WO2025050009A2

Movatterモバイル変換

Info

Publication number: WO2025050009A2
Application number: PCT/US2024/044819
Authority: WO
Inventors: Nathan SALOMONIS; Guangyuan Li
Original assignee: Cincinnati Childrens Hospital Medical Center
Current assignee: Cincinnati Childrens Hospital Medical Center
Priority date: 2023-09-01
Filing date: 2024-08-30
Publication date: 2025-03-06
Anticipated expiration: 2026-03-01
Also published as: WO2025050009A3

Abstract

Immunogenic targets associated with one or more types of cancer, wherein the immunogenic target includes a shared splicing amino acid-based neoantigen present in any of SEQ ID NOs: 1-2545, or a nucleotide coding therefor, are described, as well as uses thereof, and methods of identifying such immunogenic targets. Particular embodiments relate to immunotherapies for treating a type of cancer, and methods of treating a type of cancer by administering such immunotherapies, wherein the immunotherapy includes an immunogenic target including a shared splicing amino acid-based neoantigen present in any of SEQ ID NOs: 1-2545, or a nucleotide coding therefor.

Description

[00372] These datasets and conditions. The default cutoffs were selected in accordance with those previously recommended from a pan- cancer splicing neoantigen analysis. If a user supplies matching controls for one or more tumor samples, the matched control will be automatically concatenated to the existing GTEx database to calculate the integrated average read count. The filtered tumor-specific splice junction is referred to as a neojunction. [00373] For each neojunction, the junction RNA sequence is first retrieved by concatenating the 5′ and 3′ exon/intron sequences, followed by in silico translation into possible k-mers, considering all the three possible reading frames (Figure 2). A valid short peptide should meet two criteria: (1) absence of a stop codon, and (2) at least one amino acid should span the splice-junction site, such that the junction peptide will not derive from only the 5′ donor or the 3′ acceptor exon/intron. All possible strand-specific reading frames are used to account for both potential tumor-specific alternative protein translations and to increase the sensitivity to detect novel neoantigens. [00374] SNAF next restricts the number of T cell neoantigen candidates by considering MHC binding prediction and T cell immunogenicity. Only a few MHC-bound peptides are likely to elicit T cell response (Figure 2). Specifically, SNAF provides interfaces to two popular MHC binding prediction models – NetMHCpan4.1 and MHCflurry2.0 to prioritize junction-peptides that are likely to bind with patient-specific HLA sequences. NetMHCpan4.1 and MHCflurry2.0 have the highest reported accuracy for MHC binding based on recent benchmarking. MHCflurry is directly embedded with SNAF, while NetMHCpan4.1 is available for direct integration, following authorization of the software’s license from the software’s website. [00375] To produce the most accurate immunogenicity predictions, SNAF leverages the inventors’ recently published deep convolutional neural network approach DeepImmuno, which ranked among the best-performing models in a recent independent benchmarking study. The resulting predicted neoantigens are expected to have a high probability of MHC binding and immunogenicity (Figure 2). Prediction of ExNeoEpitopes in SNAF-B (Step 3) [00376] Extracellular epitopes (B cell antigens) identified by SNAF use the same initial neoantigen discovery workflows described above (Steps 1, 2), with the exclusion of MHC peptide binding and immunogenicity prediction SNAF module calls. To further predict full- length protein coding ExNeoEpitopes, SNAF-B can limit neojunctions to only those which encode for transmembrane spanning protein coding genes. [00377] A heuristic approach is used to infer full-length isoform from all existing transcripts and local splicing events. For example, if a tumor-specific splicing event E2.1-E3.1 is observed, along with a documented transcript whose exon composition is E1.1|E2.1|E2.2|E3.1|E4.1, the potential full-length isoform resulting from this splice-event will be E1.1|E2.1|E3.1|E4.1 (E2.2 skipping) (Figure 2). [00378] This workflow is biased to detect exon-skipping or alternative splice-sites that result in shorter peptides, but not isoforms with longer protein coding sequences (novel exons and splice-sites). To account for this class of insertion peptides, SNAF implements a parallel strategy leveraging input cancer long-read sequencing data from a gtf file. Specifically, for a specific alternative splicing event associated junction (i.e. E2.1-E3.1), SNAF identifies an associated full-length isoform by matching the exon-exon junction splice-site coordinates to those present in the long-read isoform (presumably tumor-associated) GTF file. For intron retention, SNAF determines whether both intron boundaries fall within a long-read supported single-exon region. Such matching long-read evidenced isoforms represent additional candidate ExNeoEpitopes. [00379] SNAF-B reference-augmented workflow (de novo prediction): When no matching mRNA isoform contains the neojunction, the most appropriate reference isoform is first selected for a given neojunction. This analysis (short_read mode), produces a theoretical mRNA isoform based on the software selected reference isoform. To accomplish this, SNAF-B leverages a unique feature of the software AltAnalyze, which is the explicit annotation of exon and intron blocks and regions. Specifically, all AltAnalyze documented isoforms are decomposed into program annotated exon blocks and regions, where each exon block is subdivided into regions defined by alternative splice site locations, using a series of prior described rules. When RNA-Seq data is analyzed, all exon-exon and exon-intron junctions are annotated according to this schema, with novel exons or splice sites assigned unique IDs relative to the block and region they are situated within. In SNAF, if the neojunction represents the excision of an exon block or specific exon region, the software identifies the the longest protein- coding isoform from AltAnalyze’s reference isoform collection that contains both exon regions that comprise the junction. The collection consolidates mRNA and protein isoforms from Ensembl the UCSC genome browser database (Ensembl version 91 and matched UCSC database for this paper). For example, the neojunction ENSG00000056291:E2.2-E5.1 represents an exon skipping isoform. The corresponding longest protein coding isoform that contains these two exons is ENST00000395999 (ENSP00000379321), which is represented as E2.1|E2.2|E3.1|E4.1|E5.1|E7.1|E7.2|E7.3. SNAF represents the ExNeoEpitope from this reference form as E2.1|E2.2|E5.1|E7.1|E7.2|E7.3, by excluding the skipped exon block and regions. If the isoform is an inclusion form, occurring due to the presence of a novel splice-site, exon or intron-retention event, this function lacks the specificity to infer the full-length isoform and neojunction prediction is skipped. [00380] SNAF-B Long-Read Alignment workflow: As an alternative approach to directly identify novel mRNA isoforms associated with short-read bulk RNA-Seq-derived neojunctions from SNAF, the SNAF-B pipeline includes specialized methods to identify novel mRNA isoforms that can encode for full-length transmembrane proteins. This analysis (long_read mode), produces an inferred protein isoform based on a long-read sequence match. This workflow leverages an externally provided mRNA isoform gene transfer format (GTF) text files, containing the genomic coordinates corresponding to the isoform exon positions for the genome database being analyzed (e.g., hg38). The SNAF-B predicted neojunction genomic coordinates are matched to the provided GTF to identify the longest cDNA junction match. These can include bulk or single-cell long-read sequencing predictions or ESTs from one or more studies/repositories. In this study, two GTF long-read isoform files from a prior cancer cell- line compendium dataset produced using Iso-Seq (PacBio Sequel) were used, along with a separate internal melanoma cell line Iso-Seq dataset, detailed below. [00381] To assess the translational potential of these candidate ExNeoEpitopes, an efficient Open Reading Frame (ORF) finding algorithm was implemented based on three criteria: (1) ORF length, (2) GC content, and (3) codon usage bias. ORFs of higher length are more likely to be the most valid translation. When the length of two alternative ORF predictions are less than 8nt, the ORF with higher GC content and higher codon usage bias is prioritized. As guanine-cytosine (GC) pairs have three hydrogen bonds compared to two between adenine- thymine (AT) pairs, higher GC content is an indication of stable transcripts in most gene-finding prediction algorithms. [00382] Codon usage bias measures the species-specific tendency to prefer one codon over the other, which can be used to prioritize valid protein coding genes. The codon frequency table is obtained from https <colon slash slash> www <dot> genscript <dot> com <slash> tools <slash> codon-frequency-table. The cumulative frequency value across the ORF (each triplet will have a codon frequency value) is normalized to a value between 0-1 and added to the GC content percentage as the score for ORF prioritization. The predicted ORF is assessed for Nonsense Mediated Decay (NMD) potential (Figure 2). [00383] To assess NMD, SNAF measures the distance between the end position of the ORF and the last nucleotide of the transcript and removes any ORF predictions that result in a premature stop code more than one exon away from the terminal exon in that transcript (n_stride=2), where n_stride is an adjustable parameter. The NMD check is performed for all inferred short-read predicted inferred isoforms, as a reference isoform with an established ORF, but not for long-read isoforms, as the ORF must be inferred. Any predicted isoforms that have a 100% match with a documented membrane protein isoform in the Uniprot database are excluded, as these are presumed to be expressed in non-malignant tissues. [00384] To increase the accuracy of these predictions, the resultant novel proteins are subjected to TMHMM (Figure 2), a Hidden Markov Model (HMM)-based method to predict the topology of the membrane proteins. Proteins without predicted transmembrane domains will be eliminated for further analysis. As SNAF is modular, these can be optionally skipped by the user to identify additional candidates. The user can optionally supply long-read sequencing data (GTF) to the program to validate predicted full-length isoforms derived from junctions alone (Figure 2). The resultant novel membrane isoform will be reported as high-confidence B cell neoepitopes. [00385] To determine whether a predicted novel protein will result specifically in extracellular impacts or alter transmembrane domain composition, the individual TMHMM predictions can be explored and compared to different possible reference protein isoform sequences from the SNAF interactive viewer (external linkout) on a case-by-case basis. In this analysis, a secondary manual inspection of the included or deleted polypeptide was performed in UniProt to determine which topological domain and transmembrane features they localize with, relative to the reference protein curation for that gene. [00386] To further identify candidate ExNeoEpitopes in these analyses, peptides that were contained within existing isoform protein sequences from Ensembl or UCSC that represent a 100% string match (AltAnalyze build database) were eliminated. These principally comprise novel alternative promoter exons that are not predicted to encode for protein sequences. Alignment Software and Reference Evaluation [00387] The previously described GDC-TCGA workflow was replicated for sequence alignment for the alignment of control samples to GENCODE version 36. To confirm the validity of these predictions, two TCGA breast cancer samples, FASTQ files (bedtools) were converted using the TCGA workflow and alternative settings or genome reference versions, followed by processing in AltAnalyze. Benchmark of MultiPath-PSI on Splicing Detection [00388] Benchmarking of MultiPath-PSI was performed against several recently reported methods for local splicing variation analyses (MAJIQ (version 1.0), LeafCutter (version 0.2.1), and rMATS (version 3.0.9)). As a gold-standard to evaluate prediction accuracy and error rates, a simulated RNA-Seq dataset derived from known and novel isoform predictions was created for pluripotent stem cells (PSC) and in vitro derived day 30 cardiomyocytes (CM) (https://www.synapse.org/#!Synapse:syn12104338). [00389] To create this simulation dataset, known and novel isoforms and expression estimates were predicted using Cufflinks on biological triplicate samples (PSC and CM). The Cufflinks isoform GTF and expression values (isoforms.fpkm_tracking) were supplied as input for the software Polyester to produce simulated RNA-Seq reads at a depth of 100 million paired end reads using the software default parameters. These subsequent FASTA files were supplied as input for genome alignment (hg19) with the software STAR to produce BAM files with aligned, stranded, paired end reads. The resulting aligned sequencing data were evaluated with each analysis method and benchmarked against isoform-derived alternative splicing events (exon- exon and exon-intron junctions) comparing CM to PSC. Supervised group comparison analyses with MultiPath-PSI were performed using an FDR corrected (Benjamini-Hocherg) empirical Bayes moderated t-test P-value < 0.05. [00390] To evaluate the performance of the different algorithms in reporting the differential splicing events, precision-recall curves were generated using Matlab for the synthetic dataset. For each method, the significant splicing events were ranked in descending order according to the empirical score or p-value reported by each algorithm. Each algorithm reported only a portion of the gold-standard splicing events as significant. To generate precision-recall curves that extend to recall=1, the non-reported gold-standard splicing events were included additionally with a p-value=0 or lowest empirical score. This inclusion leads to the sudden peak in precision with recall identified in the tail of the precision-recall curves for each tool. Comparison of unique splice junction clusters produced by each algorithm (determined from the junction-graph of junction genomic coordinates output by each method) to these gold-standards indicated that MultiPath-PSI has the greatest overall accuracy and lowest error rates, based on precision and recall estimates (Figure 4). Bulk and Single-Cell Melanoma Splicing Evaluation Datasets [00391] To evaluate shared and unique splicing neoantigens identified by SNAF, prior reported bulk and single-cell RNA-Seq datasets were reanalyzed through SNAF, using the same genome alignment and splicing quantification workflows applied to TCGA samples. To assess the association of melanoma splicing neoantigens with non-cancerous proliferative skin cell splicing events, five proliferative melanocyte RNA-Seq datasets were reanalyzed in the GEO database (GSE102983, GSE111786, GSE149189, GSE197471, GSE202700) (Figure 17). [00392] To determine the cell of origin for melanoma splicing antigens, the analysis obtained access to the controlled access raw sequencing data (DUOS-000002), for 4,645 individual cell transcriptomes corresponding to 19 patients with melanoma (GSE72056). Only 3,877 with cell annotations were retained for further analysis. These individual cell-level FASTQ files were re-analyzed in STAR and AltAnalyze to produce aggregate junction read counts for each patient and author annotated cell-populations. These junction read counts were summed per cell-population to identify tumor versus immune neojunction enrichments (fold>2 enriched). These analyses are biased towards immune cells, as twice as many immune cells (n=2,605) versus tumor (n=1,174) were present. Calculating Tumor Specificity [00393] SNAF can calculate two optional continuous scores (maximum likelihood estimation and Bayesian hierarchical), measuring the tumor specificity of each splice junction based on its expression across diverse tissues. These methods are designed to rank and prioritize SNAF neoantigens for experimental validation, using different normal tissue references. These scores can be optionally used to filter results by the user (not by default). Mathematically, tumor specificity for a neojunction (nj) is defined as given its raw read counts as a vector where indicates the total

of samples in GTEx and TCGA paratumor samples (in the present case, =3,644). Since GTEx contains over 50 different body site regions, the collection of normal tissues was defined as , so that tissue-specific raw read counts can be expressed

[00394] Maximum Likelihood Estimation Approach): To account for

the sequencing depth bias of different GTEx samples, first the normalized raw read counts are computed, as follows in Equation 3:

[00395]

Per Million (CPM) in single-cell sequencing, as there is no need to account for length bias for each splice junction. Next, the following is assumed as in Equation 4: [00396] The

can be formalized as in Equation 5:

(5) [00397]

Likelihood Estimation (MLE) according to Equations 6-

- ) (6)

the Python scipy.stats.halfnorm.logpdf function. [00399] Bayesian Hierarchical Approach: Although the MLE approach takes into account the whole expression vector across GTEx samples, the distribution of splicing expressions across tissue types was not explicitly modeled. The present study therefore implemented a hierarchical Bayesian model to consider the different distributions of the splicing expression values, which has been further developed into an unified tumor specificity score model called BayesTS for all types of molecular tumor targets including protein-coding genes. Briefly, this method assigns a higher tumor specificity score to predicted neojunctions with infrequent versus frequent expression in healthy controls (GTEx) across all tissues and donors. While an ideal neojunction will never be detected in healthy tissues, low-level and infrequent expression in a subset of donor tissues should not disqualify a potential neoantigen, while frequent low-level neojunction expression represents an unacceptable risk of off-target effects. [00400] Herein, a random variable has been defined, representing the number of samples with non-zero read counts for a

( ) in each tissue ( ). For example, if there are 5 out of 25 hippocampus samples that have non-zero read counts, then . This assumes:

[00401] The rate parameter , is

by the underlying tumor specificity score σ and coefficient β^c_nj to account for the magnitude difference between σ and observed count C: λ nj = β^c nj σ (9) [00402] where β^c follows a

(25,1) with the intuition that a NeoJunction present at a medium level should have around 12.5 samples expressing it within each tissue. Leveraging the advantages of Bayesian approach to incorporate multiple types of observations, this analysis assumes the underlying σ also gives rise to the normalized count data as defined in the MLE section above, such that: where β^x follows another

(10,1) to account for the magnitude difference. Log Norm distribution was chosen due to its exponentially increased space, which shows better convergence in bayesian inference than half norm distribution. A weakly informative prior Beta(2,2) was chosen for the underlying σ. And finally,

[00403] Using the model, this again assumes follows the Harf Norm distribution:

[00404] This erence solver implemented in Python pymc3 package to efficiently estimate the and the confidence interval. The mean value of the posterior distribution of serves as tumor specificity score.

[00405] In this study,

from both scoring methods are provided; however, to be inclusive, the analysis has not restricted neoantigens using either tumor specificity exclusion method. Each tumor specificity measurement has its own strength and weakness. While the MLE approach is fast and can distinguish splice-junctions that are frequently expressed in normal tissue versus being tumor specific, it tends to report zero for junctions that are extremely rare in normal tissues (i.e., splice junctions with an average read count < 1.5 - empirically observed). This is partially due to the fact that MLE does not take into account the distribution of reads across all tissues. Alternatively, the Bayesian approach can identify difficult-to-identify non- tumor-specific junctions, such as in the context of lowly expressed genes, by considering the distribution of junction expression across tissue types, but results in a significantly longer runtime. While still an ongoing area of evaluation, these methods can be used in combination when desiring the highest confidence tumor-specific junctions (i.e., initial MLE and Bayesian on remaining junctions). To verify tumor specificity predictions, IGV visualization was performed on an independent deeply sequenced RNA-Seq dataset of healthy control tissue (ArrayExpress: E-MTAB-2836). RBP Activity Inference [00406] For both training and testing of RBP activity inference methods, extensive prior RBP knockdown, CRISPR and eCLIP in vitro profiling datasets were downloaded and reanalyzed. To produce the prior network for evidenced RBP-junction interactions, K562 cell line knockdown and CRISPR RNA-Seq data (pair-end) were downloaded from the ENCODE website (https://www.encodeproject.org/). With a few exceptions, all RBPs analyzed had two replicates present and two matching controls. [00407] In total, 484 RNA-Seq samples were downloaded spanning 191 RBPs. These FASTQ files were aligned to human genome version hg38 using STAR version 2.6.1, followed by AltAnalyze 2.1.4.2 to obtain splicing junctions from the resultant BAM files. As batch effects in these data were previously reported and corrected for, these corrections were reproduced using limma on the splice junction count results prior to computing the PSI matrix in AltAnalyze. Only RBP perturbation alternative splicing events that were predicted using both the non-batch effects corrected data (comparing the two replicates and matched controls) and the batch effects corrected data (versus all 194 combined controls), were retained to ensure specificity of the predictions. K562 eCLIP bed files were downloaded from the ENCODE website (https://www.encodeproject.org/) spanning over 120 RBPs. Only high-confidence eCLIP peaks were retained from technical replicate samples, using Irreproducible Discovery Rate (IDR) analysis (idr 2.0.4 software) using an FDR threshold of 0.1. [00408] From the resultant IDR peaks, RBPs that have direct regulatory evidence for an alternative splicing event were defined, where >1 IDR peak is proximal (exon/intron plus the intermediate downstream and upstream flanking regions). From these predictions, a K562 prior splicing regulatory network was produced using rule-based criteria. Specifically, edge weight = 1 if the RBP is evidenced by both eCLIP and perturbation for a given splice junction, 0.5 if only one evidence modality is present and 0 if neither. [00409] To assess the performance of commonly used TF activity inference algorithms, the K562 prior network was used. Combined with the context-specific PSI splicing matrix (AltAnalyze), activity inference algorithms will compute the activity of each RBP in each queried sample. Ten commonly used TF activity inference algorithms implemented in decouplerPy package were tested: AUCell, GSEA, GSVA, Multivariate Decision Tree (MDT), Multivariate Linear Model (MLM), Over Representation Analysis (ORA), Univariate Linear Model (ULM), VIPER, Weighted Sum (Wsum) and Weighted Mean (Wmean). GSEA analysis was run using either the “estimate” (GSEA) or the normalized GSEA score (GSEA_norm). GSVA analysis was run using mx_diff = False or True, the former setting will penalize the events with opposite predictions than expected (increased versus decreased regulation), whereas the latter will erase the directionality, resulting in a total of 12 tested methods. The accuracy of inferred RBP activity was evaluated in ENCODE HepG2 cell line RNA-Seq knockdown data, spanning 193 RBPs across 474 samples. Splicing events with less than 75% non-zero PSI values were filtered out and the missing PSI values were imputed by the median of present values per event. Finally, splicing events with Coefficient of Variation (CV) less than 0.1 were excluded (uninformative). [00410] An accurate inference should reflect the reduced activity of each RBP in the knockdown samples compared to controls, based on the K562 prior network. This is analogous to the multi-label ranking problem and hence, Label Ranking Average Precision (LRAP) and Normalized Discounted Cumulative Gain (NDCG) can be used to quantify performance. For each RBP, the ground-state truth label in the knockdown sample should equal 0, whereas the controls should equal 1. The total number of knockdown and control samples was denoted as S, the ground state truth label as a binary vector X_S, and the predicted RBP activity as a vector Y_S where the higher value indicates higher RBP activity. The metrics are defined as below: .

identify samples whose predicted RBP activity is higher than itself. A higher LRAP means most of the control samples receive a higher RBP activity score and vice versa. [00412] NDCG is originally used to assess the usefulness of a web research result, by determining the fraction of top results which are relevant to the query, times a decay factor based on the rank of the results. First, all samples are ranked based on the predicted RBP activity in descending order, so the highest sample will receive the rank 1. The Discounted Cumulative Gain (DCG) is computed as below: (16) [00413] NDCG

the perfect DCG score it can get, such that NDCG is ranging from 0 to 1 where as a higher values indicate a higher accuracy. MS Validation of Ovarian and Melanoma Tumor Biopsies [00414] Immunopeptidome Mass Spectrum sequencing data from 14 ovarian tumor biopsies was obtained from ftp <colon slash slash> ftp <dot> ebi <dot> ebi <dot> ac <dot> uk <slash> pride-archive <slash> 2017 <slash> 11 <slash> PXD007635. Matched RNA-Seq data was downloaded from the Sequence Read Archive project PRJNA398141. This single-end RNA- Seq data was processed in AltAnalyze as described above (Step 1). To increase sensitivity for this single-end data, SNAF was run on all quantified exon-exon and exon-intron junctions, rather than additionally filtering on differential splicing events from MultiPath-PSI. Patient HLA type was provided previously. [00415] The MS profiles were analyzed using the software MaxQuant 2.0.3.1 (https <colon slash slash> www <dot> maxquant <dot> org), run on a Linux High Performance Compute (HPC) environment using mono version 5.20.1. A customized MHC-bound splicing neoantigen database was built using snaf.proteomics.remove_redundant and snaf.proteomics.compare_two_fasta function. The snaf.proteomics.compare_two_fasta function removes all redundant short peptides from the prediction, whereas the snaf.proteomics.remove_redundant function discards splicing neoantigens overlapping with non- diseased human proteome sequences (UniProt). The custom splicing neoantigen database (FASTA files) used for Peptide Spectrum Matching (PSM) search is available in synapse (https <colon slash slash> www <dot> synapse <dot> org <slash> #!synapse:syn32057176), and the MaxQuant configuration files mqpal.xml and the code SNAF uses to automatically generate the xml files are available in GitHub (https <colon slash slash> github <dot> com <slash> frankligy <SNAF> reproduce). [00416] The configuration file was created using the MaxQuant Windows Desktop application and modified using snaf.proteomics.set_maxquant_configuration function. Specifically, MaxQuant was run using the custom splicing neoantigen database as described above. Key parameters are as following: enzyme=None, enzyme_mode=5 (no digestion), protein_fdr=1, peptide_fdr=0.05, site_fdr=1, minPepLen=8, minPeptideLengthForUnspecificSearch=8, maxPeptideLengthForUnspecificSearch=25. Other parameters are the default value automatically set by the program. [00417] Raw MS sequencing data from 25 independent melanoma patients was obtained from ftp <dot> ebi <dot> ac <dot> uk <slash> pride-archive <slash> 2017 <slash> 04 <slash> PXD004894 and analyzed using MaxQuant as described above. Patient 16 (Mel–16) was excluded due to potentially corrupted raw data. [00418] The custom splicing neoantigen database (FASTA files) used for Peptide Spectrum Matching (PSM) search is available in synapse (https <colon slash slash> www <dot> synapse <dot> org <slash> #!Synapse:syn32057176). In brief, all observed shared splicing neoantigens (n=114) and unique (patient-specific) neoantigens were combined into a single combined database, to ensure these two sets had an equal chance to be detected. Here, neoantigens which correspond to peptides in UniProt (non-diseased) were eliminated, and only non-redundant peptides (from different junctions) were reported. Peptide Synthesis and MS Spike-In Aalidation [00419] The 36 peptide candidates were synthesized (GenScript, Piscataway, NJ) at minimum 70% purity with an average yield of 0.2-0.5 mg. Peptides were reconstituted with water to a final stock concentration of 1 pmol/µL. Peptides were pooled (except for LELLVKGTV and STLEFGLRV, which did not solubilize sufficiently) at a concentration of 1 pmol/µL and then diluted 1:10 for a 100 fmol/µL working solution. LC-MS analysis was performed on a 50 fmol injection of pooled peptides using a Ultimate 3000 nanoflow HPLC (Dionex) and Orbitrap Eclipse Tribrid mass spectrometer (ThermoFisher Scientific) as described below. [00420] Injections were loaded onto an Acclaim PepMap 100 trap column (300 µm x 5 mm x 5 µm C18) and gradient-eluted from an Acclaim PepMap 100 analytical column (75 µm x 25 cm, 3 µm C18) equilibrated in 96% solvent A (0.1% formic acid in water) and 4% solvent B (80% acetonitrile in 0.1% formic acid). The peptides were eluted at 300 nL/min using the following gradient: 4% B from 0-5 min, 4 to 10% B from 5-10 min, 10-35% B from 10-60 min, 35-55% B from 60-70 min, 55-90% B from 70-71 min, 90% B from 71-73 min, 90-4%B from 73-74 min and 4% B from 74-90 min. [00421] The Orbitrap Eclipse was operated in positive ion mode with 2.0 kV at the spray source, RF lens at 30% and data dependent MS/MS acquisition with XCalibur version 4.3.73.11. Positive ion Full MS scans were acquired in the Orbitrap from 375-1500 m/z with 120,000 resolution. Data dependent selection of precursor ions was performed in Cycle Time mode, with 3 seconds in between Master Scans, using an intensity threshold of 2 x 104 ion counts and applying dynamic exclusion (n=1 scans within 30 seconds for an exclusion duration of 60 seconds and +/- 10 ppm mass tolerance). Monoisotopic peak determination was applied and charge states 2-6 were included for HCD MS2 scans (quadrupole isolation mode; 1.6 m/z isolation window, Normalized collision energy at 30%). The resulting fragments were detected in the Orbitrap at 15,000 resolution with Standard AGC target and Dynamic maximum injection time mode. [00422] The MS2 raw file was imported into Skyline v.22.2.0.351 and a template for synthetic peptide sequences was applied to visualize MS2 spectra. The ion intensities for MS2 b- and y- fragment ions were plotted for each synthetic peptide sequence. Peptides with coverage for at least six fragment ions were considered “observed”. The Skyline software enables visualization of the observed fragment ions in the context of the entire raw MS2 spectrum via color-coding and labeling of the b- and –y ions. The synthetic peptide MS2 spectra were then compared to experimental spectra which were identified by search algorithm software to have the same peptide sequence. The b- and y- ions identified from both the synthetic and experimental spectrums were selected, and the quantitative correspondence were calculated based on their associated ion densities. Particularly, denoting each fragment ions in synthetic spectrum as x and the corresponding ions in experimental spectrum as y, with the total number of I, and the associated relative sundance is dxi and dyi . Pearson correlation (r) and the cosine similarity (cos) was computed as below. where

The p-value was reported as the probability of observing a higher correlation of randomly drawn X and Y from two independent normal distribution (zero correlation). The analysis used r > 0.70, p <= 0.05 and cos > 0.9 as a rough threshold instead of a hard criterion to evaluate the similarity of observed spectrums. [00423] Amongst all the MS2 spectrums that were assigned to a splicing neoantigen by the program (FDR < 0.05), the matched ions and the associated ion density were manually inspectws. A series of rules were applied to rule out potential false positive match and only retain high-confidence peptides for MS spike-in experimental validation. Criteria: [1] At least 6 matched fragment ions; [2] No more than 25% unassigned ions are greater than 10% relative abundance; [3] Fragment ions make up to at least 20% of total ions in the spectrum; [4] Presence of sequential fragment ions (i.e. y4-y5-y6); and [5] y-ions ending in proline were higher density than other ions. Peptide Spectrum matchings (PSM) that meet all the of the above were retained as high- confident peptides for further experimental validation using synthetic sequence. Validation of Peptide-MHC Binding by MHC Stabilization Assay [00424] To test MHC-I binding of synthesized neoantigen peptides, TAP deficient T2 cells that are defective in transporters were used for endogenous peptide loading. T2 cells were obtained from ATCC and grown at 37°C, 5% CO2 in Iscove's Modified Dulbecco's Medium supplemented with 20% FBS and pen/strep. 1x105 T2 cells were used without and loaded with 100ug/ml peptides and thereafter incubated overnight. All T2 cells were harvested and stained with HLA-A2-PE antibody (clone BB7.2; Biolegend) for 30 mins on ice. Cells were washed once with cell culture medium and acquired on a Fortessa II flow cytometer. Median fluorescence intensity was determined using FlowJo software. Immunogenicity Assay [00425] Immunogenicity of predicted splicing neoantigen peptides was determined as described previously. In short, HLA-typed PBMCs from leukocyte reduction system (LRS) chambers were isolated using Ficoll hypaque density gradient centrifugation, aliquoted in 20x10⁶ cells per vial and frozen in liquid nitrogen until use. At day 0, PBMC were thawed and used to set up monocyte derived dendritic cells by plating 4x10⁶ PBMC in a 24 well. Cells were incubated at 37°C, 5% CO2 in DC medium (RPMI 1640 supplemented with 10% FBS, 1% L- Glutamine (200mM) + IL-4 (1000 U/ml) and GM-CSF (800U/ml). After 4h, the non-adherent fraction was removed by rinsing the wells twice with PBS. Adherent cells were cultured for 7 days in DC medium. On day 7, DCs were loaded with 10ug/ml peptides dissolved in DC medium and incubated at 37°C, 5% CO2. After 4h, 1.5ml DC maturation medium was added (RPMI 1640 supplemented with 10% FBS, 1% L-Glutamine, IL-4 (1000U/ml), GM-CSF (800U/ml), IL-1β (10ng/ml), IL-6 (10ng/ml), TNF-α (10ng/ml) and LPS (30ng/ml)). After 16h of DC maturation, peptide loaded DCs were used to stimulate autologous PBMC, by adding 1x10⁶ PBMC of the same donor to the DC cultures. DC and PBMC co-cultures were grown in T cell medium (60% RPMI 1640, 40% Click’s medium supplemented with 10% FBS, 1% L-glutamine, IL-6 (100ng/ml), IL-7 (10ng/ml), IL-12 (10ng/ml), and IL-15 (5ng/ml). Medium was changed on day 3 and day 6 based on medium color change. On day 14 and 21 T cells were harvested and stimulated with new autologous peptide loaded DCs. [00426] After three rounds of T cell priming, at day 28, T cells were harvested and tested for their IFNγ response to peptide loaded 721.221 (221) single HLA antigen target cells. Fist the target cells 221.A*02:01, 221.C*04:01, and 221.C*08:01 were loaded with relevant peptides at 10µg/ml in separate wells.221 and peptides were incubated for 4h at 37°C, 5% CO2 in RPMI 1640 supplemented with 10% FBS. Peptide loaded 221 target cells with and without peptides were co-cultured with primed T cells at 3:1 ratio in the presence of 50 ng/ml PMA for 6h at 37°C, 5% CO2 in RPMI 1640 supplemented with 10% FBS and monensin. PMA and ionomycin stimulation (each at 1µg/ml) was used at positive control. Thereafter the cells were harvested and stained for extracellular CD45-BV786 (Biolegend), CD14-PerCP (Biolegend) and CD8-PE (biolegend) for 30 mins on ice. Cells were fixed and permeabilized using the CytoFix/CytoPerm kit (BD) according to manufacturer instructions and thereafter stained for intracellular IFNγ-APC expression (clone 4S.B3; Biolegend) for 20 mins on ice and directly analyzed on a BD Fortessa flow cytometer. Analysis of CD45+CD14-CD8+ IFNγ+ cells was determined using FlowJo software. Validation of ExNeoEpitope Localization [00427] To confirm the expression and cell membrane localization of SNAF-B novel isoforms, the long-read sequencing evidenced alternative isoforms and their annotated reference isoforms that are C-terminal tagged with either mNeon-Green or eGFP on a plasmid vector (VectorBuilder, USA). Streak LB agar plates with 100 µg/mL Ampicillin were made for each isoform. A single colony was picked from each plate and expanded in 1 mL of LB broth for 8 hours at 37 Celsius respectively. 20 µL of the pre-expansion broth was then taken and pipetted into a Elenmyer flask with 50 mL of LB broth. The competent cells were expanded overnight at 37 Celsius. Medi-preps were performed for each construct with the ZymoPure II plasmid midiprep kit. On an Ibidi 4-well chamber µ-slide, HEK-293T cells were seeded in prior with a concentration of 0.15 M/mL. When HEK-293T cells reached 60% of confluency, these constructs (CMV promoter) were transfected into the cells separately with 1 μg of plasmid (99% pUC19 negative control plasmid+1% engineered plasmid) with TransIT-LT1 (Mirus) following the manufacturer's protocol. [00428] The cells were fixed and permeabilized with 4% PFA 24 hours post- transfection. After two rounds of washing, the cells were treated with a warm 1x citrate buffer (diluted from 10X stock; Sigma-Aldrich) to break the protein cross-links. The cells were then washed once with PBS again and a membrane actin stain was performed with 1x Phalloidlin 647 in PBS with 1% BSA (Abcam) at room temperature for an hour. To stain the nuclei of the fixed cells, the cells were washed with PBS and stained with DAPI (Thermo Scientific) (1:4000 diluted in PBS with 1% BSA) at room temperature for 5 minutes. The stained cells were immediately washed with PBS, and the PBS removed by vacuum. Prolong gold mounting media (Thermo Fisher Scientific) was evenly applied to the fixed cells surface. 24 hours post transfection, the expression and co-localization of the reference and alternative isoforms were assessed by a confocal spinning disk microscopy (Yokogawa SoRa W1 dual camera system) using the Nikon Elements software. The confocal raw imaging files with all z-stacks are provided in Synapse (https://www.synapse.org/#!Synapse:syn52063953). Melanoma Cell-Line Long-Read Isoform Sequencing [00429] For long-read mRNA isoform sequencing, cells were grown and isolated from five independent Melanoma cell lines: A375, SKMEL, MeWo, UACC62, and UACC257 (ATCC). Total RNA was isolated (Trizol) and analyzed on a Thermo Nanodrop UV-Vis and an Agilent Bioanalyzer to confirm the nominal concentration and ensure RNA integrity. From the RNA, cDNA was synthesized using the Clontech SMARTr cDNA Synthesis Kit, in which a barcode was added to the oligo-dT at the 3′ end. Each melanoma cell line cDNA was pooled and then converted into a SMRTbell library using the Iso-Seq Express Kit SMRT Bell Express Template prep kit 2.0 (Pacific Biosciences). Each library was sequenced on a SMRT cell on the Sequel II system using a 30 hour movie collection time. The “ccs” command from the PacBio SMRTLink suite (SMRTLink version 9) was used to convert Raw reads into Circular Consensus Sequence (CCS) reads. The resulting data was analyzed in SQUANTI to assign reads to full- length collapsed reference or novel isoforms. The isoform GTF files, barcode sequences and raw data are available in Synapse (https <colon slash slash> www <dot> synapse <dot> org <slash> #!Synapse:syn32057176). Results are shown in Table 1. N N N N N V^V V N T_T V^V N N N N N N V N N T^{T T T} V_T V^V V T^V T^V V^T V^T^T_{T T}_T^{T T}_T^T^T_T^{T T} T^{T T}_T^T T_T^T T_T_T^T_T T^T_{T T}^T T^{T T} T^{T T}^T_{T T}_T^{T T}^T_{T T}^T_T_T^T_{T T}^{T T} T^T_T^T_{T T}_T^{T T} T_T^T^{T T}_T^T T_T^T T_T_T^{T T T} T^T_T^T_{T T}^T T^{T T} T^{T T}^T_{T T}^T T^{T T T}_{T T T}^{T T} T^T_T^T_{T T} T^T_{T T}_T^{T T}_T^T^T_T^{T T} T_T T^T_T^T T_T^T T_T_T^T_T^T_{T T}^T T^{T T} T^{T T}^T_{T T}^T T^{T T}^T_{T T}^T_T_T^T_{T T}^{T T}^{T T}_T^T_T^T_{T T}_{T T}^T^{T T}_T^T T_T^T T_T_T^{T T}^T_{T T}^T_T T_T^T_{T T}^T T^{T T} T^{T T}_T T_T^T_T^{T T} T_T^{T T} T^T_T^T_{T T} T^T_{T T}_T^{T T}_T^T^T_T^{T T}_T T^T_T^T^T_{T T} T_{T T}^T_{T T}^T T_T_T^{T T} T^{T T}^T_{T T T} T_T^T T^{T T}_{T T}^{T T}^T_{T T T}^{T T T}_T T_T^T_T^T_{T T} T_T^T T_T^T T^T T_T^{T T}^T_{T T}^T_T_T^T_{T T} T_{T T}^T T^{T T T T}^T_T^T T_T^{T T} T^T_{T T} T_{T T}_T^T^{T T}^T_{T T}_T^{T T}_T^T T^T_T^{T T}_T^T_T^T T^T_{T T} T^T T_T^T_{T T}^T T_T_T^{T T T T}^T_T^T T_T^{T T} T^T_{T T} T_{T T} T_T_T^{T T} T_T^{T T T}_T^T_T^T T_T^T T^T T G_T^{G A}_T T_{C T}^T A_{C C}^T T G^{G G} A A^A T_{C T} A_C A G^{T T} A^C A_T G_T G T G C A_T G A_{C C} T G^A T A^T C^{A G} C_T^C G^A T_C G^A T_T^{A A} C^{A G G A}_{T C}^A^A_C A^G_{T T C} G^{A A} A^C T^A_T A G_{C T} G A^{G A} T_{C T} G A^G_C^A C A A A C C G_C G G^C G^A C_C^G C A G^G T G A T_T A A^C_T^C T^T C A G^T A^A^{T G}_T A^G T G^G C_T^A^C G G^{T C}_C A_{C T} A C T^{G C} A_C G_{C T} G^G G_T G^A_{C T} A G^G T_C A^{C A}_C^C_T^{A G} T^G C G A C^C^{T G}_T A_T A A T_T A A G_C G A A_{T T}_{T T T T T T}^G T^C T^A T^C T^T T^G T^C T^G T G G G G G G G G G G G G G G A A A A A A A A A A A A A A G G G G G G G G G G G G G G A C^{A A A A A A A A A}^G_{C C C}^A C^A C_{C C C}^A C^A C_{C C C}_C^G C^G C^G C^G C^G C^G C^G C^G C^G C^G C^G C^G C^G C A A A A A A A A A A A A A A A C A A A A A A A A A A A A A T^C T^C T^C T^{C C C C C C C C C C}^{A A A A}_{T T} A_T A_T A_T A_{T T} A_{T T T}_{T T T T}^A T_{T T T T}^A_T^A T^A T^A T G G G G G G G G G_T G G G G G G T^G T^G T^G T^G T^G T^G T^G T^G T^G T^G T^G T^G T^G T_e G G G G G G G G G G G G G G d^A C^{A A A A A A A A A A A A A}_.^g_o_n^c_{C C C C C C C C C C C} G G_{C C} Gⁱ_r G G G G G G G G G G G_c^a A A A A A A A A A A A A A Aⁿ_B A A A A A A A A A A A A A A_e^u_q^e_s²₂-m^r_o^{L 2}₇^f_e^l₆⁵₂²_{2 2 2 -} -²₇₆⁵₂_o^s_{p 5 E l 2 3 o}^C -_l -₂ -_{3 o}^{C C} i m^{a 7}₃ M^{T T T C}^W_{C C}^{T T T W}_{C C} K_{C C C e} A A_{C C C e} A A^d_{a S} A_S M M M M U U M M M M U U^e_r_gⁿ_o O^{L. N}₁^D_{e I}^l_b Q⁶^{a E}₄⁷₄⁸₄⁹₄⁰₅¹₅²₅³₅⁴₅⁵₅^{6 7 8 9}_{T S}⁵₂⁵₂⁵₂⁵₂⁵₂⁵₂⁵₂⁵₂⁵₂⁵₅₂⁵₅₂⁵_{5 5}₂⁵₂⁵₂ Melanoma RNA-Seq Analyses [00430] RNA-Seq paired-end BAM files from 472 Melanoma patients collected by TCGA (SKCM) were obtained from the GDC portal following dbGAP approval (phs000178.v10.p8). A second collection of 40 RNA-Seq FASTQ files from patients who underwent immunotherapy (archival formalin-fixed, paraffin-embedded) in Van Allen cohort (33) files were obtained from the dbGAP database (phs000452.v2.p1). These FASTQ files were aligned to the reference human genome (hg38) and transcriptome (Ensembl 91) using STAR. Among these 40 Van Allen RNA-Seq samples, patient 41 was excluded (only partial sequencing data available), as previously reported. The HLA genotype of each patient sample was determined from the RNA-Seq FASTQ files using the software Optitype 1.3.3. Optitype was chosen based on its superior performance on calling HLA-I alleles (over 99% accuracy) from RNA-Seq data based on recent large-scale benchmark, evaluated on “gold-standard” HLA genotyping data. [00431] AltAnalyze v 2.1.4 was used to quantify splicing independently in these two cohorts using the Ensembl version 91 database. MultiPath-PSI identified splicing events were used as inputs for SNAF. The TCGA survival and mutation data were downloaded from Xena Browser. Survival analysis was performed using the snaf.survival_analysis function with stratification argument n=2 (high burden is equivalent to > the median burden, low burden < the median burden). Mutation analysis was conducted using snaf.mutation_analysis function. [00432] To identify individual neojunctions or neoantigens associated with survival, a univariate Cox Regression analysis was used to identify events/antigens with a parental PSI value that are positively or negatively associated with patient outcome. Here, the neojunction and its parental PSI value is ignored if the neoantigen was not predicted to be presented in that sample, resulting in different survival associations for different neoantigens produced from the same neojunction. For analysis of Melanoma RNA-Seq TCGA samples in the SNAF-B workflow, long-read Iso-Seq cDNA sequences were obtained from pan-cancer cell line sequencing, using the PacBio provided isoform GTF file (https <colon slash slash> downloads <dot> pacbcloud <dot> com <slash> public <slash> dataset <slash> UHRRisoseq2021 <slash> Final-MappedTranscripts). Normal Tissue Reference RNA-Seq [00433] A subset of represented samples (n=1,215 samples) for all GTEx healthy tissues were obtained as RNA-Seq FASTQ files from dbGAP (phs000424.v8.p2). These FASTQ files were aligned to the reference human genome (hg38) and transcriptome (Ensembl 91) using STAR prior to analysis in SNAF. Selective Amino Acid Motif Enrichment Analysis [00434] The shared and unique neoantigens were defined by accounting for the frequency of their parental splicing junction. The occurrence frequency of a neoantigen were denoted as and its parental splicing junction (neojunction) as . By default, a splicing neoantigen classified as shared if: Instead, a splicing

(13) [00435] Each amino

vector of length 12 using a reduced representation of over 500 physicochemical properties derived from the AAindex database. A detailed description of the feature selection criteria and dimension reduction procedures is provided in DeepImmuno. Briefly speaking, each amino acid was associated with over 500 numerical properties. A Principal Component Analysis (PCA) was performed to reduce the dimension to 12, which was further used for encoding the peptide sequence. [00436] The encoded neoantigens were embedded into a 2D space using Uniform Manifold Approximation and Projection (UMAP). The most statistically significant motifs were derived using the MEME suite differential enrichment mode on the official website, the minimum and maximum length of the motif were set to 9 to reflect the queried 9mer sequences. The plots were produced in Python or Prism version 9. SNAF Interactive Web Application [00437] An interactive web application for both SNAF-T and SNAF-B visualization results was created using the Python Dash framework. The front-end was written in React.js and the backend in the Flask framework. Plotly-py was used for interactive visualization and the Ensembl REST API used for retrieving transcript reference sequences and Needleman-Wunsch global alignment. The peptide weblogo plot was implemented. For each position , the height of the stack (i) represents the difference between the maximum entropy and the actual entropy among the observed amino acids distribution. [00438] the observed

frequency of amino acids at position . Within each stack, the

of each amino acid was determined by their relative frequencies. AlphaFold23D modeling [00439] AlphaFold2 was obtained from https <colon slash slash> github <dot> com <slash> deepmind <slash> alphafold and implemented as a Singularity container from the provided Dockerfile, to be run on a Linux HPC, along with all training databases. For the indicated examples, the ranked_0.pdb was chosen as the 3D modeling result, and the protein cartoon visualization and coloring were performed using PyMol software. EXAMPLE 2 SNAF Computational Workflows – Inferring New Classes of Neoepitopes from RNA-Seq [00440] The SNAF workflow begins with user-supplied BAM files from tumor samples or cancer cell lines, followed by the identification and quantification of diverse classes of post-transcriptional regulation. In particular, the workflow applies a highly accurate approach for local splicing variation (MultiPath-PSI) from the AltAnalyze framework, to detect known and novel alternative splicing (cassette exon, 3′/5′ splice site exon, intron retention, alternative terminal exon, trans-splicing) and alternative promoter regulatory events, which can produce unique exon-exon or exon-intron junctions for in silico translation (Figure 3). [00441] This approach has been extensively benchmarked against diverse local- splicing variation (LSV) approaches, including rMATs, Leafcutter and MAJIQ, with specialized methods to accurately quantify retained introns (Figure 4). The produced splice-junction/sample count matrix is compared against a MultiPath-PSI pre-processed database of normal human healthy tissues (GTEx and TCGA) to identify those that are tumor-specific (23) (Figure 5). Tumor-specific splice junctions can be analyzed in parallel with SNAF-T and SNAF-B. [00442] SNAF-T consists of: 1) HLA type prediction from sample FASTQ files (user provided); 2) in-silico translation; 3) MHC-binding prediction (NetMHCpan or embedded calls to MHCflurry); and 4) HLA-allele specific immunogenicity prediction (DeepImmuno). [00443] SNAF-B consists of: 1) full-length isoform prediction for each tumor-specific splice-junction by augmenting existing isoform references; 2) exclusion of isoforms predicted to induce nonsense-mediated decay (NMD); 3) transmembrane topology prediction; and 4) long-read isoform sequence validation and augmented prediction (optional). [00444] For both of the SNAF-T and SNAF-B workflows, a Maximum Likelihood Estimation and separate hierarchical Bayesian model (BayesTS) can optionally be applied to assess the tumor specificity of each neojunction, e.g. relative to GTEx, in SNAF’s default and an optional custom healthy tissue reference RNA-Seq data, with custom tissue weighting assigned by the user. Normal healthy tissue references can include but are not limited to 54 tissue regions from the GTEx consortium, human fetal development through adult independent pan-tissue bulk RNA-Seq cohorts, single-cell RNA-Seq of diverse human tissues, and long-read isoform sequencing from normal healthy tissues (e.g., GTEx). Finally, to identify causal regulators of splicing neoantigen production, RNA-SPRINT (RNA-based Splicing PRotein activity INference from multivariate decision Trees) was developed to infer splicing factor activity directly from tumor RNA-Seq splicing profiles (see Example 1). [00445] The workflows described herein are unique in both design and functionality (see Table 2). Unlike prior T cell-based splicing neoantigen prediction approaches, SNAF is fully automated, supports any human genome version, has an embedded database of healthy reference RNA-Seq profiles (such as, for example, GTEx and fetal human development), identifies intron retention associated antigens, and enables more accurate prediction of immunogenicity. As the program has a modular design with well-described Python classes, it can be extensively customized to incorporate additional reference datasets for verification (e.g., control RNA-Seq, long-read sequencing) and alternative algorithms (e.g,. MHC binding prediction). [00446] While existing splice-neoantigen methods only consider T cell antigens, SNAF-B provides independent evidence of tumor-specific transmembrane proteins (ExNeoEpitopes) that can uncover novel extracellular epitopes for antibody recognition. SNAF predicts ExNeoEpitopes from a reference mRNA transcript database through the removal of alternative spliced-out transcript sequences in the longest reference model and through matching of tumor-specific splice-junctions to large long-read isoform sequencing experiments or reposited expressed sequence tags (ESTs). [00447] SNAF automates the analysis of survival within a cancer patient cohort to prioritize neoantigens for further evaluation. SNAF further automates comparative proteomics analysis of produced candidate splicing neoantigen peptide sequences to produced mass- spectrum profiling datasets, such as immunopeptidomics and surface proteomics datasets, to confirm neoantigen cell-surface or MHC presentation. Such predictions are used to derive a ranking of splicing neoantigen importance, in conjunction with other SNAF produced outputs, including, but not limited to, immunogenicity score, HLA-binding score, neoantigen tumor specificity scores relative to healthy control samples (maximum likelihood Estimation, Bayesian hierarchical model, control cohort mean counts), alternative splicing PSI value, mRNA junction occurrence in the reference transcriptome (Ensembl version 91), patient cohort occurrence frequency, recurrence in independent cancers and cohorts, splice-event type, full-length peptide prediction length versus the reference, peptide sequence similarity to reference peptides, the presence of signal peptide sequence, secondary structure, 3D protein structure prediction, insertion versus deletion of amino acids, absolute junction read counts, topology predictions and other programmatic outputs. [00448] SNAF can identify both precision (patient specific) and shared neoantigens, which are frequently presented by MHC and are immunogenic or which produce ExNeoEpitopes. The present inventors have found that shared splicing neoantigens can be distinguished from patient-specific neoantigens in their amino acid sequence and corresponding physicochemical parameters, further enabling systemic prioritization. [00449] SNAF has a modular design with well-described Python classes, enabling extensive customization to incorporate additional reference datasets for verification (e.g., custom control RNA-Seq, long-read sequencing) and alternative algorithms (e.g,. MHC binding prediction). Further, interactive visualization tools enable the user to explore sequence features shared among common versus unique neoantigens or explore background signals in specific GTEx tissues (outlier junction reads) on a case-by-case basis, following prioritization. [00450] A central component of the SNAF prediction workflow is the ability to more accurately predict which splicing neoantigens will selectively induce a T-cell response based on which HLA alleles are present in a given patient specimen. As previously demonstrated, the DeepImmuno workflow can predict immunogenic tumor neoantigens with up-to a 2-fold greater sensitivity than alternative approaches to assess antigen immunogenicity. This workflow has been independently validated as a top-performer in predicting immunogenicity versus other publicly available tools. Unpublished and commercial analytical platforms also designed to identify splicing neoantigens for T-cell and B-cell based immunotherapies include SpliceCore and IRIS, but such tools are limited in their features and capabilities, which limits their utility (see Table 2).

^._s w^o_l^f_k^r_o^w

ⁿ_e^g_i^t_n^a_o^e_n_gⁿ_i^ci_l^p_s_d^e_h^si_l^b_u^pr_e^h_t^o^o_t_FA^N_S_nⁱ_s^e_r^u_t^a_e^f_f^oⁿ_o^si_r^a_p^m_o^C.₂_e^l_b^a_T EXAMPLE 3 Preliminary Evaluation of the Computational Workflows – Validation of Predicted MHC-Bound Neoantigens [00451] The DeepImmuno workflow can predict immunogenic tumor neoantigens with up-to a 2-fold greater sensitivity than alternative approaches. To determine whether SNAF- T can identify bonafide MHC-presented splicing neoantigens, two prior produced cancer immunopeptidome datasets to validate its predictions were selected. First, bulk RNA-Seq (single-end) was evaluated and immunopeptidome profiling data (HLA-bound peptides) was matched from 14 ovarian cancer patients (27). To expand these predictions, SNAF-T was applied to skin cutaneous melanoma (SKCM) biopsy RNA-Seq from The Cancer Genome Atlas (TCGA) initiative (n=472), and unmatched immunopeptidome data from 24 melanoma patients, herein referred to as the Bassani Sternberg cohort (28) (Figure 6A). In ovarian tumors, SNAF identified 46 splicing neoantigens on average per sample detected by MS as MHC presented in all 14 patients (7.40% of all detected), ranging from 12 to 160 (Figure 6B). [00452] This number is expected to be an underestimate, as HLA-bound peptides arise from non-specific protease cleavage (which alters the resultant MS spectra), and hence traditional tryptic-based search engines cannot confidently recover all neoantigens. Here, the absolute number of MS-supported splicing neoantigens is higher than previously reported (average 2 peptides per sample) using untargeted proteome data (CPTAC), indicating increased sensitivity of targeted immunopeptidome for identifying valid and rare neoantigens. [00453] As MS-based predictions are subject to inherent type-1 errors, select ovarian and melanoma SNAF-T neoantigens were validated using targeted MS on synthesized neopeptides. Specifically, 14 ovarian and 22 melanoma peptides were selected with high- confidence spectra (see Example 1). Of the 36 tested, 27 were sufficiently detected by MS. Comparison of the synthetic to the original MS2 spectra found 11 matches, of which 7 were high scoring based on all match criteria (see Example 1). [00454] These seven neoantigens derive from multiple mechanisms, including known cassette exons, alternative 3’ and 5’ splice sites, intron retention and novel cassette exons (AltAnalyze defined). For example, the shared melanoma splicing neoantigen HAAASFETL in the gene FCRLA occurs due to a known cassette exon-exon junction in an isoform that is weakly detected in blood and spleen (average read count = 0.51, BayesTS: 0.03) (Figure 6C). [00455] Expression of FCRLA, which is a member of the F-receptor-like immunoglobulins, is correlated with good prognosis in Melanoma. Interestingly, FCRLA gene expression is only weakly tumor-specific (BayesTS: 0.13). However, this junction is detected in >34% of TCGA melanoma patients (162/472, Average read count = 52.91) (Figure 6D). The resultant MS2 spectrum of this antigen has a high-confidence match (Andromeda score: 149.06, P-value: 0.0), with a synthetic spectrum Pearsonr similarity of 0.63 (p-value: 0.05, cosine similarity: 0.87). The other MS2 confirmed neoantigens derive from ubiquitin protein ligase complex (FBXO7), asparagine amidase (NGLY1), cytoskeletal motor protein (DYNLT5), negative regulator of RAS signaling (RASA3) and currently uncharacterized protein coding genes (C20orf204, C6orf52) (Figure 7). Lower spectrum similarity scores are observed in the Ovarian cohort (Pearsonr 0.55) compared to the Melanoma cohort (Pearsonr: 0.84), which can be attributed to differences in the MS technology applied (Ion trap MS versus Fourier transform MS) and fragmentation methods (CID versus HID). EXAMPLE 4 Splicing Neoantigen Burden Predicts Overall Survival and Response to Immunotherapy [00456] To broadly assess the clinical relevance of splicing neoantigens, SNAF was applied to >500 melanoma patient biopsies with (i.e., Van Allen cohort) (33) and without immunotherapy (TCGA-SKCM). These analyses separately considered the number of predicted neojunctions, MHC-bound peptides, immunogenic neoantigens and overall neoantigen burden, considering RNA-Seq determined HLA alleles for each patient. [00457] In the TCGA cohort of 472 samples, 528 tumor-specific splicing junctions (neojunctions) were found per patient on average, ranging from 28 to 1,549 neojunctions in each patient. From these neojunctions, an average of 1,090 MHC-bound peptides were predicted, ranging from 75-2,981 peptides per patient. DeepImmuno predictions reduced the number of neoantigens to 915 on average (ranging from 74-2,486/patient), filtering out 16% of potentially non-immunogenic bound peptides, as current immunogenicity approaches still suffer from low precision, which may explain the relatively low filtering rates from netMHCpan-reported bound peptides. To only prioritize the most high-confident immunogenic neoantigens for downstream experimental validation, SNAF further allows users to increase the default immunogenicity cutoffs (i.e. Deepimmuno default cutoff = 0.5) and supply external immunogenicity predictions (i.e. IEDB, DeepHLApan). [00458] To investigate the relationship between splicing neoantigen burden and clinical outcome, survival analyses were performed on both the TCGA and Van Allen cohorts. This analysis surprisingly found that patients in TCGA with high MHC-bound neoantigen burden trends towards poor overall survival (log rank p<0.05) (Figure 8A). Conversely, the analysis found that patients with a high neoantigen burden that received ICB (Van Allen cohort) have not statistically significant but improved overall survival (log rank p=0.18). These trends were also reflected in neojunction and immunogenic peptide burden (Figure 8B). Since the bulk RNA-Seq data profiled in these cohorts was obtained prior to therapy, one plausible explanation is that patients with high neoantigen burden exhibit frequent tumor immune escape, which is overcome by CTLA-4 inhibition. [00459] Examination of differential gene expression in TCGA patients with high splicing neoantigen burden versus low found 597 up- and 227 down-regulated genes (fold>1.5 and eBayes t-test p<0.05, FDR) (Figure 8C). Regarding the survival results, the top most- differentially up-regulated genes were those previously implicated in poor response to immunotherapy, immune evasion and tumor invasion (e.g., ADAM10, PTPN11, TGFBR1, TNPO1, ANKRD52, MIB1, KIF3B). [00460] In contrast, down-regulated genes were markers of active tumor-immune cell infiltration, in particular plasma cells and T-cells (Figure 8C) with corresponding enrichment of this immune infiltration signature, along with genes involved in antigen binding and complement activation (Figure 8D). Enrichment of this immune infiltration signature was observed in the low-burden patients by gene-set enrichment. [00461] Enrichment analysis of high burden induced transcripts identified upregulation of p53 signaling, mitotic cycle-cycle, cell motility and signaling MET, among others (Figure 9A,). Hence, patients with splicing neoantigen burden express genes that promote immune evasion while low burden patients show evidence of immunoreactivity. [00462] Since both chemotherapy and radiotherapy target cancer cells via directly inducing DNA damage or reactive oxygen species, reprogramming of metabolic pathways and increased DNA repair machinery have been recognized as possible mechanisms of chemo- and radio-therapy resistance. Combination radiotherapy and immunotherapy have been proposed as a strategy to overcome single therapy drug resistance, which aligns with the observation of elevated immune evasion genes in high-burden groups. Thus, high splicing neoantigen patients (TCGA) can represent chemo- and radio-therapy resistant tumors with upregulated immune evasion capacity, representing candidates for combination therapy. In contrast, low neoantigen burden patients evidence immune reactivity, which can partially explain their favorable prognosis. EXAMPLE 5 CAMMK2 Mutations Correlate with Increased Neoantigen Burden in Melanoma [00463] To determine whether splicing neoantigen burden in melanoma is associated with specific mutations, mutations in high- versus low-burden neoantigen patients were compared. While no individual mutations were significant based on FDR correction, one of the most significant hits relates to mutations in the gene CAMKK2 versus wild type. Specifically comparing patients with and without CAMKK2 mutations, a significantly higher neoantigen burden was found with mutations (Mann Whitney test p=0.0001) (Figure 8E). Out of 19 patients with reported CAMKK2 mutations, 18 were found to occur in the high burden group, out of a total of 222 patients. [00464] CAMKK2 has been previously identified as a target to improve immunotherapy outcomes, as its mutation or inhibition leads to increased anti-PD1 immunotherapy efficacy. This effect is believed to occur via CAMKK2’s ability to negatively regulate ferroptosis, a mechanism of cell death that is induced by iron-dependent lipid peroxidation, via the AMPK‒NRF2 pathway. These results indicate that CAMKK2 mutations can contribute to splicing neoantigen burden and immunotherapy outcome, either through direct or indirect splicing regulatory networks, in a subset of melanoma patients. EXAMPLE 6 Individual Splicing Neoantigens can Predict Response to Immunotherapy [00465] In addition to neoantigen burden associations with survival, the individual PSI profiles of SNAF immunogenic neoantigens were compared with patient survival. This analysis identified 2,970 parental junctions associated with poor overall survival in TCGA (likelihood- ratio test (LRT) p<0.05 and z ≥ 1) (Figure 8F). Among these neoantigens, it was found that a subset (n=108 junctions) was present in over 15% of patients (shared neoantigens), indicating that these represent novel survival biomarkers. Although a much smaller number of neoantigens was associated with survival in the Van Allen cohort due to the limited sample size, the analysis found 1,755 poor and 227 good prognosis associated neoantigen junctions following CTLA-4 inhibitor treatment (LRT p<0.05 and z ≥ 1 or z ≤ 1, respectively). [00466] Importantly, 7 unique neoantigen junctions in TCGA and Van Allen were found to be associated with opposite survival associations (poor in TCGA and good in Van Allen) and present in >10 TCGA patients (Figure 8F). For example, a splicing neoantigen in the melanin synthesis associated glucose transporter SLC45A2 (TEFQTRRAM) was detected in 212/472 TCGA SKCM patients; this neoantigen is associated with poor overall survival in TCGA (Wald p <0.05, z-score>2) and is associated with good overall survival in the Van Allen cohort (Wald p<0.05, z-score<-2), indicating that it represents a relatively common target for therapy and prognosis (Figure 8F). [00467] Other splicing neoantigens were associated with poor overall survival in TCGA; these were identified in the majority of patients in both cohorts (e.g., SLC45A2 - FQTRRAMTL). Other overlapping neoantigens produced from this same junction were present in >64% patients in both melanoma cohorts (FQTRRAMTL), with differences in their abundances due to varying MHC-I allele binding and immunogenicity preferences. Dozens of other shared neoantigens were associated with poor prognosis in TCGA and trending towards good prognosis in the Van Allen cohorts. The occurrence and tumor specificity of these and other similar neojunctions were readily confirmed by IGV visualization, and frequently associated with alternative splice sites in conjunction with tumor-specific gene expression (Figure 9B). These data find that individual shared splicing neoantigens can predict outcomes in patients prior to and after ICB. EXAMPLE 6 Disruptive splicing factor activity underlies high splicing neoantigen burden [00468] The coordinated regulation of alternative splicing and alternative promoters are orchestrated by the expression or activity of conserved cis- and trans-regulatory interactions (e.g., splicing factor binding to RNA recognition elements). Splicing neoantigens are likely produced by modulation of such interactions, driven by direct mutations or RNA-editing that result in unique tumor-specific mRNAs. [00469] The most likely mediator of differential splicing neoantigen burden is a change in the expression or activity of one or more splicing factors. Examination of differential gene expression in the high versus low splicing neoantigen group found a large number of upregulated splicing regulators (n=20) (Figure 8C). These included 117 mRNA splicing regulators, upregulated with a fold > 1.2 and eBayes t-test p<0.05, FDR. Patients with splicing factor mutations in SF3B1 were further enriched in the high-burden group (Mann Whitney p=0.02). [00470] This surprising finding led to an investigation of whether splicing factor activity was also increased with burden. Existing methods for transcription regulatory network inference, which rely on a defined set of target genes or regulons, were used to infer splicing factor activity. The up- and down-regulation of these regulons can indirectly provide evidence of TF activity. A large dataset of RNA-Binding Proteins (RBP) knockdowns (n=191) in K562 cells from the ENCODE project (see Example 1) was used to establish a reliable link between splicing factors and downstream splicing junctions. Initial splicing regulons were derived by observing the changes in splicing events upon the knockdown of specific RBP and accounting for batch effects. To refine these regulons, evidence of direct RBP-target regulation was incorporated using eCLIP sequencing from 120 in the K562 cell line. The resulting data were used to construct a prior network to infer splicing factor activity (see Example 1, Figure 10A). [00471] Multivariate Decision Tree (MDT) was found to produce the most accurate RBP activity predictions, compared to 11 other highly-used transcription factor (TF) activity inference methods, when considering 116 RBP knockdown splicing profiles from HepG2 cells (ENCODE) (Figure 11A-C). Application of this approach, called RNA-SPRINT, to all TCGA melanoma patient PSI profiles, finds that 209 out of 221 examined RBPs have reduced versus increased activity in the high versus low splicing neoantigen group (Mann-Whitney p<0.05) (Figure 10B-10C). [00472] While contrary to expectations based on RBP gene expression, these data indicate that there exists a coordinated failure to properly splice transcripts, as previously described. This result is further supported by an observed increase in intron-retention in the high- burden patients (failure to excise introns) (Figure 10D). It is also possible that broad upregulation of splicing factors is a compensatory mechanism to counteract global splicing defects or simply a byproduct of cell-type heterogeneity within the tumor (e.g., immune infiltration). EXAMPLE 7 Shared splicing neoantigens are presented by MHC and display amino acid compositional bias [00473] As an important observation from the SNAF-T analyses is the co-occurrence of splicing neoantigens among many patients, the study examined broadly shared splicing neoantigens observed in melanoma, versus those observed in individual patients, to understand such regulation. Intriguingly, the analysis found 940 T-cell neoantigens predicted in >15% of patients in TCGA (shared) (Figure 12A). Although a much smaller cohort, 439 of 8,422 shared neoantigens were found in Van Allen, overlapping with the 940 in TCGA. [00474] TCGA shared neoantigens were found to more frequently result from cassette exon as compared to unique neoantigens (only found in one patient) (Figure 12B). Genes with shared neoantigens were significantly enriched in gene sets for melanocyte biology, such as melanocyte differentiation (p = 9.3 e-4, FDR), melanin biosynthetic process (p = 3.4 e-3, FDR) and cell division (p = 1.0 e-4, FDR) (Figure 12C). Thus, this analysis revealed that splicing neoantigens tend to occur in resident melanocyte genes and those which regulate NMD regulation and are predominantly associated with alternative splice-site selection. [00475] As shared splicing neoantigens are rare relative to all unique predicted neoantigens, it would be expected that such peptides are more frequently detected using immunopeptidome profiling in an independent cohort. Hence, a previously published melanoma dataset of HLA-I bound peptides detected by MS in 24 independent patients was analyzed (the Bassani-Sternberg cohort). [00476] Considering both shared (n=613) and unique (n=16,753) peptides in the reference peptide database (combined unique reference database, excluding peptides in UniProt), the study indeed found that shared splicing neoantigens can be detected at a higher rate than unique neoantigens (paired t-test, p=0.006) (Figure 12D). Additionally, of the 613 shared neoantigens examined, immunopeptidome evidence was found for 34% (n=210), with 98 shared neoantigens observed in at least 15% of patients by immunopeptidome analysis in Bassani- Sternberg, as illustrated using kernel density estimates or estimator of the empirical cumulative distribution (eCDF) function (Figure 12E, Figure 13A). Inspection of neojunctions produced from these MHC-presented neoantigens showed that they were tumor-specific (Figure 13B). Taken together, this MS validation from an independent melanoma cohort further demonstrates that those common neoantigens can be leveraged as high-value therapeutic targets. [00477] In addition to exploring the regulation of alternative splicing, the study investigated whether selective amino acid preferences exist between shared and unique splicing neoantigens. While HLA genes are highly polymorphic in humans and have different binding preferences for neoantigens, the existence of shared splicing neoantigens suggests that diverse HLA genotypes can bind in a more promiscuous manner to these peptides versus those present only in one or few individuals. [00478] To evaluate the potential recurrence of amino acid sequences, the shared and unique neoantigens were first redefined by normalizing the frequency of their parental splicing junctions (Figure 12F, see also the methods described in Example 1). Since 9-mer neoantigens have a balanced number of shared and unique neoantigens, the focus was specifically on these peptides. To derive a physicochemical profile of the amino acids associated with each, each 9- mer was encoded as a numerical vector based on all amino acid physicochemical parameters in the AAIndex database (566 parameters) (see the methods described in Example 1), and each neoantigen vector was projected as a point in UMAP space (Figure 12G). While the majority of shared and unique splicing neoantigens have similar physicochemical characteristics, a few empirically observed clusters were found to be enriched in shared versus unique neoantigens. [00479] To find preferentially detected amino acids present in the shared and unique neoantigens, motif analysis was performed with MEME in these two sets. MEME finds that shared neoantigens frequently end in lysine and phenylalanine (e-value=9.8e-14), whereas the most dominant motif found in unique neoantigens disproportionately end in arginine (e- value=0.14); the less significant E-value also indicates the more diverse binding modes for unique neoantigens (Figure 12H). Near identical motif enrichments were observed for shared vs. unique splicing neoantigens in the Van Allen cohort (Figure 13C). These data indicate that amino acid sequence bias can be used to find shared neoantigens that are bound by most MHC alleles. [00480] To demonstrate that such splicing neoantigens represent viable targets for broadly applicable immunotherapies, their ability to directly bind MHC and induce T-cell reactivity was experimentally verified. For validation, 5 shared splicing neoantigens were selected, derived from three neojunctions in PMEL, SLC45A2 and CDH19. While PMEL is an existing target for immunotherapy in melanoma (in phase I/II clinical trials), the predicted splicing neoantigen has not been previously described (novel exon-exon junction). These neoantigens were selected based on availability of HLA matched donor cells, high neoantigen frequency in melanoma and tumor specificity. [00481] An MHC stabilization assay using TAP deficient HLA-A*01 containing T2 cells confirmed the binding of two neoantigens to HLA-A*02 (Figure 14A-B). To assess immunogenicity, HLA-typed healthy blood PBMC was leveraged to prime T cells and test IFNγ responses upon pMHC stimulation (Figure 15). Frozen PBMC were primed 3 times with autologous peptide loaded DCs and thereafter tested for their IFNγ response against single HLA expressing 721.221 (221) cells, to ensure immunogenicity is specific for a single HLA genotype. The 5 neoantigen peptides loaded on 221 cells expressing their respective allele, generated similar IFNγ responses compared to known immunogenic FLU and/or HMCV peptide antigens in 3 donors (Figure 14C-D). Only weak IFNγ responses were detected in unstimulated T cells and in response to 221 without peptide, showing peptide specificity. Thus, these analyses provide strong evidence of T-cell immunogenicity by all 5 shared splicing neoantigens tested. Notably, prior described neoantigen prediction workflows have reported far lower validation accuracies. EXAMPLE 8 Shared splicing neoantigens derive from tumor cells rather than the tumor microenvironment [00482] Emerging data suggests that neoantigens can also be derived from the tumor microenvironment, including immune cells. While bulk tumor RNA-Seq enables the detection of splicing neoantigens, it cannot clarify the precise cellular source. To determine which cell-types splicing neoantigens derive from, single-cell (sc)-RNA-Seq from melanoma tumor biopsies from 17 patients (4,454 cells) was re-analyzed. These data were profiled using SmartSeq2 chemistry, allowing for the detection of over 340,000 exon-exon and exon-intron junctions present throughout the gene body, for 7 prior annotated cell-types (tumor, endothelial, cancer associated fibroblasts, B cell, T cell, natural killer cells, and macrophages). [00483] This analysis found that only a small proportion of scRNA-Seq detected junctions are enriched in tumor cells >2 fold (n=30,523) versus those enriched in immune cells (n=236,954) (Figure 16A). Intriguingly, roughly equal numbers of immune and tumor enriched neojunctions were associated with TCGA melanoma neoantigens (1,289 and 979, respectively). However, restricting this analysis to shared splicing neoantigens in >15% of patients found that these neojunctions are almost entirely derived from tumor cells (n=195) versus immune (n=10) (Figure 16B). Tumor specific neojunctions included experimentally confirmed splicing antigens (SCL45A2, CDH19, FCRLA) (Figure 16C). Thus, the tumor microenvironment appears to be a significant contributor to splicing neoantigen burden, but not a primary source for shared neoantigens. [00484] To ensure that such neoantigens are not only a byproduct of a normal cellular proliferative program, thus limiting their therapeutic use, multiple in vitro bulk RNA-Seq datasets, in which human epidermal cells were induced to proliferate, were analyzed. This analysis found that proliferation associated junctions can only account for a small percentage (~6%) of all TCGA melanoma identified neojunctions (Figure 17A-C). Neojunctions from shared splicing neoantigens were frequently observed in only one sample from these data, specifically from 16 week fetal fibroblasts, but not from other fetal fibroblasts profiled. Furthermore, these overlapping melanoma neojunctions were only weakly detected (1-50 reads) in any cultured epidermal sample and were associated with cell cycle regulation (Figure 17D). EXAMPLE 9 SNAF accurately predicts full-length mRNAs and stable proteoforms [00485] An intriguing set of SNAF-T splicing neoantigens were those that occur in transmembrane proteins, such as GPR143 and SLC45A2, which occurred due to novel in-frame alternative splice-sites (5′ or 3′). Such splicing events can result in novel cell-surface expressed proteins. As an alternative source of neoantigens that do not have degradation, MHC presentation, and T-cell receptor recognition, SNAF-B was applied to the same TCGA melanoma cohort. The SNAF-B workflow can be used to identify ExNeoEpitopes that have predicted transmembrane domains, but altered N-terminal or other extracellular sequences that could serve as novel epitopes for specifically designed monoclonal antibodies. [00486] To predict full-length isoforms, novel exon-exon or exon-intron junctions (not in the Ensembl or UCSC mRNA database), can be inserted into the best matching isoform models based on exon composition, followed by in silico translation (Figure 18A, see also the methods described in Example 1). This workflow can optionally exclude predicted mRNAs that are expected to result in NMD and selectively include those that have high-confidence transmembrane domains based on a prior published Hidden Markov Model based topology prediction approach (TMHMM). Putative ExNeoEpitopes that result in deleted or novel extracellular polypeptides can be determined using the SNAF-B interactive viewer. [00487] This analysis found 378 initial ExNeoEpitopes in the TCGA melanoma cohort, using prior well annotated transcripts as reference models (Ensembl, UCSC mRNAs). To initially assess the validity of such ExNeoEpitopes, long-read RNA isoform sequencing (“Iso- Seq”) was performed in four commonly used melanoma cell lines using the PacBio Sequel II platform. This Iso-Seq analysis found 17 of the predicted ExNeoEpitopes isoforms that perfectly matched the in silico predictions, and an additional 20 which partially matched (overlapping neojunction). One example is a novel alternative 3′ site in Signal Regulatory Protein Alpha (SIRPA), which results in an in-frame deletion of the 5′ end of exon 13, missing 21AA. The predicted full-length isoform was initially confirmed by Iso-Seq (Figure 18B). [00488] In addition to predicting full-length isoforms, SNAF-B can match short-read identified junctions to other supplied isoform models (e.g., long-read or EST). Matching exon- exon and exon-intron junctions from the TCGA melanoma cohort to a publicly available PacBio Iso-Seq dataset of 10 commonly used cancer cell lines (Universal Human Reference) found 1207 additional full-length neo-isoforms associated with junctions not readily detected in GTEx. This included the same SLC45A2 neojunction, predicted by SNAF-T to be associated with poor survival, shared in 69% of patients, found to be MHC-presented (MS) and only expressed in tumors (Figure 19A-B). This alternative 5′ donor site results in the deletion of 80AA which disrupts the 6th transmembrane domain, with the deleted region (AA 215-295) appearing to be composed of a transmembrane segment (AA 215-237) and a cytoplasmic segment (AA 237-295) in the reference protein (Figure 13C-D). As a result of the deletion, a region of polypeptide sequence normally positioned at the cytoplasmic face of the membrane is now predicted to reside in the extracellular domain, representing a potential novel neoepitope for CAR-T therapy. [00489] Considering all SNAF-B short- and long-read predictions from 10 tumor cell lines (pan-cancer) together, SNAF-B was run to identify a total of 514 unique ExNeoEpitope proteins. These were filtered to 187 predictions in which the neojunction overlapped with the extracellular domain (UniProt) and was not contained within any other Ensembl or UCSC protein isoforms. In addition to proteoforms with missing polypeptide sequences, the study identified 12 initial candidate long-read supported isoforms with high GTEx evidence of tumor- specificity (mean reads/sample <0.1) that result in the inclusion of novel alternative first or cassette exons. These were initially identified through manual inspection of BLAT sequence matches to the human genome and mRNA transcript databases (UCSC and Ensembl) and biased to ExNeoEpitopes detected in >15% of patients. [00490] Visualization in the SNAF-B viewer found that 5 out of these 12 ExNeoEpitopes result in new inserted polypeptide sequences that impact a cytoplasmic region of the protein (OCA2, SLC2A10, TMEM9, IL13RA1, ATP13A1), one occurring within a transmembrane domain (ANO10), and five predicted to alter the extracellular region of the protein and result in stable transmembrane predictions (DCBLD2, NALCN, MET, SEMA6A, IGSF11). [00491] While orthogonal long-read RNA sequencing demonstrates the validity of these transcripts from neojunction inference, it is possible that such isoforms are not properly folded or inserted into the cell membrane. First, to initially evidence the ability of these isoforms to produce functional transmembrane proteins, 2D and 3D protein structures were predicted for the tumor-specific and reference mRNA isoforms using Protter and Alphafold2 (57), respectively. In each of the examples, high-confident 2D and 3D structures of both the tumor and reference isoforms were observed, indicating that the deleted or inserted polypeptide sequence selectively impacts the extracellular portion of these proteins (Figures 18-21). [00492] Finally, the ability of ExNeoEpitopes, with distinct in silico evidence, to traffic to the plasma membrane was evaluated. Specifically, C-terminal fluorescent tagged cDNAs were synthesized and transfected for three ExNeoEpitopes that gain new peptide sequences (SEMA6A, ANO10, IGSF11, DCBLD2) and three isoforms with deletions (SIRPA, MET, SLC45A2), along with the reference isoform for each. The synthesized reference isoforms for 5 out of the 7 ExNeoEpitopes (SEMA6A, SIRPA, MET, SLC45A2, DCBLD2) were expressed and found to co-localize to the transmembrane using a selective cell surface stain in HEK-293 cells. Confirming their expression and trafficking, partial transmembrane localization was also observed in 4 of 7 of the evaluated ExNeoEpitopes predicted by SNAF-B (SEMA6A, SIRPA, MET, SLC45A2) (Figure 18F-G, Figure 21B-F). [00493] These data illustrate that tumor-specific splice isoforms can be higher precision candidates for CAR-T or monoclonal antibodies over existing targets. This can include conformational epitopes that impact protein structure. EXAMPLE 10 Neoantigen target-containing peptides and proteins identified via SNAF-T and SNAF-B, respectively [00494] To identify shared splicing neoantigens that are specific to one cancer or shared among multiple cancers, the SNAF-T and SNAF-B workflows were applied to 5 cancers: 1) melanoma, 2) ovarian cancer, 3) head and neck cancer (HNC), 4) breast cancer, and 5) myelodysplastic syndrome (MDS). All analyses were performed on publicly available datasets comprising: 1) bulk RNA-Seq patient individual tumor biopsies (NCI Genomics Data Commons or NIH Sequence Read Archive), 2) immunopeptidomics mass spectrometry profiles, 3) single- RNA-Seq data, 4) cancer cell lines long-read mRNA sequencing isoform models, and/or 5) associated open-access clinical metadata. The applied processes included the SNAF software in combination with the prioritization/selection/in silico confirmation processes of the SNAF-T and STAF-B workflows, as described herein, for all steps, with the following exceptions: 1) integrated immunopeptidomics analyses were applied in melanoma and ovarian cancer, 2) cell- of-origin single-cell splicing analyses were applied in melanoma and MDS, and/or 3) confirmatory independent cohort analyses were performed in melanoma (Van Allen cohort). [00495] With the exception of MDS, all analyses were performed on solid cancer tumor biopsies from the TCGA project. In the case of MDS, analyses were performed on bulk RNA-Seq from an independent MDS cohort of 44 patients and matched healthy controls bone marrow liquid biopsies. [00496] MDS is a heterogeneous disorder of the hematopoietic stem cell (HSC) resulting in bone marrow failure and frequent progression to acute myeloid leukemia (AML). MDS is diagnosed in over 10,000 patients yearly (98). Survival varies greatly by the subtype of MDS, with survival ranging from few months (high risk MDS, t-MDS) to several years (low risk). Despite the introduction of new therapies, such as hypomethylating agents and immunomodulatory drugs, the only curative therapy to date remains allogeneic stem cell transplant. However, marrow transplant is limited to a subset of patients due to its toxicity and limited donor availability (in particular in older individuals). A striking finding in MDS is the frequent recurrence of somatic mutations in core factors that mediate pre-mRNA splicing (>50%). Such splicing factor mutations lead to global alterations in mRNA products shared in MDS and AML. These mRNA isoforms represent largely unexplored targets for new immunomodulatory therapies for the prevention of cancer in aging adults. [00497] Over 4,000 dysregulated and coordinated splicing events are induced in MDS patients with evidence of chronic TGF-beta signaling. This splicing was attributed to a loss of HNRPK activity, which is selectively impaired in HSC (patients, engineered mouse models and single cell splicing). Thus, coordinated signaling-mediated splicing changes and splicing factor mutations, are a major contributor to HSC splicing dysfunction in MDS, prior to transformation to AML. MDS clones loose niche associations and mobilize from the bone marrow to blood. Thus, liquid biopsies can be readily collected from blood for RNA- and protein-based diagnostics. [00498] To evidence the existence of splicing antigens in MDS for the creation of new immunoprevention strategies, SNAF was applied to 44 patients and 23 controls bone morrow biopsies (RNA-Seq). The analyses of this MDS cohort demonstrate the existence of 5,473 splicing antigens, including 78 shared in 10-75% of patients. These include targets for antigen vaccines (predicted SNAF-T immunogenic) and those that alter the sequence composition of extracellular protein domains of transmembrane encoding proteins (ExNeoEpitopes, SNAF-B) (Figures 22 and 23). While the latter predictions are derived from short-read sequencing, they are secondarily evidenced from long-read single-cell RNA-Seq profiling of MDS bone marrow. [00499] In head and neck cancer tumor biopsies form individual patients profiled in TCGA, SNAF-T splicing neoantigens corresponding to 155 neojunctions were found to be predicted to be presented in > 15% of patients. Further, 84 prioritized ExNeoEpitope full-length isoforms present in >15% of patients were identified. In breast cancer, 155 SNAF-T shared splicing neoantigens present in >10% of patients and 118 SNAF-B ExNeoEpitopes present in >10% of patients were identified. [00500] The definition of shared splicing neoantigens was set at either >10% or >15%, depending on the size of the cohort and previous extent of tumor heterogeneity. Taken together, 2241 unique shared predicted immunogenic splicing neoantigens were found in these cohorts (SNAF-T) (SEQ ID NOs: 1-2241, included in the sequence listing submitted concurrently herewith; see also Table 3 of U.S. Provisional Application No. 63/580,300, IDENTIFICATION OF TARGETS FOR IMMUNOTHERAPY IN MELANOMA USING SPLICING-DERIVED NEOANTIGENS, filed on filed September 1, 2023, which is currently co-pending herewith and which is incorporated by reference in its entirety). Among these 2241 predictions, 124 shared splicing neoantigens were present in 2 or more cancers. Of these 2241 shared splicing neoantigens identified by SNAF-T, four have been experimentally validated for immunogenicity, as shown in Table 3, namely SEQ ID NOs: 1-4. [00501] Using SNAF-B, 304 unique shared predicted ExNeoEpitope protein isoforms were identified (SEQ ID NOs: 2242-2545, included in the sequence listing submitted concurrently herewith; see also Table 4 of U.S. Provisional Application No. 63/580,300, IDENTIFICATION OF TARGETS FOR IMMUNOTHERAPY IN MELANOMA USING SPLICING-DERIVED NEOANTIGENS, filed on filed September 1, 2023, which is currently co-pending herewith and which is incorporated by reference in its entirety). Among these 304 predictions, 54 were found to be present in 2 or more cancers, including the in vitro experimentally evidenced ExNeoEpitope in the gene MET (Melanoma and Head and Neck cancer).

²_A_e^{5 9}₁^L_A n₄^L e CH_E D^M_R C

ⁿ₃- -⁵_{- -}_u^d m_r^o_{4 9}⁹₈²₉_o⁰₅¹₉^{1 4} mⁱ_r C^{4 0}₆⁵₀¹^o_{f n 5} o⁹₅⁶₉⁵₇ d_eⁱ_{3 6 5}¹₆ t_t_a^{c 3}_:^:₈^:₂¹_: d_n^{5 1 1 1} i^u_r l_J^h_r_c^h_r_c^h_r_c^h_c_a^v^yl_l^a_tⁿ_e m_i^r_e^p_x^eⁿ_e^e_b_e^v_a^h^h_cⁱ_h w T-_F A N_S_y^b^d_eⁱf_i^t_n^e_dⁱ s K n^L e_T g_i^C^t M_R_n^A_R^R^a_o^R_L e^V n_{T P}_g^Qⁿ_F^L_P_i^E^ci^M lⁿ^p_{e T}_s^gi G_V_{d t}^{L D}^{e n}_{L L}_{r a}^R|^a_{I L}^Q_|_h^o_e D^{F R L}^S N N_V P_L^T_T E^._{3 g} G_{E A}_eⁿ^l_i^T_F^S b^{c Q}ⁱ_{l S} Q_{I A} DN^S_E^A^{a p}_S^P_I D^{I T}_A_{T I S} H 1_{2 3 4} Q E^. S^D_{I o} 160 N EXAMPLE 11 Exemplary Splicing-Derived Neoantigens (RLLGTEFQTRRAMTLK, IPDSQGNDI, QRETDFKMKF, TVTAGRQGIY, VILQDAGIGFI) [00502] The processes described in Examples 1-10 above have resulted in the identification of various splicing-derived neoantigens, some candidates of which are described below. [00503] Candidate 1: [00504] Two splicing-derived neoantigens (RLLGTEFQTRRAMTLK and IPDSQGNDI), encoded by the gene SLC45A2 (ENSG00000164175:E3.2_33963931-E4.2), were identified and found to be associated with poor overall survival (Wald test p<0.05, z-score >2) in a large diagnostic melanoma RNA-Seq cohort (TCGA) and good overall survival (Wald test p<0.05, z-score <-2.0) in a melanoma immunotherapy RNA-Seq cohort (phs000452.v3). This neoantigen is not readily detected in normal healthy tissues (GTEx, see Figure 6B) but is frequently detected in melanoma (328/472 patients) by SNAF and by immunopeptidomics (4 independent melanoma patients). [00505] Accordingly, this neoantigen is predictive of good response with standard anti-PD-1 checkpoint blockade therapy. As such, this candidate can be used: A) as a diagnostic (quantitative PCR for exon-exon junction chr5:33954504-33963931) to identify patients that are likely to benefit from standard therapy. [00506] Candidate 2: [00507] A splicing-derived neoantigen (QRETDFKMKF), encoded by the gene TBC1D16 (ENSG00000167291:E38.6-E39.1, chr17:79950594-79950756), was identified and found to be associated with poor overall survival (Wald test p<0.0005, z-score >3) in a large diagnostic melanoma RNA-Seq cohort (TCGA) and decent overall survival (Wald z-score <-1.5) in a melanoma immunotherapy RNA-Seq cohort (phs000452.v3). This neoantigen is not readily detected in normal healthy tissues (GTEx, see Figure 7 and Figure 8) but is frequently detected in Melanoma (158/472 patients) by SNAF. [00508] Accordingly, this neoantigen is a predictive of poor survival. As such, this candidate can be used: A) as a diagnostic (quantitative PCR for exon-exon junction chr17:79950594-79950756) to identify patients that may benefit (mixed outcome) from standard therapy, and/or B) targeted immunotherapy using this neoantigen. [00509] Candidate 3: [00510] A splicing-derived neoantigen (TVTAGRQGIY), encoded by the gene GPR143 (ENSG00000101850:E8.1-E9.1_9743655, chrX:9743655-9746044), was found to be associated with poor overall survival (Wald test p<0.05, z-score >2) in a large diagnostic melanoma RNA-Seq cohort (TCGA). This neoantigen is not readily detected in normal healthy tissues (GTEx, see Figure 6B) but is frequently detected in melanoma (192/472 patients) by SNAF. [00511] Accordingly, this neoantigen is a predictive of poor response with standard anti-PD-1 checkpoint blockade therapy. As such, this candidate can be used: A) as a diagnostic (quantitative PCR for exon-exon junction chrX:9743655-9746044) to identify patients that are candidates for new therapeutics, and/or B) targeted immunotherapy using this neoantigen. [00512] Candidate 4 [00513] A splicing-derived neoantigen (VILQDAGIGFI), encoded by the gene MCF2L (ENSG00000126217:E18.4-E19.1, chr13:113044888-113045271), were found to be associated with poor overall survival (Wald test p = 0.05, z-score = 2) in a large diagnostic melanoma RNA-Seq cohort (TCGA) and poor overall survival (Wald test p = 0.05, z-score = 1.9) in a melanoma immunotherapy RNA-Seq cohort (phs000452.v3). This neoantigen is not readily detected in normal healthy tissues (GTEx) but is frequently detected in melanoma (136/472 patients) by SNAF. [00514] Accordingly, this neoantigen is a predictive of poor response with standard anti-PD-1 checkpoint blockade therapy. As such, this candidate can be used: A) as a diagnostic (quantitative PCR for exon-exon junction chr13:113044888-113045271) to identify patients that are unlikely to benefit from standard therapy but are candidates for new therapeutics, and/or B) targeted immunotherapy using this neoantigen. [00515] Other exemplary neoantigens identified by the processes described herein include RETDFKMKF, YQDWQQNLY, FNEDGLRQV, as well as the immunogenic targets listed in SEQ ID NOs: 1-2545 submitted concurrently herewith. EXAMPLE 12 Interactive Neoantigen Web Explorer [00516] To facilitate the exploration and prioritization of the predicted neoantigens from SNAF, two interactive web applications were developed to visualize both T-cell and B-cell neoantigens (Figure 24). The SNAF-T viewer allows users to explore different global neoantigen features (i.e., amino acid length, and frequency within a cohort) in a projected 2D UMAP space. Here, each neoantigen is embedded based on its physicochemical properties. Users can manually select clusters to identify enriched web logo AA motifs and explore individual neojunctions and neoantigens. The SNAF-B viewer can be used to interactively visualize GTEx and tumor counts for neojunctions, generate protein sequence alignments between a putative ExNeoEpitope and selected reference proteins, compare protein feature composition, secondary structure and solvability prediction, and perform topology modeling. Hence, these tools can be used to select optimal targets for experimental validation. [00517] The various methods and techniques described above provide a number of ways to carry out the disclosure . Of course, it is to be understood that not necessarily all objectives or advantages described can be achieved in accordance with any particular embodiment described herein. Thus, for example, those skilled in the art will recognize that the methods can be performed in a manner that achieves or optimizes one advantage or group of advantages as taught herein without necessarily achieving other objectives or advantages as taught or suggested herein. A variety of alternatives are mentioned herein. It is to be understood that some preferred embodiments specifically include one, another, or several features, while others specifically exclude one, another, or several features, while still others mitigate a particular feature by inclusion of one, another, or several advantageous features. [0100] Furthermore, the skilled artisan will recognize the applicability of various features from different embodiments. Similarly, the various elements, features and steps discussed above, as well as other known equivalents for each such element, feature or step, can be employed in various combinations by one of ordinary skill in this art to perform methods in accordance with the principles described herein. Among the various elements, features, and steps some will be specifically included and others specifically excluded in diverse embodiments. [0101] Although the application has been disclosed in the context of certain embodiments and examples, it will be understood by those skilled in the art that the embodiments of the disclosure extend beyond the specifically disclosed embodiments to other alternative embodiments and/or uses and modifications and equivalents thereof. [0102] In some embodiments, the numbers expressing quantities of ingredients, properties such as molecular weight, reaction conditions, and so forth, used to describe and claim certain embodiments of the application are to be understood as being modified in some instances by the term “about.” Accordingly, in some embodiments, the numerical parameters set forth in the written description and attached claims are approximations that can vary depending upon the desired properties sought to be obtained by a particular embodiment. In some embodiments, the numerical parameters should be construed in light of the number of reported significant digits and by applying ordinary rounding techniques. Notwithstanding that the numerical ranges and parameters setting forth the broad scope of some embodiments of the application are approximations, the numerical values set forth in the specific examples are reported as precisely as practicable. [0103] In some embodiments, the terms “a” and “an” and “the” and similar references used in the context of describing a particular embodiment of the application (especially in the context of certain of the following claims) can be construed to cover both the singular and the plural. The recitation of ranges of values herein is merely intended to serve as a shorthand method of referring individually to each separate value falling within the range. Unless otherwise indicated herein, each individual value is incorporated into the specification as if it were individually recited herein. All methods described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The use of any and all examples, or exemplary language (for example, “such as”) provided with respect to certain embodiments herein is intended merely to better illuminate the application and does not pose a limitation on the scope of the application otherwise claimed. No language in the specification should be construed as indicating any non-claimed element essential to the practice of the application. [0104] Preferred embodiments of this application are described herein. Variations on those preferred embodiments will become apparent to those of ordinary skill in the art upon reading the foregoing description. It is contemplated that skilled artisans can employ such variations as appropriate, and the application can be practiced otherwise than specifically described herein. Accordingly, many embodiments of this application include all modifications and equivalents of the subject matter recited in the claims appended hereto as permitted by applicable law. Moreover, any combination of the above-described elements in all possible variations thereof is encompassed by the application unless otherwise indicated herein or otherwise clearly contradicted by context. [0105] All patents, patent applications, publications of patent applications, and other material, such as articles, books, specifications, publications, documents, things, and/or the like, referenced herein are hereby incorporated herein by this reference in their entirety for all purposes, excepting any prosecution file history associated with same, any of same that is inconsistent with or in conflict with the present document, or any of same that may have a limiting affect as to the broadest scope of the claims now or later associated with the present document. By way of example, should there be any inconsistency or conflict between the description, definition, and/or the use of a term associated with any of the incorporated material and that associated with the present document, the description, definition, and/or the use of the term in the present document shall prevail. [0106] In closing, it is to be understood that the embodiments of the application disclosed herein are illustrative of the principles of the embodiments of the disclosure . Other modifications that can be employed can be within the scope of the application. Thus, by way of example, but not of limitation, alternative configurations of the embodiments of the application can be utilized in accordance with the teachings herein. Accordingly, embodiments of the present application are not limited to that precisely as shown and described.

Claims

Claims What is claimed is: 1. An immunogenic target associated with one or more types of cancer, wherein the immunogenic target comprises a shared splicing amino acid-based neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1-2545, or a nucleotide coding therefor.

2. The immunogenic target of claim 1, wherein the target comprises a shared splicing neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1- 2241.

3. The immunogenic target of claim 2, wherein the target comprises a shared splicing neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1-4.

4. The immunogenic target of claim 1, wherein the target comprises a shared splicing neoantigen present in a protein having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 2242-2545.

5. The immunogenic target of claim 4, wherein the target comprises a shared splicing neoantigen present in a protein having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 2242-2245.

6. The immunogenic target of claim 1, wherein the target comprises a nucleotide coding for a shared splicing neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1-2545.

7. The immunogenic target of claim 1, wherein the target comprises a nucleotide coding for a shared splicing neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1-4 or SEQ ID NOs 2242-2245.

8. The immunogenic target of any one of claims 1-7, wherein the immunogenic target comprises at least 90% sequence homology to any of SEQ ID NOs: 1-2545, or a nucleotide coding therefor.

9. The immunogenic target of claim 1, wherein the target comprises a shared splicing neoantigen selected from SEQ ID NOs: 1-2545.

10. The immunogenic target of claim 9, wherein the target comprises a shared splicing neoantigen selected from SEQ ID NOs: 1-2241.

11. The immunogenic target of claim 9, wherein the target comprises a shared splicing neoantigen selected from SEQ ID NOs: 1-4.

12. The immunogenic target of claim 9, wherein the target comprises a shared splicing neoantigen selected from SEQ ID NOs: 2242-2545.

13. The immunogenic target of claim 1, wherein the target consists of a shared splicing neoantigen selected from SEQ ID NOs: 1-2545.

14. The immunogenic target of claim 13, wherein the target consists of a shared splicing neoantigen selected from SEQ ID NOs: 1-2241.

15. The immunogenic target of claim 13, wherein the target consists of a shared splicing neoantigen selected from SEQ ID NOs: 1-4.

16. The immunogenic target of claim 13, wherein the target consists of a shared splicing neoantigen selected from SEQ ID NOs: 2242-2545.

17. The immunogenic target of any one of the preceding claims, wherein the target is in vitro.

18. A cell comprising the immunogenic target of any one of the preceding claims.

19. An immunotherapy for treating one or more type of cancer, wherein the immunotherapy is specific for, or comprises, one or more immunogenic target of any preceding claim.

20. The immunotherapy of claim 19, wherein the immunotherapy is specific for, or comprises, one or more shared splicing neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1-2241, or one or more nucleotide coding therefor, and wherein the type of cancer comprises skin cancer, breast cancer, head and neck cancer, ovarian cancer, and/or myelodysplastic syndrome (MDS).

21. The immunotherapy of claim 20, wherein the immunotherapy is specific for, or comprises, one or more shared splicing neoantigen selected from SEQ ID NOs: 1-2241, or one or more nucleotide coding therefor, and wherein the type of cancer comprises skin cancer, breast cancer, head and neck cancer, ovarian cancer, and/or myelodysplastic syndrome (MDS).

22. The immunotherapy of claim 19, wherein the immunotherapy is specific for, or comprises, one or more shared splicing neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1-4, or one or more nucleotide coding therefor, and wherein the type of cancer comprises skin cancer.

23. The immunotherapy of claim 22, wherein the immunotherapy is specific for, or comprises, one or more shared splicing neoantigen selected from SEQ ID NOs: 1-4, or one or more nucleotide coding therefor, and wherein the type of cancer comprises skin cancer.

24. The immunotherapy of claim 19, wherein the immunotherapy is specific for, or comprises, one or more shared splicing neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 2242-2545, or one or more nucleotide coding therefor, and wherein the type of cancer comprises skin cancer, breast cancer, head and neck cancer, ovarian cancer, and/or myelodysplastic syndrome (MDS).

25. The immunotherapy of claim 24, wherein the immunotherapy is specific for, or comprises, one or more shared splicing neoantigen selected from SEQ ID NOs: 2242-2545, or one or more nucleotide coding therefor, and wherein the type of cancer comprises skin cancer, breast cancer, head and neck cancer, ovarian cancer, and/or myelodysplastic syndrome (MDS).

26. The immunotherapy of claim 19, wherein the immunotherapy is specific for, or comprises, one or more shared splicing neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 2242-2245, or one or more nucleotide coding therefor, and wherein the type of cancer comprises skin cancer and/or head and neck cancer.

27. The immunotherapy of claim 26, wherein the immunotherapy is specific for, or comprises, one or more shared splicing neoantigen selected from SEQ ID NOs: 2242-2245, or one or more nucleotide coding therefor, and wherein the type of cancer comprises skin cancer and/or head and neck cancer.

28. The immunotherapy of any one of claims 19-27, wherein the immunotherapy comprises an RNA-, DNA-, or peptide-based vaccine, dendritic cell vaccine, chimeric antigen receptor T cell (CAR-T) therapy, T cell receptor-engineered T cell (TCR-T) therapy, antibody or antibody derivative, cell-based immunotherapy, immune checkpoint inhibitor, immunomodulator, topical immunotherapy, injection immunotherapy, adoptive cell transfer, oncolytic virus therapy, immunosuppressive drug, helminthic therapy, and/or other non-specific immunotherapy.

29. The immunotherapy of any one of claims 19-28, wherein the immunotherapy comprises an RNA-, DNA-, or peptide-based vaccine, dendritic cell vaccine, CAR-T therapy, and/or antibody or antibody derivative.

30. A composition comprising the immunogenic target of any one of claims 1-18.

31. A composition comprising the immunotherapy of any one of claims 19-29.

32. The composition of any one of claims 30 or 31, wherein the composition is formulated as a vaccine.

33. The composition of any one of claims 30-32, wherein the composition further comprises one or more cytokines, growth factors, or adjuvants.

34. An in vitro isolated dendritic cell comprising the immunogenic target of any one of claims 1-18.

35. The dendritic cell of claim 34, wherein the dendritic cell comprises a mature dendritic cell.

36. The dendritic cell of any one of claims 34 or 35, wherein the cell is a cell with an HLA type selected from HLA-A, HLA-B, or HLA-C.

37. An engineered T-cell receptor (TCR) or chimeric antigen receptor (CAR) that specifically recognizes the immunogenic target of any one of claims 1-18.

38. A cell comprising one or more TCR or CAR of claim 37.

39. An antibody, or antigen binding fragment thereof, that specifically recognizes the immunogenic target of any one of claims 1-18.

40. A method of treating one or more type of cancer, the method comprising: administering, to a patient having a type of cancer, a treatment comprising therapeutically effective amount of a composition comprising the immunotherapy according to any one of claims 8-14.

41. A method of stimulating an immune response in a subject, the method comprising: administering, to a subject having a type of cancer, a treatment comprising therapeutically effective amount of a composition comprising the immunotherapy according to any one of claims 8-14

42. The method of claim 40 or 41, wherein the immunotherapy is administered in combination with one or more additional therapies.

43. The method of claim 42, wherein the immunotherapy and the one or more additional therapies are administered together in one administration or composition.

44. The method of claim 42, wherein the immunotherapy and the one or more additional therapies are administered separately in more than one administration or more than one composition.

45. The method of claim 42, wherein the one or more additional therapies comprise surgical intervention, chemotherapy, radiation therapy, hormone therapies, immunotherapy, and/or adjuvant systematic therapies.

46. The method of any one of claims 40-45, wherein the subject has a tumor resistant to chemotherapy and/or radiation therapy.

47. The method of any one of claims 40-46, wherein the subject has been determined to be resistant to the previous treatment.

48. The method of any one of claims 40-47, wherein the cancer comprises stage I, II, III, or IV cancer.

49. The method of any one of claims 40-48, wherein the cancer comprises metastatic and/or recurrent cancer.

50. The method of any one of claims 40-49, wherein the subject is enrolled in a clinical trial.

51. A method of identifying one or more immunogenic targets for treating a type of cancer, wherein the method comprises: receiving a sample dataset, wherein the sample dataset comprises mRNA from a cancer cell line, from a tumor cohort, or from a tumor sample from an individual patient; analyzing the sample dataset by a machine learning model comprising: identifying and quantifying post transcriptional regulation to detect one or more neojunctions and/or alternative splicing and/or alternative promoter regulatory events; determining one or more candidate neoantigen targets based on the neojunctions and/or alternative splicing and/or alternative promoter regulatory events; filtering the candidate neoantigen targets by: a) setting a minimum threshold for antigen abundance in the dataset; and/or b) assessing the specificity of exon-exon or exon-intron junctions for tumor samples versus a healthy reference cell line or cohort dataset based on junction counts and isoform structure association; and identifying a validated tumor- or cancer-specific immunogenic target when the candidate neoantigen target meets the filtering conditions.

52. The method of claim 51, the method further comprising identifying candidate neoantigen targets by one or more of the following: a. immunopeptidomics validation; b. long-read RNA isoform sequence validation; c. single-cell short-read and/or long-read RNA-Seq validation; d. identification of unique tumor-specific extracellular protein sequences; e. splicing event type filtering; f. BayesTS probability; g. survival association h. SashimiPlot visualization; i. BLAT searching; j.3D topology evaluation; and/or i. filtering the candidate neoantigen targets based on one or more of the above steps and/or one or more additional filtering step.

53. The method of claim 52, wherein: immunopeptidomics validation comprises filtering candidate neoantigen targets using one or more immunopeptidomics datasets comprising LC/MS data, with or without matching bulk RNA-Seq data; long-read RNA isoform sequence validation comprises confirming candidate neoantigen targets in long-read isoform sequencing data from the cancer cell line, tumor cohort, or tumor sample and/or identifying tumor-specific full-length transmembrane protein isoforms; single-cell short-read and/or long-read RNA-Seq validation comprises identifying common candidate neoantigen targets which are detected in one or more patients in a tumor sample versus microenvironment in a matching cancer and/or identifying tumor-specific full- length transmembrane protein isoforms; identification of unique extracellular tumor-specific protein sequences comprises sequence comparison and alignment with a prior database, determination of genomic correspondence, and prediction of secondary protein structures and protein features; splicing event type filtering comprises detecting splice-junctions not present in a reference database; Bayes TS probability comprises evaluating BayesTS for the healthy reference cell line or cohort dataset and/or the sample dataset; survival association comprises using a univariate Cox regression analysis to identify candidate neoantigen targets with a parental PSI value that is positively or negatively associated with patient outcome; SashimiPlot visualization comprises determining the specificity of exon-exon or exon- intron junctions for the healthy reference cell line or cohort dataset vs the sample dataset via junction counts and isoform structure association relative to known isoforms in a reference database; BLAT searching comprises evaluating full-length predicted mRNA sequences to confirm exon-region isoform specificity relative to prior annotated mRNA and ESTs and evaluating transposable element associations; and/or 3D topology evaluation comprises predicting 3D conformational differences in protein folding and domain architecture each candidate neoantigen target relative to a reference isoform.

54. The method of any one of claims 51-53, wherein filtering the candidate neoantigen targets comprises one or more of the following: a. determining if the splicing event is previously unknown; b. setting a threshold percentage of the sample dataset having the candidate neoantigen target; c. predicting 2D and/or 3D folding; d. predicting extracellular domain impact; e. SashimiPlot validation; f. evaluating secondary evidence from long-read sequencing; g. exclusion of protein isoform sequence data from a reference database; h. setting a threshold BayesTS probability score; i. setting a threshold average control tissue count; j. setting a threshold MHCNetPan binding; and/or k. setting a threshold DeepImmuno score > 0.5.

55. The method of claim 54, wherein: a. the candidate neoantigen target is present in greater than 5%, 10%, 15%, 20%, 25%, 30%, or higher, of the sample dataset; b. the BayesTS probability score is less than 10, less than 5, less than 4, less than 3, less than 2, less than 1, or lower; c. the average control tissue count mean is less than 10, less than 5, less than 3, less than 2, less than 1, less than 0.5, or lower; and/or d. the MHCNetPan binding is less than 5%, less than 4%, less than 3%, less than 2%, less than 1%, less than 0.5%, less than 0.1%, or lower.

56. The method of claim 55, wherein: a. the candidate neoantigen target is present in greater than 15% of patients in the cohort; b. the BayesTS probability score is less than 3; c. the average control tissue count mean is less than 1; and/or d. the MHCNetPan binding is < 2% as a weak binder and < 0.5% as a strong binder.

57. The method of any one of claims 51-56, wherein filtering the candidate neoantigen targets comprises: a. long-read single-cell genomics data; b. patient survival association data; c. immunopeptidomics data; and/or d. patient survival association with antigen presence.

58. The method of any one of claims 51-57, the method further comprising validating candidate neoantigen targets by LC/MS, peptide-MHC binding, proteomics, and/or immunogenicity.

59. The method of claim 58, wherein: a. validation via LC/MS comprises spectral matching; b. validation via peptide-MHC binding comprises performing an MHC stabilization assay; and/or c. validation via immunogenicity comprises testing pMHCs on HLA genotyped healthy blood PMBC.

60. The method of any one of claims 51-59, wherein a ranking is derived for the candidate neoantigen target.

61. The method of claim 60, wherein the ranking further comprises immunogenicity score, HLA-binding score, neoantigen tumor specificity scores relative to healthy control samples (maximum likelihood Estimation, Bayesian hierarchical model, control cohort mean counts), alternative splicing PSI value, mRNA junction occurrence in a reference transcriptome, patient cohort occurrence frequency, recurrence in independent cancers and cohorts, splice-event type, full-length peptide prediction length versus the reference, peptide sequence similarity to reference peptides, the presence of signal peptide sequence, secondary structure, 3D protein structure prediction, insertion versus deletion of amino acids, absolute junction read counts, topology predictions, and/or other programmatic outputs.

62. The method of any one of claims 51-61, wherein alternative splicing and alternative promoter regulatory events comprises cassette exon, 3′/5′ splice site exon, intron retention, alternative terminal exon, trans-splicing, and/or other event which can produce unique exon-exon or exon-intron junctions for in silico translation.

63. The method of any one of claims 51-62, wherein the method identifies intron retention associated antigens, and/or predicts immunogenicity of the candidate neoantigen target.

64. The method of any one of claims 51-63, wherein the neojunctions comprise gene- associated exon-exon and exon-intron junctions.

65. The method of any one of claims 51-64, wherein the machine learning model comprises a deep learning model, probabilistic algorithms (Maximum Likelihood Estimation (MLE), hierarchical Bayesian model), and/or long-read sequencing concordance analysis.

66. The method of any one of claims 51-65, wherein the machine learning model comprises a Maximum Likelihood Estimation (MLE), hierarchical Bayesian model, and long- read sequencing concordance analysis.

67. The method of any one of claims 51-66, wherein the immunogenic target is located on a transmembrane protein.

68. The method of any one of claims 51-67, wherein the sample dataset comprises mRNA from a cancer cell line.

69. The method of claim 68, wherein the immunogenic target is shared among a patient population.

70. The method of any one of claims 51-69, wherein the sample dataset comprises mRNA from a tumor sample.

71. The method of claim 70, wherein the immunogenic target comprises a patient-specific neoantigen.

72. The method of any one of claims 51-71, wherein the method further comprises one or more additional reference datasets and/or algorithms.

73. The method of any one of claims 51-72, wherein the method is automated.

74. The method of any one of claims 51-73, wherein the immunogenic target comprises a neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1- 2545, or a nucleotide coding therefor.

75. The method of claim 74, wherein the immunogenic target comprises a neoantigen selected from SEQ ID NOs: 1-2545, or a nucleotide coding therefor.

76. The method of any one of claims 51-73, wherein the immunogenic target comprises a neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1- 2241.

77. The method of claim 76, wherein the immunogenic target comprises a neoantigen selected from SEQ ID NOs: 1-2241.

78. The method of any one of claims 51-73, wherein the immunogenic target comprises a neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 1-4.

79. The method of claim 78, wherein the immunogenic target comprises a neoantigen selected from SEQ ID NOs: 1-4.

80. The method of any one of claims 51-73, wherein the immunogenic target comprises a neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID NOs: 2242-2545.

81. The method of claim 80, wherein the immunogenic target comprises a neoantigen present in a protein selected from SEQ ID NOs: 2242-2545.

82. The method of any one of claims 51-73, wherein the immunogenic target comprises a neoantigen having at least 80% sequence homology to the sequence in any of SEQ ID Nos: 2242- 2245.

83. The method of claim 82, wherein the immunogenic target comprises a neoantigen present in a protein selected from SEQ ID Nos: 2242-2245.

84. The method of any one of claims 51-73, wherein the immunogenic target comprises a nucleotide coding for a shared splicing neoantigen selected from SEQ ID NOs: 1-2545.

85. The method of any one of claims 51-73, wherein the immunogenic target comprises a nucleotide coding for a neoantigen selected from SEQ ID NOs: 1-4 or SEQ ID NOs 2242-2245.

86. The method of any one of claims 51-85, wherein the method further comprises treating the type of cancer by administering, to a subject with the type of cancer, a therapeutically effective amount of a treatment, wherein the treatment targets the tumor or cancer-specific immunogenic target.

87. The method of any one of claims 51-86, wherein the sample is a tumor sample from the subject, and wherein the treatment targets the tumor-specific immunogenic target.

88. The method of any one of claims 51-87, wherein the type of cancer comprises melanoma, ovarian, head and neck, breast, skin, lung, adult and pediatric acute myeloid leukemia, myelodysplastic syndrome (MDS), bladder, cervical, colorectal, kidney, low grade glioma, glioblastoma, liver, prostate, pancreatic, esophageal, rectal, sarcomas, stomach, thymus, thyroid, testicular, and/or uterine cancer.

89. The method of claim 88, wherein the type of cancer comprises melanoma, ovarian, head and neck, breast, skin, and/or MDS.

90. The method of any one of claims 51-89, wherein the treatment comprises an immunotherapy.

91. The method of claim 90, wherein the immunotherapy comprises an RNA-, DNA-, or peptide-based vaccine, dendritic cell vaccine, chimeric antigen receptor T cell (CAR-T) therapy, T cell receptor-engineered T cell (TCR-T) therapy, antibody or antibody derivative, cell-based immunotherapy, immune checkpoint inhibitor, immunomodulator, topical immunotherapy, injection immunotherapy, adoptive cell transfer, oncolytic virus therapy, immunosuppressive drug, helminthic therapy, and/or other non-specific immunotherapy.

92. The method of claim 91, wherein the immunotherapy comprises an RNA vaccine, chimeric antigen receptor T cell (CAR-T) therapy, and/or antibody or antibody derivative.

93. The method of any one of claims 51-92, wherein the immunotherapy comprises the immunotherapy of any one of claims xx-xx.

94. The method of any one of claims 51-93, wherein the type of cancer is responsive to targeting of the tumor- or cancer-specific immunogenic target.

95. A method of treating a type of cancer in a subject in need thereof, the method comprising administering to the subject a therapeutically effective amount of an immunotherapy comprising an immunogenic target selected from the peptides and proteins listed in SEQ ID NOs: 1-2545, or a nucleotide coding therefor.

96. A method of treating a type of cancer in a subject in need thereof, the method comprising administering to the subject a therapeutically effective amount of a compound, or a composition comprising a compound, directed to an immunogenic target selected from the peptides and proteins listed in SEQ ID NOs: 1-2545, or a nucleotide coding therefor.

97. The method of claim 96, wherein the compound binds to an immunogenic target selected from the peptides and proteins listed in SEQ ID NOs: 1-2545.