent to minimizing the likelihood that the phenomenon does not occur. Reconstructing the process of selecting one or more amino acid sequences as minimizing the likelihood of a vaccine that does not induce an immune response against cancer cells advantageously allows the use of a mathematical optimization algorithm based on minimizing the flow in a network where a set of nodes corresponds to candidate neoantigen amino acid sequences, a set of nodes corresponds to multiple cancer cells, and there is one sink. Thus, the optimized vaccine component minimizes the likelihood of no response across the entire population of cancer cells.

[0032] The variables of the mathematical optimization algorithm preferably include (a) a binary indicator variable for each candidate neoantigen amino acid sequence indicating whether the candidate amino acid is included in the vaccine, and (b) a continuous variable for each cancer cell giving the log-likelihood that there is no immune response elicited by the candidate neoantigen amino acid sequence against said cancer cell.

[0033] A continuous variable giving the log-likelihood for each cancer cell of the absence of an immune response induced by a candidate neoantigen amino acid sequence against said cancer cell can be estimated using the predicted likelihood of each candidate neoantigen amino acid sequence to induce an immune response against the cancer cell, simplifying the optimization problem to find a binary indicator variable that minimizes the total flow from amino acid sequence nodes to cancer cell nodes.

[0034] Alternatively, the step of selecting one or more amino acid sequences for inclusion in the vaccine may involve applying a mathematical optimization algorithm to minimize the estimated likelihood of the vaccine not eliciting an immune response to cells with the greatest estimated likelihood of no vaccine-elicited immune response.

[0035] This approach uses a mathematical optimization algorithm that is also based on minimizing the flow in a network with a set of nodes corresponding to candidate neoantigen amino acid sequences and a set of nodes corresponding to multiple cancer cells. However, in this case, each cell is a sink and the flow to each sink is minimized. Thus, the optimized vaccine composition minimizes the likelihood that there is no immune response elicited against any one cancer cell. Thus, while the first mathematical optimization algorithm results in a vaccine composition that kills the maximum number of cancer cells, this alternative mathematical optimization algorithm results in a vaccine composition that maximizes the likelihood of eliciting at least some immune response against all cancer cells.

[0036] The variables of this second mathematical optimization algorithm preferably include (a) a binary indicator variable for each candidate neoantigen amino acid sequence indicating whether the candidate amino acid is included in the vaccine, and (b) a continuous variable for each cancer cell giving the log-likelihood that there is no immune response induced by the candidate neoantigen amino acid sequence against said cancer cell. These are the same variables as discussed above. The variables preferably further include (c) a continuous variable for each cancer cell giving the log-likelihood that there is no immune response induced by a vaccine comprising a subset of the set of candidate neoantigen amino acid sequences, and (d) a continuous variable giving the maximum log-likelihood that no cancer cell responds to a vaccine comprising a subset of the set of candidate neoantigen amino acid sequences.

[0037] A continuous variable for each cancer cell that gives the log-likelihood of the absence of an immune response induced by a vaccine containing a subset of the set of candidate neoantigen amino acid sequences is used to calculate a continuous variable that gives the maximum log-likelihood that no cancer cell will respond to a vaccine containing a subset of the set of candidate neoantigen amino acid sequences.

[0038] Whatever mathematical optimization algorithm is used, preferably this is an integer linear program.

[0039] As will be appreciated, the vaccine platform used constrains the total length of the amino acid sequences included. Therefore, it is preferred that the method further comprises assigning a cost to each candidate amino acid sequence, and the step of selecting one or more amino acid sequences for inclusion in the vaccine is constrained based on the cost assigned to each candidate amino acid sequence, such that the selected one or more amino acid sequences have a total cost below a pre-set threshold budget.

[0040] This constraint is most preferably used to constrain the mathematical optimization algorithm used to select one or more amino acid sequences for inclusion in the vaccine. Integer linear programming is particularly well suited to solving such constrained optimization problems.

[0041] According to a second aspect of the present invention, there is provided a method of making a vaccine, the method comprising the steps of selecting one or more amino acid sequences for inclusion in the vaccine from a set of predicted neoantigen candidate amino acid sequences by a method according to an embodiment of the first aspect of the present invention, and synthesizing the one or more amino acid sequences or encoding the one or more amino acid sequences into a corresponding DNA or RNA sequence and / or integrating the DNA or RNA sequence into the genome of a bacterial or viral delivery system to make the vaccine.

[0042] According to a third aspect of the present invention, there is provided a system for selecting one or more amino acid sequences for inclusion in a vaccine from a set of predicted neoantigen candidate amino acid sequences, the system comprising at least one processor in communication with at least one storage device having instructions stored thereon for causing the at least one processor to perform a method according to an embodiment of the first aspect of the present invention.

[0043] According to a fourth aspect of the invention there is provided a computer readable medium having stored thereon computer executable instructions for carrying out a method according to an embodiment of the first aspect of the invention.

[0044] Embodiments of the present invention will now be described with reference to the figures. [Brief description of the drawings]

[0045] [Figure 1] FIG. 1 shows a specific example of a method used to select neoantigen candidate amino acid sequences for a neoantigen vaccine. [Diagram 2] FIG. 1 illustrates a network flow problem to be solved to select the best neoantigen vaccine composition in an embodiment of the present invention. [Diagram 3]FIG. 1 illustrates a network flow problem to be solved to select the best neoantigen vaccine composition in another embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0046] The following provides a specific example for the selection of neoantigen candidate amino acid sequences for a neoantigen vaccine with reference to Figure 1. Please note that in the proposed implementations presented below, all references set forth herein are incorporated by reference.

[0047] For each step discussed below that involves sampling from a distribution, these distributions can also be learned by machine learning algorithms, if suitable data is available.

[0048] Input data In step S101, a set of input data is obtained relating to the patient, and advantageously including a representation of the patient's HLA-I alleles, gene expression information, a set of identified genetic variants, a binding affinity index for each tuple of candidate neoantigen amino acid sequences and HLA-I alleles, a presentation index for each tuple of candidate neoantigen amino acid sequences and HLA-I alleles, as well as TCR repertoire and associated gene expression data, which data may be associated with the patient and used to simulate cancer cells present in the patient.

[0049] A set of predicted candidate neoantigen amino acid sequences is also obtained in step S101. The neoantigen candidates may be proteins similar to neoantigen proteins formed in cancer cells in response to mutations in the tumor's DNA, allowing for the creation of protein-based vaccines, or may include DNA or RNA sequences, allowing for the creation of DNA or RNA-based vaccines, such as mRNA vaccines.

[0050] Thus, the phrase "candidate neoantigen amino acid sequence" refers to a sequence of amino acids that form a neoantigen protein, although the vaccine itself need not contain the selected amino acid sequence, but may contain the corresponding DNA or RNA sequence.

[0051] Candidate neoantigen amino acid sequences are called "candidates" because, in principle, they could be selected as vaccine elements. However, a given neoantigen is not typically present in all potential cancer cells. Furthermore, many neoantigens do not elicit an immune response. Thus, the goal of the present invention is to identify the best subset of neoantigen candidates that maximizes the likelihood of eliciting an immune response.

[0052] The neoantigen candidates may also be used to simulate cancer cells in step S102. This simulation can be performed in multiple ways, but preferably involves simulating the processes of transcription, translation, intracellular processing, HLA binding, and cell surface presentation to simulate a population of cancer cells, also referred to as a cancer digital twin.

[0053] Cancer Digital Twin: Probabilistic Simulation of Cancer Cells In a first stage of transcription, the presence of one or more genetic variants in each simulated cancer cell is determined by sampling from a statistical distribution (e.g., a Bernoulli distribution). A population of simulated cancer cells is then assigned genetic variants according to this sampling.

[0054] As mentioned above, the input data includes a set of identified genetic variants, which are preferably somatic variants that can be derived from whole genome sequencing data. Identifying somatic variants is well understood in the art and involves comparing exome data for tumor samples with exome data for healthy tissues to identify genetic variants that appear in tumor samples but not in healthy tissues. For example, GATK Best Practices (https: / / software.broadinstitute.org / gatk / best-practices / workflow?id=11146 13.10.2021) can be used to perform this step.

[0055] A parametric distribution of gene variants is then obtained by using the DNA variant allele frequency (VAF), which is the percentage of reads that matched the variant divided by the sum of all reads that matched the gene.

[0056] The translation step then aims to estimate protein abundance in each simulated cancer cell from the gene expression information. This involves sampling from the distribution of FPKM (fragments per kilobase per million mapped reads) values ​​and FPKM variance values ​​(based on 95% confidence intervals) and multiplying this value by the RNA VAF, since the FPKM value is calculated based on all reads that match a gene, only a small fraction of which (=VAF) contain actual mutations.

[0057] For this purpose, the gene expression information obtained in step S101 advantageously comprises a table of FPKM RNA-sequencing-based gene expression data. Each gene is identified by its name and ENSG-identifier and has an FPMK value and an FPKM variance value. This input type may for example be based on Cufflinks (http: / / cole-trapnell-lab.github.io / cufflinks / cufflinks / #fpkm-tracking-format 14.10.2021), which provides smaller and larger ranges of the 95% confidence interval and can be used to calculate the FPKM variance value.

[0058] After estimating protein abundance in each simulated cancer cell, intracellular processing of each simulated cancer cell is simulated to estimate peptide abundance in each simulated cancer cell.

[0059] To aid in this step of simulating a cancer cell population, step S101 includes receiving a binding affinity index for each tuple of candidate neoantigen amino acid sequence and HLA-I allele and a presentation index for each tuple of candidate neoantigen amino acid sequence and HLA-I allele, but alternatively step S101 may include deriving a binding affinity index for each tuple of candidate neoantigen amino acid sequence and HLA-I allele and a presentation index for each tuple of candidate neoantigen amino acid sequence and HLA-I allele.

[0060] These indices, also called HLA binding and presentation scores, can be predicted by machine learning models, such as NetMHC (Gapped sequence alignment using artificial neural networks: application to the MHC class I system; Andreatta M, Nielsen M, Bioinformatics 32.4 (2016): pp. 511-517) or MHCflurry (MHCflurry 2.0: Improved Pan-Allele Prediction of MHC Class I-Presented Peptides by Incorporating Antigen Processing, Timothy J. O'Donnell, Alex Rubinsteyn, Uri Laserson, Cell systems 11.1 (2020): pp. 42-48) for MHC binding, as well as other predictors for other steps of intracellular processing to derive the presentation score.

[0061] In addition to being based on neoantigen candidates, these indices are predicted based on the patient's representation of HLA-I alleles, also called HLA typing, which can be determined from WXS (whole exome sequencing) data of healthy cells. HLA typing can be at the gene level (HLA-A, HLA-B, HLA-C, etc.) and, where possible, allele-specific.

[0062] The intracellular processing steps themselves then simulate peptide processing and assign a weight to each peptide that reflects the likelihood that the peptide will be present after the protein to which it belongs has been processed. The weights are calculated from the presentation scores, but may also be implemented by using individual weights for each relevant processing step, such as cleavage, trimming, transport to the endoplasmic reticulum, etc.

[0063] Once the peptide abundance in each simulated cancer cell is estimated, an HLA binding step is performed to simulate the competitive binding of peptides to available MHC (HLA) molecules by taking into account the MHC-peptide binding prediction score and also the MHC molecules (derived from HLA typing).

[0064] Finally, the cell surface presentation of each cancer cell is simulated. The presentation score is used as a sampling probability to determine whether the peptide-HLA (also called peptide-MHC) complex present in each simulated cancer cell is also presented on the cell surface. In other words, the binding likelihood of the peptide-HLA complex present in each simulated cancer cell and each peptide-HLA complex presented on the surface of the cancer cell is used to simulate the cell surface presentation of each simulated cancer cell.

[0065] After simulating cell surface presentation of a population of simulated cancer cells, the TCR recognition likelihood for all presented neoantigen-MHC-I complexes is predicted in step S103.

[0066] TCR recognition Rather than directly predicting TCR-peptide-HLA binding, it is preferable to estimate the likelihood of neoantigen-presenting cancer cells being killed by T cells, i.e. the probability of the presence of T cells that can reach the tumor and recognize the neoantigen. This can be calculated by considering the information of tumor infiltrating lymphocytes (TILs), i.e. T cells present in tumor sequencing data. This data consists of TCR information such as V, D and J alleles, CDR3 sequences, TIL marker genes and corresponding cancer cell markers. This is advantageous over existing algorithms, since predicting TCR-peptide-HLA binding is not very reliable when dealing with neoantigens.

[0067] The information used in this step is present in the TCR repertoire and associated gene expression data received in step S101. The TCR repertoire data is a table containing CDR3 sequences (nucleotides and amino acids), V, D and J alleles and clone or read counts or comparable quantification information, and may be provided by MiXCR, for example (Antigen receptor repertoire profiling from RNA-seq data; Dmitriy A Bolotin, Stanislav Poslavsky, Alexey N Davydov, Felix E Frenkel et al., Nature Biotechnology 35.10 (2017): pp. 908-911). The associated gene expression may be provided by the FPKM table discussed above in connection with the gene expression information, or alternatively or in addition, a comparable gene expression table may be provided for use in the TCR recognition step S103. These further gene expression tables include TCR gene-specific references. For example, they may contain more different V and J allele sequences and T cell surface marker genes. In addition, genes involved in the interaction between T cells and cancer cells, in particular checkpoints such as CTLA4 and PD1 (and PDL1 in cancer cells), are taken into account.

[0068] Thus, step S103 allows for a prediction of the likelihood of an immune response against each simulated cancer cell being elicited by each candidate neoantigen amino acid sequence, which can then be used in step S104 to select one or more amino acid sequences for inclusion in the vaccine that maximize the estimated likelihood of the vaccine eliciting an immune response against the cell.

[0069] Optimization: Selecting the best neoantigen vaccine composition

[0070]

number

[0071] Let be the set of neoantigen vaccine element candidates (one of the system inputs).

[0072]

number

[0073] Let be a set of cancer cells simulated by a stochastic simulation of cancer cells. C is sometimes called the “cancer digital twin.” Let V ⊂ NeoAg be any subset of NeoAg.

[0074] The optimization block of the system aims to find the best set V that maximizes the likelihood P(R=+|V,C) of a vaccine inducing an immune response against a simulated population of cells C. Hence, the optimization problem can be formulated as:

[0075]

number

[0076] This optimization problem can be addressed in several different ways, and two approaches based on network flows are specifically discussed here, although it will be appreciated that other approaches can also be used within the scope of the present invention.

[0077] EMBODIMENT 1 This embodiment is directed to designing a vaccine that aims to induce an immune response that is intended to kill the maximum number of cancer cells.

[0078] From equation (1), we define:

[0079]

number

[0080] By modeling the probability of not having an immune response in all cells as the joint probability of having an immune response in each cell, and assuming conditional independence, one can write:

[0081]

number

[0082] The set of selected vaccine elements V is then passed through a set of Boolean selectors.

[0083]

number

[0084] where

[0085]

number

[0086] Therefore, the problem of finding the best set V is the same as finding the best set X.

[0087] Vaccine V is the element v i A positive response is considered to occur if at least one of the following induces a positive immune response: j The probability of no response for P(R=-|V,c j ) is the joint likelihood that all elements fail:

[0088]

number

[0089] In the formula, x iis the selector of the i-th neoantigen, defining its inclusion in the vaccine. In other words, for each neoantigen that is not selected, this term in the binding likelihood is set to 1, which can be understood to define that a non-existent neoantigen cannot induce an immune response.

[0090] therefore:

[0091]

number

[0092] The vaccine contains the vaccine element v as follows: i By including j Define the log probability of not having an immune response for: p ij :=logP(R=-|v i ,c j ) (6)

[0093] Now we can rewrite objective (2) as follows:

[0094]

number

[0095] Each neoantigen v i About cost k i Define , which may be a constant or a function of the peptide length. Objective (7) is therefore constrained by:

[0096]

number

[0097] where b is the budget determined by the vaccine platform used.

[0098] We approach this problem as a kind of network flow problem, illustrated in Figure 2, with one set of nodes corresponding to vaccine elements, one set corresponding to cancer cells, and one sink. The goal is to select a set of vaccine elements such that the likelihood of no response across the population of cells is minimized. Integer Linear Programming (ILP) allows us to optimize the objective (7) subject to constraints (8).

[0099] EMBODIMENT 2 While the same formulation of the problem still persists (see Equation 1), in this embodiment vaccine design is approached by minimizing the probability of non-response for cells that have the greatest probability of non-response, i.e.:

[0100]

number

[0101] This approach will result in the design of a vaccine that is intended to induce at least some response against all cancer cells.

[0102] From Equation 4 and Equation 6 we derive:

[0103]

number

[0104] Standard ILP solvers cannot directly solve this minimax problem; however, we use standard techniques for a set of surrogates to tackle this problem. In particular, the cell c j is the log-likelihood of no response for

[0105]

number

[0106] Define:

[0107]

number

[0108] Additionally, we define:

[0109]

number

[0110] That is, z is the maximum log-likelihood that any cell will not respond to the vaccine (or, alternatively, the minimum log-likelihood that any cell will respond to the vaccine). Finally, our objective is then to minimize z:

[0111]

number

[0112] Our problem is effectively a minimum-flow problem with multiple sinks, where each cell is a sink, as shown in Figure 3; however, our objective is to minimize the flow to each individual sink, rather than the flow to all sinks.

[0113] The best vaccine composition The outcome of the method is then step S105, where the best vaccine composition is selected. This vaccine composition can be protein-based, if the vaccine is made from amino acid sequences similar to proteins displayed on the surface of cancer cells, or DNA or RNA-based, if the vaccine is made from DNA or RNA sequences to induce the production of said proteins in the patient's cells. [Explanation of symbols]

[0114] S101 process S102 process S103 process S104 process S105 process

Claims

1. 1. A computer-implemented method for selecting one or more amino acid sequences for inclusion in a neoantigen vaccine from a set of candidate neoantigen amino acid sequences, comprising: obtaining a set of input data relating to a patient; simulating a plurality of cancer cells based on a set of input data, where simulating each cancer cell comprises predicting a cell surface presentation of said cancer cells; For each candidate neoantigen amino acid sequence, predicting the likelihood of said candidate neoantigen amino acid sequence to elicit an immune response against cancer cells based on the predicted cell surface presentation of each cancer cell; and selecting one or more amino acid sequences for inclusion in the vaccine that maximizes the likelihood of the vaccine inducing an immune response against the cancer cells based on the predicted likelihood of each candidate neoantigen amino acid sequence to induce an immune response against the cancer cells. A computer-implemented method comprising:

2. 2. The computer-implemented method of claim 1, wherein the set of input data comprises one or more of: a representation of the patient's HLA-I alleles; gene expression information; a set of identified genetic variants; a binding affinity index for each tuple of candidate neoantigen amino acid sequences and HLA-I alleles; and a presentation index for each tuple of candidate neoantigen amino acid sequences and HLA-I alleles.

3. 3. The computer-implemented method of claim 2, wherein simulating the plurality of cancer cells comprises predicting the presence or absence of each of the identified genetic variants in each of the plurality of cancer cells based on a statistical distribution of the identified genetic variants.

4. 4. The computer-implemented method of claim 3, wherein simulating the plurality of cancer cells comprises estimating the abundance of one or more proteins synthesized in each cancer cell based on gene expression information and genetic variants predicted to be present in each cancer cell.

5. 5. The computer-implemented method of claim 4, wherein simulating the plurality of cancer cells comprises estimating the abundance of one or more peptides processed in each cancer cell based on the estimated abundance of one or more proteins synthesized in the cancer cell and the likelihood of each of the one or more proteins being divided into one or more peptides.

6. 6. The computer-implemented method of claim 5, wherein simulating the plurality of cancer cells comprises simulating binding of one or more peptides to HLA molecules to estimate the likelihood of one or more peptide-HLA complexes being present in each cancer cell, and simulating binding of peptides to HLA molecules is based on the abundance of the one or more peptides and a binding affinity index for each tuple of candidate neoantigen amino acid sequence and HLA-I allele.

7. 7. The computer-implemented method of claim 6, wherein simulating the plurality of cancer cells comprises predicting cell surface presentation for each cancer cell based on the likelihood of one or more peptide-HLA complexes being present in each cancer cell and a presentation index for each tuple of candidate neoantigen amino acid sequence and HLA-I allele.

8. 8. The computer-implemented method of any one of claims 1 to 7, wherein predicting the likelihood of each candidate neoantigen amino acid sequence to elicit an immune response against cancer cells comprises estimating the likelihood of the patient's immune system comprising T cells having receptors that bind to the predicted cell surface presentation of each cancer cell.

9. 9. The computer-implemented method of any one of claims 1 to 8, wherein selecting one or more amino acid sequences for inclusion in the vaccine comprises applying a mathematical optimization algorithm to minimize the likelihood of the vaccine not eliciting an immune response against cancer cells.

10. The variables of the mathematical optimization algorithm are (a) a binary indicator variable for each candidate neoantigen amino acid sequence indicating whether the candidate amino acid is included in the vaccine; and (b) a continuous variable giving, for each cancer cell, the log-likelihood that there is no immune response elicited by the candidate neoantigen amino acid sequence against said cancer cell; 10. The computer-implemented method of claim 9, comprising:

11. 9. The computer-implemented method of any one of claims 1 to 8, wherein the step of selecting one or more amino acid sequences for inclusion in the vaccine comprises applying a mathematical optimization algorithm to minimize the likelihood of the vaccine not eliciting an immune response relative to cells that have the greatest likelihood of there being no vaccine-elicited immune response.

12. The variables of the mathematical optimization algorithm are (a) a binary indicator variable for each candidate neoantigen amino acid sequence indicating whether the candidate amino acid is included in the vaccine; (b) a continuous variable giving, for each cancer cell, the log-likelihood of the absence of an immune response elicited by the candidate neoantigen amino acid sequence against said cancer cell; (c) a continuous variable for each cancer cell that gives the log-likelihood of the absence of an immune response induced by a vaccine containing a subset of the set of candidate neoantigen amino acid sequences; and (d) A continuous variable that maximizes the log-likelihood that no cancer cells will respond to a vaccine containing a subset of the set of candidate neoantigen amino acid sequences.

12. The computer-implemented method of claim 11, comprising:

13. 13. The computer-implemented method of any one of claims 9 to 12, wherein the mathematical optimization algorithm is integer linear programming.

14. 14. The computer-implemented method of any one of claims 1 to 13, wherein the method further comprises the step of assigning a cost to each candidate neoantigen amino acid sequence, and wherein selecting one or more amino acid sequences for inclusion in the vaccine is constrained based on the cost assigned to each candidate neoantigen amino acid sequence, such that the selected one or more amino acid sequences have a total cost less than a pre-set threshold budget.

15. 1. A method of making a vaccine comprising the steps of: Selecting one or more amino acid sequences for inclusion in a vaccine from a set of candidate neoantigen amino acid sequences by the method of any one of claims 1 to 14; and synthesizing the selected amino acid sequence(s) or encoding the selected amino acid sequence(s) into a corresponding DNA or RNA sequence and / or incorporating the DNA or RNA sequence into the genome of a bacterial or viral delivery system to produce the vaccine. A method comprising:

16. 15. A system for selecting one or more amino acid sequences for inclusion in a vaccine from a set of candidate neoantigen amino acid sequences, the system comprising at least one processor in communication with at least one memory device having instructions stored thereon for causing the at least one processor to perform the method of any one of claims 1 to 14.

17. A computer readable medium having stored thereon computer executable instructions for performing the method of any one of claims 1 to 14.

Images