A neoantigen discovery method based on multi-omics data and progressive candidate screening

Abstract

The application discloses a neoantigen discovery method based on multi-omics data and progressive candidate screening. The application constructs a complete analysis process from multi-omics data input to high-confidence neoantigen candidate output, uses a unified associated candidate neoantigen analysis library as a data carrier in each stage, realizes the continuous association and whole-process tracking of somatic mutation information, expression information, functional annotation information, protein sequence information and binding capacity evaluation results, and overcomes the defects in the prior art that data processing links are mutually fragmented and candidate objects lack unified organization. On the basis of the aforementioned structure-unified analysis library, further multi-stage progressive candidate compression is carried out, biological characteristic constraints, immunogenicity evaluation and multi-dimensional candidate information aggregation are introduced in sequence, the candidate space is gradually reduced, high-confidence candidates with verification evidence are retained, candidate redundancy is significantly reduced, and the neoantigen screening efficiency and result stability are improved.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of bioinformatics, genomics, immunomics and artificial intelligence, and specifically to a method for discovering new antigens based on multi-omics data and progressive candidate screening. Background Technology

[0002] Tumor-specific antigens (TSA), also known as neoantigens, are antigenic peptides formed from novel protein sequences generated by somatic mutations, insertion / deletion variations, fusion gene mutations, or frameshift mutations in tumor cells. Because these antigens are not present in normal tissues, they can be processed and presented by the major histocompatibility complex (MHC) and further recognized by CD8+ cytotoxic T cells or CD4+ helper T cells, thereby inducing a specific anti-tumor immune response. Therefore, they are considered important targets for personalized tumor vaccines and tumor immunotherapy.

[0003] With the development of high-throughput sequencing technology, a neoantigen discovery workflow based on sequencing data from tumor and normal tissues has been established. A typical workflow includes multiple steps such as somatic mutation detection, human HLA typing, candidate peptide generation, MHC binding affinity prediction, antigen processing and presentation assessment, immunogenicity assessment, and candidate antigen sequencing. Researchers have successively developed prediction tools such as NetMHCpan, MHCflurry, pVACtools, Seq2Neo, and TSNAD to assist in neoantigen screening. However, neoantigen discovery relies on multi-stage, end-to-end collaboration, and accurately identifying reliable neoantigens from a massive pool of candidate peptides remains challenging. Layer-by-layer compression and efficient screening of massive candidate peptides has become a key technical problem limiting the efficiency and reliability of personalized tumor vaccine development.

[0004] While current neoantigen discovery technologies have developed relatively complete analytical workflows, they primarily focus on improving the predictive capabilities of individual steps, such as increasing the accuracy of MHC binding affinity prediction, constructing new presentation models, or developing new immunogenicity prediction models. However, insufficient attention is paid to the systematic compression of the candidate space and multi-stage synergistic screening. Specifically, existing workflows typically employ fixed-threshold screening strategies, such as filtering based on MHC binding affinity thresholds or percentile rankings. While this approach can eliminate some low-quality candidates, it still retains a large number of candidate peptides, resulting in a large final result size. This increases the cost of subsequent experimental validation and the difficulty of vaccine design, lacking an effective progressive compression mechanism. Simultaneously, most methods treat mutation detection, expression analysis, affinity prediction, presentation assessment, and immunogenicity analysis as independent steps, lacking a unified candidate compression strategy across stages. Some candidates, although possessing high predictive affinity, may originate from low-expression, low-clonal, or transcriptionally unsupported mutation sites, leading to a large number of false-positive candidates entering subsequent analysis processes, making it difficult to achieve synergistic multi-dimensional biological constraints.

[0005] Furthermore, because different prediction models employ different training data and modeling strategies, significantly different prediction results often emerge between different tools. Existing technologies typically rely on a single prediction model as the primary screening criterion, making candidate results susceptible to model bias and lacking stable and reliable candidate compression capabilities. At the final output level, most existing methods output a ranked candidate list, lacking a systematic design for candidate size control, validation priority allocation, and vaccine development needs. Researchers still need to rely on manual experience for secondary screening, resulting in low efficiency in neoantigen discovery and limited reproducibility of results. Overall, most current research focuses on performance optimization of a single prediction stage, lacking a unified compression framework from mutation discovery, candidate generation, affinity assessment to final candidate output, making it difficult to achieve continuous shrinkage of the candidate space and stable output of high-quality candidates. Summary of the Invention

[0006] To address the problems in existing neoantigen discovery technologies, such as fragmented data processing, lack of unified organization of candidate objects, insufficient connection between different analysis stages, high redundancy of candidate results, and insufficient reliability and traceability of final candidates, this invention provides a neoantigen discovery method and system based on multi-omics data and progressive candidate screening.

[0007] This invention does not aim to improve single peptide-MHC binding prediction models, immunogenicity evaluation models, or single screening indicators. Instead, it constructs a complete analytical workflow from multi-omics sequencing data input to high-confidence neoantigen candidate output. Based on tumor tissue DNA, paired normal tissue DNA, and tumor tissue RNA sequencing data, this workflow sequentially completes somatic mutation discovery, candidate peptide construction, candidate neoantigen analysis object establishment, peptide-MHC binding capacity evaluation, multi-stage candidate constraint, candidate status update, candidate association aggregation, and result output.

[0008] In this process, candidate analysis objects are jointly determined by somatic cell mutations, source transcripts, mutant candidate peptides, corresponding wild-type candidate peptides, and individual MHC alleles. Sequencing support information, expression information, functional annotation information, protein sequence information, and evaluation results at each stage are continuously linked to the same candidate analysis object. The candidate results, stage status, and newly added information formed in the previous stage serve as the basis for processing in the next stage, thereby achieving continuous data transfer between different analysis steps, maintenance of candidate object consistency, and full-process tracking of candidate results.

[0009] This invention aims to establish a unified, phase-coordinated, and scalable neoantigen discovery and screening system. While maintaining the integrity of candidate information and the ability to effectively identify candidates, it gradually reduces low-confidence and redundant candidates, improves the screening efficiency, result stability, and interpretability of high-confidence neoantigen candidates, reduces the candidate scale and implementation costs required for subsequent immunological experimental verification and personalized vaccine design, and provides repeatable and traceable technical support for tumor immunotherapy.

[0010] To achieve the above objectives, this invention provides a method for discovering new antigens based on multi-omics data and progressive candidate screening, comprising: S1. Obtain multi-omics sequencing data from the subjects, perform mutation detection, expression quantification, functional annotation and MHC typing, and obtain mutation information, annotation information, expression information and MHC allele information; S2. Based on the mutation information and annotation information, construct mutant candidate short peptides and corresponding wild-type candidate short peptides, and combine the candidate short peptides with the MHC alleles to obtain candidate peptide-MHC combinations. S3. Associate the mutation information, annotation information, expression information, and MHC allele information with the candidate short peptides to construct a candidate neoantigen analysis library; S4. Evaluate the MHC binding capacity of mutant and wild-type candidate short peptides in the candidate neoantigen analysis library, and write the evaluation results back to the corresponding candidate records to form an enhanced candidate neoantigen analysis library. S5. Based on the enhanced candidate neoantigen analysis library, the candidate space is progressively compressed to obtain a compressed candidate neoantigen set; S6. Organize the results and prioritize the compressed candidate neoantigen set to generate a high-confidence neoantigen candidate set.

[0011] Preferably, S1 specifically includes: S1.1 Obtain raw sequencing data of tumor tissue DNA, paired normal tissue DNA and tumor tissue RNA from the subjects, and perform quality control; S1.2 Align the quality-controlled tumor DNA and normal DNA sequencing reads to the reference genome to obtain tumor tissue DNA alignment files and normal tissue DNA alignment files; S1.3 Using the normal tissue DNA alignment file as a pairing control, perform somatic mutation detection on the tumor tissue DNA to obtain a somatic mutation set; S1.4 Perform quality control on the RNA sequencing data of the tumor tissue and align it to the reference genome, then quantify the expression to obtain RNA expression level information; S1.5. Perform functional annotation on the somatic mutation detection results obtained in S1.3 to obtain transcript annotation information; S1.6. Using the RNA sequencing data of the tumor tissue, MHC-I allele typing is performed to obtain the subject's MHC allele set.

[0012] Preferably, S2 specifically includes: S2.1 Based on the protein position, reference amino acid, and mutant amino acid in the transcript annotation information obtained in S1.5, determine the position of the mutation site in the protein sequence, extract a predetermined number of amino acids centered on the mutation site, and construct mutant local protein fragments and wild-type local protein fragments respectively. S2.2 Using a sliding window method, candidate short peptides containing mutated amino acids are generated from the mutant local protein fragments, and wild-type short peptides corresponding to the positions are generated simultaneously to obtain mutant candidate short peptides and corresponding wild-type candidate short peptides. S2.3. Combine each candidate short peptide record with all MHC alleles obtained in S1.6 to construct a candidate peptide-MHC combination space.

[0013] Preferably, in S3, the following combined identifiers are used to determine a peptide-MHC candidate record: Chrom, Pos, Ref, Alt, Transcript, MT_SHORT_PEPTIDE, and MHC; for the same mutation formed under different transcripts, different peptide windows, different peptide lengths, or different MHC alleles, each candidate is saved as an independent record.

[0014] Preferably, in step S4, the binding affinity assessment step includes: obtaining the predicted MHC binding affinity of mutant peptides and wild-type peptides respectively; after completing the prediction, writing the binding affinity results of mutant and wild-type peptides back to the corresponding candidate records to form an enhanced candidate neoantigen analysis library.

[0015] Preferably, in step S5, the progressive compression includes: S5.1. Based on the frequency of mutated alleles, expression level, filter field and sequencing depth information, the basic credibility of candidate records is judged, and the functional validity of candidate records is judged in combination with transcript function type constraints, thus completing the biological characteristic constraints. S5.2 Extract the mutant peptide-MHC binding affinity characteristics, the binding change ratio between mutant and wild type characteristics, and the interaction characteristics between the two from the candidate records after biological feature constraints, construct an immunogenicity feature vector, and call the pre-constructed immunogenicity evaluation module to calculate the immunogenicity probability of the candidate records. S5.3. Based on the immunogenicity probability and candidate comprehensive information, perform fine screening of the candidate space to obtain a compressed set of candidate neoantigens.

[0016] Preferably, in step S6, the output results include at least the mutation information corresponding to the candidate neoantigen, the mutant peptide and wild peptide sequences, MHC alleles, binding affinity evaluation results, VAF, TPM, and Biotype.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a complete analytical workflow from multi-omics data input to high-confidence neoantigen candidate output. Using a unified and interconnected candidate neoantigen analysis library as the data carrier for each stage, it achieves continuous correlation and full-process tracking of somatic mutation information, expression information, functional annotation information, protein sequence information, and binding capacity evaluation results. This overcomes the shortcomings of existing technologies, such as fragmented data processing stages and a lack of unified organization of candidate objects. Based on the aforementioned unified analysis library, a multi-stage progressive candidate compression process is further implemented, sequentially introducing biological characteristic constraints, immunogenicity evaluation, and multi-dimensional candidate information aggregation. This gradually reduces the candidate space, significantly decreasing candidate redundancy while retaining high-confidence candidates with validating evidence. This improves neoantigen screening efficiency and result stability, and reduces the candidate size and implementation cost for subsequent immunological experimental validation and personalized vaccine design. Attached Figure Description

[0018] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] Example The present invention will be further illustrated below through specific embodiments. This embodiment uses tumor tissue DNA, normal tissue DNA, and tumor tissue RNA sequencing data from a subject (number P4166) as input to detail the complete implementation process of a neoantigen discovery method based on multi-omics data and progressive candidate screening. The method flow is as follows: Figure 1 As shown, the steps include: S1. Obtain multi-omics sequencing data from the subjects, perform mutation detection, expression quantification, functional annotation and MHC typing, and obtain mutation information, annotation information, expression information and MHC allele information; S1.1 Quality control of raw sequencing data.

[0023] In this embodiment, for the test samples, the raw sequencing data in FASTQ format is first subjected to quality control. Quality control includes adapter sequence removal, low-quality base trimming, filtering of excessively short reads, and sequencing quality statistics. Specifically, the Fastp tool is used to perform the above processing, obtaining tumor DNA, normal DNA, and tumor RNA sequencing reads that meet the requirements for subsequent alignment. It should be noted that this embodiment is not limited to Fastp; other data processing programs capable of achieving the same quality control functions can also be used.

[0024] S1.2 Sequencing data alignment and standardization.

[0025] After quality control, tumor DNA and normal DNA sequencing reads were aligned to the GRCh38 / hg38 reference genome, respectively. In this embodiment, BWA was used for sequence alignment to generate alignment results in BAM format. The alignment results were then sorted by coordinates, read group information was added, and duplicate reads were marked. Subsequently, the base quality value of the BAM file was recalibrated using the Genomics Analysis Toolkit (GATK) to obtain standardized tumor tissue DNA alignment files and normal tissue DNA alignment files.

[0026] For any genomic location x, its sequencing depth is expressed as: in This indicates the number of reads supporting the reference allele. This indicates the number of reads supporting the mutant allele. Mutant allele frequency is expressed as: In the data structure of this embodiment, the mutation support depth, total sequencing depth, and mutation allele frequency are stored as SampleAD, SampleDP, and SampleAF fields, respectively.

[0027] S1.3, Somatic cell mutation detection.

[0028] Using normal tissue DNA alignment files as paired controls, somatic mutation detection was performed on tumor tissue DNA. In this embodiment, GATK Mutect2 was used to identify single nucleotide variants (SNVs) and short fragment insertion / deletion variants (InDels), and FilterMutectCalls was used to filter the initial variant results.

[0029] The detected set of somatic mutations is represented as: Among them, the i Individual cell mutations are represented as: in, Represents chromosomes. Indicates genomic location, Indicates the reference allele. This indicates a mutated allele.

[0030] This embodiment ultimately generates a somatic mutation file containing 1000 somatic cell mutations. Each mutation record includes at least Chrom, Pos, dbSNP, Ref, Alt, Qual, Filter, SampleAD, SampleDP, and SampleAF information.

[0031] S1.4 RNA sequencing data processing and expression level annotation.

[0032] RNA sequencing data underwent quality control and alignment to the GRCh38 / hg38 reference genome, followed by quantification of gene or transcript expression. In this embodiment, expression levels were expressed as TPM (Transcripts Per Million) and obtained from Salmon expression quantification results.

[0033] For transcript t, its TPM is represented as: in, This indicates the number of reads assigned to transcript (t). This indicates the effective length of the transcript. This represents the total number of transcripts involved in the expression level calculation.

[0034] In the subsequent VEP annotation file, this embodiment uses the TX field to store the corresponding expression level information. This field is used to determine whether the transcript containing the somatic mutation has RNA expression support.

[0035] S1.5, Somatic cell mutation function annotation.

[0036] Ensembl Variant Effect Predictor (VEP) was used to perform functional annotation on the mutation detection files. In this embodiment, after VEP annotation, for each annotation record, at least the following information was extracted: mutation type (Variant_Type), impact level (IMPACT), gene symbol (SYMBOL), gene identifier (Gene), feature type (Feature_type), transcript, transcript biotype (BIOTYPE), transcript support level (TSL), coding sequence alteration (HGVSc), protein alteration (HGVSp), protein position (Protein_position), reference amino acid (RefAA), and mutant amino acid (AltAA).

[0037] For the The mutation and its first Each transcript annotation represents the annotation object as follows: in, For transcripts, Types of mutation consequences As a transcriptome organism type, Location of the protein. and These are the reference amino acid and the mutant amino acid, respectively. This represents the corresponding expression level.

[0038] Furthermore, records that simultaneously meet the following conditions are filtered from the VEP annotation results: After the above screening, a total of 274 missense mutation transcript records with expression support were obtained. Since a single genomic mutation may correspond to multiple transcripts, further screening was conducted according to the combination of Chrom, Pos, Ref, and Alt. After deduplication, 146 missense mutations with expression support were obtained.

[0039] S1.6 and MHC allele typing identification.

[0040] MHC-I allele typing was performed using tumor RNA sequencing data. In this embodiment, six HLA-I alleles from the P4166 sample were obtained using OptiType: HLA-B07:02, HLA-B55:01, HLA-C07:02, HLA-A03:01, HLA-C03:03, and HLA-A02:01. The MHC allele set of the subjects is represented as follows: In this embodiment, m=6.

[0041] S2. Based on mutation information and annotation information, construct mutant candidate short peptides and corresponding wild-type candidate short peptides, and combine the candidate short peptides with MHC alleles to obtain candidate peptide-MHC combinations. S2.1 Construction of mutant and wild-type local protein sequences.

[0042] For the obtained missense mutations with expression support, the position of the mutation in the protein sequence is determined based on fields such as Transcript, HGVSp, Protein_position, RefAA, and AltAA.

[0043] Suppose the protein sequence containing a certain mutation is... The mutated amino acid is located at position k, and the reference amino acid is... The mutated amino acid is Centered on the mutation site, a predetermined number of amino acids were extracted upstream and downstream of the mutation site to form wild-type and mutant local protein fragments, respectively, as follows: Where N represents the number of amino acids extracted from both sides of the mutation site. In this embodiment, N=13. When the mutation is located near the start or end of the protein sequence, the local fragment boundaries are adjusted according to the actual available protein sequence. The generated wild-type and mutant local protein fragments are saved as Wildtype_Sequence and Mutant_Sequence, respectively.

[0044] S2.2 Generation of candidate short peptides containing mutation sites.

[0045] A sliding window approach was used to generate candidate short peptides containing mutated amino acids from locally mutated protein fragments, and simultaneously generate wild-type short peptides at the corresponding positions.

[0046] For a preset peptide length l, the mutant candidate peptide is represented as follows: The corresponding wild-type candidate peptide is The starting position i of the sliding window should satisfy the condition that the mutated amino acid is within the window coverage area. For MHC-I, the candidate peptide length set used in this embodiment is... .

[0047] For each mutant transcript, the lengths of the peptide segments and the starting positions of the windows that satisfy the mutation coverage criteria are traversed to form mutant and wild-type paired short peptides. The candidate peptide pairs are represented as follows: in, It is a mutant short peptide. For wild-type short peptides, The length of the peptide segment. Mutation of origin This is a source transcript.

[0048] This embodiment generated 10,198 candidate short peptide records with a length of 8–11 amino acids. These records included paired mutant and wild-type short peptides; after deduplication based on the mutant short peptide sequence, a total of 5,485 unique mutant short peptides were identified. The generated mutant and wild-type short peptides were saved as the MT_SHORT_PEPTIDE and WT_SHORT_PEPTIDE fields, respectively.

[0049] S2.3, combination of candidate peptides and MHC alleles.

[0050] Each candidate peptide record is combined with all MHC alleles obtained in step S1. For the candidate peptide pair set... and the MHC allele set Construct a candidate peptide-MHC combinatorial space Q×H, where each candidate combinatorial space is represented as: In this embodiment, the number of candidate short peptide records is 10198, and the number of MHC alleles is 6, thus forming 10198×6=61188 candidate peptide-MHC combination records.

[0051] S3. Link mutation information, annotation information, expression information, MHC allele information with candidate short peptides to construct a candidate neoantigen analysis library; The mutation information, sequencing support information, transcript annotation information, RNA expression information, and MHC typing information obtained in S1 are uniformly correlated with the local protein sequences and mutant and wild-type candidate short peptides generated in step S2 to construct a candidate neoantigen analysis library.

[0052] The candidate neoantigen analysis library is represented as follows: Where N represents the number of candidate peptide-MHC combinations. In this example, N = 61188. Each candidate record A single analytical object, identified by somatic mutations, transcripts, mutant peptides, wild-type peptides, and MHC alleles, can be represented as: in, Information on mutation sites, To provide sequencing support and expression information, This is a function annotation information. For local protein sequence information, Information on candidate short peptides, This is MHC allele information.

[0053] In this embodiment, the candidate neoantigen analysis library includes at least the following fields: Chrom, Pos, dbSNP, Ref, Alt, Qual, Filter, SampleAD, SampleDP, SampleAF, TX, Variant_Type, IMPACT, SYMBOL, Gene, Feature_type, Transcript, BIOTYPE, TSL, HGVSc, HGVSp, Protein_position, RefAA, AltAA, Wildtype_Sequence, Mutant_Sequence, MT_SHORT_PEPTIDE, WT_SHORT_PEPTIDE, and MHC.

[0054] Preferably, a peptide-MHC candidate record is identified using the following combined identifiers: For candidate antigens formed by the same mutation under different transcripts, different peptide windows, different peptide lengths, or different MHC alleles, they are saved as independent records, thereby preserving transcript differences, sequence differences, and MHC presentation differences in the candidate space. The candidate neoantigen analysis library generated in this embodiment contains a total of 61,188 records. This analysis library serves as a unified data carrier for subsequent binding ability prediction, candidate space compression, and candidate result output.

[0055] S4. Evaluate the MHC binding capacity of mutant and wild-type candidate short peptides in the candidate neoantigen analysis library, and write the evaluation results back to the corresponding candidate records to form an enhanced candidate neoantigen analysis library. Using the candidate neoantigen analysis library obtained in S3 as input, the MT_SHORT_PEPTIDE, WT_SHORT_PEPTIDE and MHC fields were extracted from each record to evaluate the binding ability of mutant peptide-MHC combinations and wild-type peptide-MHC combinations.

[0056] This embodiment uses the peptide-MHC binding affinity prediction module previously applied for by the applicant to perform binding capacity assessment. The module receives peptide sequences and MHC alleles as input and outputs the corresponding peptide-MHC binding affinity prediction results. The internal model construction, basic predictor selection, model training, and parameter optimization of this module are not part of the scope of this invention; this embodiment only uses it as a binding capacity evaluation module in an end-to-end process.

[0057] For the i-th candidate record, the binding affinity evaluation values ​​for the mutant and wild-type peptides are obtained respectively: in, This represents the peptide-MHC binding affinity prediction module. and The results represent the binding affinity evaluations of mutant and wild-type peptides, respectively.

[0058] In one implementation, the predicted value can be the IC50 predicted value, where a smaller IC50 value indicates a stronger predicted binding affinity. Alternatively, IC50 can be normalized using the following formula: After the prediction is completed, the binding affinity results of mutant and wild-type peptides are written back to the corresponding candidate records to form an enhanced candidate neoantigen analysis library: The symbol ⊕ indicates that fields are appended and results are merged according to the candidate record identifier. This embodiment yields 61,188 records after prediction, and this analysis database serves as the data carrier for subsequent multi-stage progressive compression of the candidate space.

[0059] S5. Based on the enhanced candidate neoantigen analysis library, the candidate space is progressively compressed to obtain the compressed candidate neoantigen set; Based on the enhanced candidate neoantigen analysis library formed by S4, the candidate space is progressively compressed in the order of basic biological feasibility constraints to multidimensional feature synergistic evaluation.

[0060] Let the initial augmentation candidate space be: After the k-th compression stage, we obtain: in, For the first Candidate constraint functions for each stage. Different biological features are used to constrain different stages. Each feature is not used as a simple linear weight for unified screening all at once, but plays a role in different stages of candidate space shrinkage.

[0061] S5.1 Biological characteristic constraints.

[0062] Preferably, a basic confidence assessment is performed by combining mutation and expression confidence levels, eliminating candidate records with insufficient mutation support, lack of expression support, or unacceptable sequencing quality. In this embodiment, the basic confidence assessment of candidate records is performed based on allele frequency, expression level, Filter field, and sequencing depth information. The mutation allele frequency satisfies SampleAF>0.05, the expression level satisfies TPM>0, and the Filter field is preferably PASS.

[0063] Preferably, the functional validity of candidate records is determined by combining transcript functional type constraints. In this embodiment, Variant_Type, BIOTYPE, IMPACT, TSL, and protein alteration information are used as key fields. Candidate records derived from protein_coding transcripts that can produce amino acid changes are preferentially retained, while records that do not produce effective protein products, are affected by nonsense-mediated degradation, or lack reliable transcript support are removed.

[0064] Preferably, by combining peptide-MHC binding capacity constraints, the normalized IC50 predicted values ​​of the mutant peptides are divided into different affinity levels. Based on the mutant and wild-type peptide-MHC binding capacity results obtained in S4, candidate constraints are applied, and they are respectively classified into high-binding-capacity candidates (…). Candidates for general bonding strength ( ) and weak binding candidates ( ).

[0065] Preferably, the binding affinity between the mutant peptide and the corresponding wild-type peptide is further compared, and the ratio is expressed as follows: Or, under the IC50 notation system: Here, ε is a non-negative small value set to avoid a denominator of zero. This differential characteristic is used to reflect whether the mutation improves the binding ability of the candidate peptide to the corresponding MHC molecule.

[0066] Preferably, multi-MHC presentation capability is further considered, and a range of multi-MHC presentation capability evaluation is performed. For multiple MHC combination records derived from the same mutant short peptide, the individual best harmonic ranking (PHBR) of candidate peptides is calculated based on the binding ranking or binding capability under different MHC alleles. In one embodiment, the ranking of candidate peptide p across all MHC alleles of the subject is... PHBR is represented as: PHBR can be used to reflect the comprehensive presentation potential of candidate peptides in an individual's MHC background. Candidate peptides that can form optimal binding with multiple MHC alleles can obtain higher candidate priority.

[0067] S5.2 Calculation of immunogenicity probability.

[0068] For candidate records that have passed the aforementioned compression phase, mutant affinity scores are further extracted. And the ratio of mutation / wild-type binding variation To reduce the skewness of the ratio data and enhance the comparability between ratios of different orders of magnitude, a natural logarithmic transformation was performed on the affinity ratios: Immunogenicity feature vectors for candidate records are constructed based on the absolute affinity characteristics of mutant peptides, the variation characteristics between mutant and wild-type peptides, and the interaction characteristics between the two: in, Used to reflect the MHC binding capacity of mutant candidate peptides themselves. Used to reflect the degree of binding change of mutant candidate peptides relative to wild-type candidate peptides. This is the interaction term between the two, used to characterize the synergistic relationship between absolute binding capacity and mutation effect.

[0069] The pre-built immunogenicity evaluation module is invoked to calculate the immunogenicity probability of candidate records based on the aforementioned feature vectors. The basic form of the probability evaluation is expressed as follows: in, For the intercept term, , and These are the model coefficients corresponding to affinity characteristics, mutation effect characteristics, and interaction characteristics, respectively. For the first The immunogenicity probability of each candidate record, ranging from 0 to 1.

[0070] Model coefficients can be obtained through pre-training with experimentally validated immunogenicity-positive and immunogenicity-negative samples. In this embodiment, the immunogenicity evaluation module is called as a functional module in the progressive compression process of the candidate space. Its training data construction, parameter solving, parameter optimization, and model validation methods can be implemented using the schemes disclosed in the applicant's prior applications, and are not considered as duplicate protection content in this case.

[0071] After calculating the immunogenicity probability, the candidate records are initially classified, and the initial classification labels for the candidate records are as follows: in, The immunogenicity probability threshold is a preset threshold. This threshold can be used for candidate ranking, candidate stratification, or candidate eligibility assessment, but it is not limited to using a fixed probability threshold for binary classification. In this embodiment, =0.5 Candidate records with an initial classification label of 1 will proceed to the subsequent rule correction and multi-dimensional feature aggregation stage; candidate records with an initial classification label of 0 can be removed from the current candidate space or retained as low-priority candidates depending on the actual application purpose.

[0072] S5.3 Multidimensional Candidate Information Aggregation and Candidate Space Refinement Screening Preferably, the candidate space is finely screened based on the immunogenicity evaluation results and comprehensive candidate information. The comprehensive candidate information of the i-th candidate record is represented as: The labels for candidate records are represented as follows: in, This represents the rule correction function. Indicates weak binding, mutation wild-type similarity, or other anomalous status markers for candidate records.

[0073] Preferably, for multiple candidate records belonging to the same mutant candidate peptide, the same mutation event, or the same source gene, they are aggregated, sorted, and deduplicated according to candidate priority, retaining one or more representative candidate objects, thereby reducing redundant records formed by different MHC alleles, different transcripts, and duplicate annotations.

[0074] S5.4 Implementation Results.

[0075] This embodiment compared spatial compression with the Seq2Neo full-process neoantigen prediction tool, and the specific data are shown in Table 1. Table 1 shows that this embodiment and Seq2Neo obtained completely consistent results in somatic mutation detection, VEP functional annotation, expression-supporting missense mutation screening, and mutation site deduplication stages, indicating that the two methods share the same upstream mutational basis. This embodiment further generated 10,198 candidate short peptides of 8–11 amino acids based on 146 missense mutations, and combined them with 6 MHC-I alleles to form 61,188 candidate peptide-MHC records. After evaluating peptide-MHC binding capacity, the number of candidate records remained unchanged, but each candidate record added information on mutant and wild-type binding capacity; after basic biological filtering, 35,377 records were retained, and further multi-stage candidate spatial compression yielded 67 high-confidence candidates. Compared to the 361 candidates ultimately output by Seq2Neo, this embodiment reduces the final number of candidates by approximately 81.44%, and the candidate size is approximately 18.56% of that of Seq2Neo. This demonstrates that the present invention can significantly compress the candidate neoantigen space while maintaining the same upstream mutation input, providing a technical basis for reducing the cost of subsequent experimental verification.

[0076] Table 1 Table 2 shows a comparison of the prediction results of the two methods with the experimentally validated mutation candidates in the P4166 sample in the patients of this embodiment. Of the 67 candidate peptides generated in this embodiment, two short candidate peptides containing validated mutation sites were identified, ranking 19th and 22nd respectively, both within the top 30 of the candidate list. Of the 361 candidate peptides generated by Seq2Neo, only one short candidate peptide containing a validated mutation site was identified, ranking 63rd and not within the top 30. Using the two validated candidates in this sample as a reference, the overall recall and Top 30 recall of this embodiment were both 100%, while the overall recall of Seq2Neo was 50%, and the Top 30 recall was 0%. In an application scenario where the top 30 candidates are consistently selected for experimental validation, this embodiment can cover all validated candidates, while Seq2Neo cannot. This demonstrates that while significantly compressing the candidate space, the present invention can still retain candidates with validated evidence within a higher experimental priority range.

[0077] Table 2 .

[0078] S6. Organize and prioritize the results of the compressed candidate neoantigen set to generate a high-confidence neoantigen candidate set.

[0079] After progressive candidate space compression, the results of the candidates are sorted and prioritized to generate a high-confidence neoantigen candidate set.

[0080] The output retains the mutation information, expression information, functional annotation information, candidate peptide information, MHC typing information and binding capacity evaluation results generated by the candidate objects throughout the analysis process. It also sorts the candidate objects based on the comprehensive biological characteristics formed during the progressive compression process, providing candidate basis for subsequent immunological experimental validation, personalized tumor vaccine design and clinical application.

[0081] Preferably, the output results include mutation information corresponding to the candidate neoantigen, mutant peptide and wild peptide sequences, MHC alleles, binding affinity evaluation results, VAF, TPM, Biotype, PHBR and other auxiliary information used for candidate evaluation.

[0082] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for discovering new antigens based on multi-omics data and progressive candidate screening, characterized in that, include: S1. Obtain multi-omics sequencing data from the subjects, perform mutation detection, expression quantification, functional annotation and MHC typing, and obtain mutation information, annotation information, expression information and MHC allele information; S2. Based on the mutation information and annotation information, construct mutant candidate short peptides and corresponding wild-type candidate short peptides, and combine the candidate short peptides with the MHC alleles to obtain candidate peptide-MHC combinations. S3. Associate the mutation information, annotation information, expression information, MHC allele information with the candidate short peptides to construct a candidate neoantigen analysis library; S4. Evaluate the MHC binding capacity of mutant and wild-type candidate short peptides in the candidate neoantigen analysis library, and write the evaluation results back to the corresponding candidate records to form an enhanced candidate neoantigen analysis library. S5. Based on the enhanced candidate neoantigen analysis library, the candidate space is progressively compressed to obtain a compressed candidate neoantigen set; S6. Organize the results and prioritize the compressed candidate neoantigen set to generate a high-confidence neoantigen candidate set.

2. The neoantigen discovery method based on multi-omics data and progressive candidate screening according to claim 1, characterized in that, S1 specifically includes: S1.1 Obtain raw sequencing data of tumor tissue DNA, paired normal tissue DNA and tumor tissue RNA from the subjects, and perform quality control; S1.2 Align the quality-controlled tumor DNA and normal DNA sequencing reads to the reference genome to obtain tumor tissue DNA alignment files and normal tissue DNA alignment files; S1.3 Using the normal tissue DNA alignment file as a pairing control, perform somatic mutation detection on the tumor tissue DNA to obtain a somatic mutation set; S1.4 Perform quality control on the RNA sequencing data of the tumor tissue and align it to the reference genome, perform expression quantification, and obtain RNA expression level information; S1.

5. Perform functional annotation on the somatic mutation detection results obtained in S1.3 to obtain transcript annotation information; S1.

6. Using the RNA sequencing data of the tumor tissue, MHC-I allele typing is performed to obtain the subject's MHC allele set.

3. The neoantigen discovery method based on multi-omics data and progressive candidate screening according to claim 2, characterized in that, S2 specifically includes: S2.1 Based on the protein position, reference amino acid, and mutant amino acid in the transcript annotation information obtained in S1.5, determine the position of the mutation site in the protein sequence, extract a predetermined number of amino acids centered on the mutation site, and construct mutant local protein fragments and wild-type local protein fragments respectively. S2.2 Using a sliding window method, candidate short peptides containing mutated amino acids are generated from the mutant local protein fragments, and wild-type short peptides corresponding to the positions are generated simultaneously to obtain mutant candidate short peptides and corresponding wild-type candidate short peptides. S2.

3. Combine each candidate short peptide record with all MHC alleles obtained in S1.6 to construct a candidate peptide-MHC combination space.

4. The neoantigen discovery method based on multi-omics data and progressive candidate screening according to claim 3, characterized in that, In S3, the following combined identifiers are used to determine a peptide-MHC candidate record: Chrom, Pos, Ref, Alt, Transcript, MT_SHORT_PEPTIDE, and MHC; for the same mutation formed under different transcripts, different peptide windows, different peptide lengths, or different MHC alleles, the candidates are saved as independent records.

5. The neoantigen discovery method based on multi-omics data and progressive candidate screening according to claim 1, characterized in that, In step S4, the binding affinity assessment step includes: obtaining the predicted MHC binding affinity of mutant peptides and wild-type peptides respectively; after the prediction is completed, writing the binding affinity results of mutant and wild-type peptides back to the corresponding candidate records to form an enhanced candidate neoantigen analysis library.

6. The method for discovering new antigens based on multi-omics data and progressive candidate screening according to claim 1, characterized in that, In step S5, the progressive compression includes: S5.

1. Based on the frequency of mutated alleles, expression level, filter field and sequencing depth information, the basic credibility of candidate records is judged, and the functional validity of candidate records is judged in combination with transcript function type constraints, thus completing the biological characteristic constraints. S5.2 Extract the mutant peptide-MHC binding affinity characteristics, the binding change ratio between mutant and wild type characteristics, and the interaction characteristics between the two from the candidate records after biological feature constraints, construct an immunogenicity feature vector, and call the pre-constructed immunogenicity evaluation module to calculate the immunogenicity probability of the candidate records. S5.

3. Based on the immunogenicity probability and candidate comprehensive information, perform fine screening of the candidate space to obtain a compressed set of candidate neoantigens.

7. The method according to claim 1, characterized in that, In S6, the output results include at least: mutation information corresponding to the candidate neoantigen, mutant peptide and wild peptide sequences, MHC alleles, binding affinity evaluation results, VAF, TPM and Biotype.

Images