An open search method based on open HLA pan-modified immune peptide

Abstract

The application discloses an open retrieval method based on OpenHLA pan-modification immunopeptide. The method comprises the following steps: acquiring immunopeptidomics mass spectrum data and preprocessing; constructing a target database, a decoy database and a modification rule set; performing de novo analysis on MS / MS spectrum; constructing a Tag index based on the de novo analysis result; establishing an ion index; under the database retrieval framework, fusing the de novo information, the Tag index and the ion index to generate an open modification candidate peptide spectrum matching result; performing initial scoring on the candidate result; re-scoring by a Percolator re-scoring module specially adapted for immunopeptide open modification search; and performing FDR control based on a target-decoy strategy, and outputting an immunopeptide open modification identification result. The application can support a large number of modification searches while controlling the complexity of the candidate space, and improve the sensitivity, accuracy and computing efficiency of HLA immunopeptide open modification identification.

Description

I. Technical Field

[0001] This invention relates to the field of immunopeptidomics mass spectrometry data analysis technology, and more particularly to a method for retrieving open modification databases for HLA-presented peptide data. More specifically, this invention relates to a retrieval method that, in non-enzymatic immunopeptide database retrieval scenarios, integrates de novo sequence information, tag indexes, ion indexes, open modification candidate generation, and a dedicated rescoring model to improve the sensitivity, accuracy, and computational efficiency of modified immunopeptide identification. II. Background Technology

[0002] (a) Current state of technology Immunopeptidomics primarily studies peptides presented to the cell surface by major histocompatibility complex-associated HLA molecules. These peptides typically exhibit characteristics such as being non-enzymatically cleaved, having a wide length distribution, originating from complex proteins, possessing diverse modification types, and exhibiting high sample specificity. Mass spectrometry is a crucial technique for identifying HLA-presented peptides, and database searching is one of the most commonly used computational methods in mass spectrometry data analysis.

[0003] Existing methods for retrieving immune peptide databases typically rely on reference protein databases. They generate candidate peptides under conditions of limited peptide length, quality tolerance, and a small number of variable modifications, and then match and score the theoretical fragment ions of the candidate peptides with experimental MS / MS spectra. These methods offer high computational efficiency when the number of modifications is limited and the candidate space is well-controlled.

[0004] Traditional open search or large mass shift search methods relax precursor quality difference constraints to discover unknown or unmodified peptides. De novo sequencing methods can directly infer local sequence fragments from MS / MS spectra, providing potential sequence clues for database retrieval. Re-scoring methods such as Percolator typically re-rank candidate peptide profile matching results based on target-decoy strategies and machine learning models to improve the ability to distinguish between target and decoy peptides.

[0005] (ii) Problems with existing technology First, conventional immune peptide database searches typically only support a limited number of variable modifications. When a large number of modifications are introduced simultaneously, the number of candidate peptide combinations increases rapidly, easily leading to combinatorial explosion, which significantly increases search time, memory consumption, and the risk of false positives.

[0006] Second, traditional open search methods often focus on precursor quality offset matching, failing to fully utilize local sequence information, de novo fragment information, and fragment ion index information in the immune peptide scenario, resulting in insufficient candidate space pruning efficiency.

[0007] Third, HLA-presented peptides typically do not conform to the classic rules of trypsin-digested peptides, exhibiting characteristics such as non-digestion, large length variations, and inconsistent modification sites. Directly applying conventional proteomics candidate generation and scoring logic can easily lead to insufficient candidate generation, mismatched scoring features, or decreased reliability of results.

[0008] Fourth, general percolator or general rescoring module is usually built based on conventional proteomics feature system, without fully considering factors such as a large number of modification candidates, special candidate generation sources, modification localization evidence, tag support and ion coverage features in immune peptide open modification search, which leads to limited target and decoy separation ability.

[0009] Therefore, a method specifically designed for searching open modifications of HLA immune peptides is needed, which can expand the modification search space while controlling candidate complexity, improving search efficiency, and enhancing identification accuracy through a rescoring mechanism adapted to the open modification features of immune peptides. III. Summary of the Invention

[0010] (a) Purpose of the invention The purpose of this invention is to provide an open search method based on OpenHLA panmodified immune peptides to solve the problems of limited modification search capability, candidate space expansion, low search efficiency, insufficient adaptation of general scoring models, and limited identification reliability in existing technologies for open modification search of HLA immune peptides.

[0011] This invention achieves a balance between search depth, identification accuracy, and computational efficiency by integrating de novo sequence information, tag index, and ion index within a database retrieval framework, and by constructing a dedicated Percolator rescoring module for Open HLA open modified immune peptide scenarios.

[0012] (II) Technical Solution To achieve the above objectives, this invention provides an open retrieval method based on OpenHLA panmodified immune peptides, the method comprising the following steps: S1: Acquire and preprocess immunopeptidomics mass spectrometry data. The mass spectrometry data may include raw mass spectrometry files, mzML files, mgf files, or other spectral data that can be used for database retrieval. Preprocessing may include spectral denoising, peak screening, precursor mass correction, fragment ion mass correction, spectral quality control, peak intensity normalization, and format standardization suitable for subsequent de novo analysis and database retrieval.

[0013] S2: Construct a target database, a decoy database, and a set of modification rules. The target database may include a reference protein database, a sample-specific protein database, a variant protein database, a transcriptome translation database, or a combination thereof. The decoy database can be obtained through reversal, randomization, or other decoy generation strategies. The set of modification rules may contain multiple variable modifications, which can be organized according to rules such as amino acid site, peptide N-terminus or C-terminus, protein N-terminus or C-terminus, mass offset range, chemical category, and whether combined modifications are allowed.

[0014] S3: Perform de novo analysis on MS / MS spectra. For each MS / MS spectrum, obtain one or more candidate sequence fragments, local amino acid tags, amino acid quality pathways, local confidence scores, or reliable position fragments. The de novo results do not directly replace the database search results, but rather serve as prior constraints for subsequent candidate recall, candidate ranking, and candidate selection.

[0015] S4: Construct or extract Tag information based on de novo results and establish a Tag index. The Tag can be a continuous amino acid fragment, a local sequence pattern allowing partial quality tolerance, a modification-tolerant sequence tag, or a combination thereof. Using the Tag index, candidate regions that are consistent with or approximately consistent with the local sequence in the spectrum can be quickly recalled from the database, thus narrowing the search scope of the entire library.

[0016] S5: Establish an ion index. The ion index pre-encodes and organizes candidate peptides, candidate fragments, or theoretical fragment ions from the database. During spectral matching, potential candidate peptides, candidate fragments, or modification sites can be quickly located based on fragment ion peaks observed in experimental spectra. The ion index supports b-ions, y-ions, neutral loss ions, ions with different charge states, and modification-related shift ions.

[0017] S6: Integrate de novo information, tag index, and ion index to generate open modification candidate peptides. Within the database retrieval framework, de novo results and tag indexes are first used to recall candidate regions consistent with local sequences in the spectrum. Then, ion indexes are used for rapid matching, scoring, and screening. Subsequently, in a large modification space, candidate peptide modification assignment, modification localization, and theoretical spectrum calculation are performed to obtain open modification candidate peptide spectrum matching results. The candidate recall order is not limited to the above order; ion index-based recall, multi-path parallel recall, or staged recall methods can also be used.

[0018] S7: Initial scoring of open-modification candidate peptides. The initial score can comprehensively consider factors such as precursor quality error, number of fragment ion matches, explanatory power of fragment ion intensity, length of continuous ion coverage, rationality of modification localization, consistency of modification quality shift, de novo consistency, tag matching consistency, ion index recall strength, candidate peptide length, candidate peptide charge state, and candidate generation source.

[0019] S8: Construct and apply a Percolator rescoring module specifically adapted to Open HLA. The rescoring module designs a feature set based on the open modification search characteristics of immune peptides, and uses a target-decoy strategy to train or apply a classification model to rescore candidate PSMs, peptides, and / or modification sites, outputting a discriminant score for result ranking.

[0020] S9: Perform FDR control and output results. Based on the target-decoy strategy, threshold control is applied to the results at the PSM level, peptide level, protein source level, modification level, and / or modification site level, outputting the final immune peptide open modification recognition results. Output results may include spectrum number, candidate peptide sequence, modification type, modification location, quality shift, initial score, rescore, FDR or q-value, target / decoy label, and candidate generation source.

[0021] Furthermore, the input features of the Percolator rescoring module specifically adapted to Open HLA may include at least one or more of the following: number of open modification candidates, modification location features, non-enzyme digestion sequence features, de novo support, tag support, ion coverage, precursor quality error, fragment quality error, number of modifications, modification category, candidate peptide length, candidate peptide charge state, target / decoy distribution related features, and candidate generation source features.

[0022] Furthermore, the method may also include clustering and merging the modification quality shift, performing two-stage fine screening of candidate peptides, grouping and scoring according to peptide length range or modification category, constraining the stability of the Percolator rescoring results, using multi-model fusion to rank the results, or combining sample HLA typing information to perform post-processing filtering of candidate immune peptides.

[0023] (III) Beneficial Effects Improved modification search capabilities. Compared to conventional immune peptide library search methods that only support limited modifications, this invention can identify HLA-presenting peptides within a larger modification space, which is beneficial for discovering more modified immune peptides.

[0024] Improve search efficiency. By combining de novo results, tag indexes, and ion indexes, the search space can be expanded while reducing invalid candidate comparisons and lowering computational complexity.

[0025] Improve the adaptability of immune peptides to various scenarios. This invention designs candidate generation and scoring features specifically for the characteristics of HLA peptides, such as non-enzymatic digestion, large length variations, and diverse modifications, making it more suitable for immune peptide data analysis than conventional proteomics workflows.

[0026] Improve identification accuracy. The dedicated Percolator rescoring module can leverage features such as open modifier candidates, modifier localization, tag support, ion coverage, and de novo consistency to enhance the separation of target and decoy.

[0027] Balancing search depth and computational cost, this invention achieves deeper open modification search without significantly sacrificing computability through multi-level candidate constraints and indexing strategies.

[0028] It has good scalability. This invention can be extended to different HLA categories, different species databases, different modification rule sets, different mass spectrometry platforms, and different data analysis workflows. IV. Description of the attached drawings

[0029] Figure 1 This is a schematic diagram of the overall process of an open search method for panmodified immune peptides based on OpenHLA according to the present invention.

[0030] Figure 2 This is a schematic diagram illustrating the process of generating candidate peptides in collaboration with de novo information, tag index, and ion index in this invention.

[0031] Figure 3 This is a schematic diagram of the process for rescoring and FDR control using the dedicated Percolator in this invention.

[0032] It should be noted that the above figures are only used to illustrate the main process and information relationships of the present invention, and do not constitute a limitation on the specific implementation order, module deployment method or parameter range. V. Detailed Implementation Methods

[0033] (I) Method Implementation Examples The present invention will be described below with reference to exemplary embodiments. It should be understood that the following embodiments are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0034] In one embodiment, immunopeptidomics MS / MS spectral data are acquired and preprocessed. Preprocessing may include removing low signal-to-noise ratio peaks, retaining peaks with high intensity ranking, correcting precursor and fragment ion masses based on calibration peptides or global mass error distribution, filtering low-quality spectra, and standardizing the data to a structure suitable for subsequent de novo analysis and database retrieval.

[0035] In one embodiment, a target database and a decoy database are constructed. The target database may consist of a human reference protein database, a sample-specific transcriptome translation database, a tumor mutant peptide database, or a combination thereof. The decoy database can be obtained by reversing or randomizing the target sequence. A set of modification rules is also constructed, which includes multiple variable modifications and corresponding constraints such as site restrictions, endpoint restrictions, quality shifts, whether combinations are allowed, and the maximum number of modifications.

[0036] In one embodiment, de novo analysis is performed on each MS / MS spectrum to obtain local sequence fragments, amino acid quality pathways, and their confidence levels. Tags that meet the confidence thresholds are extracted from the de novo results and used as candidate recall clues. Local fragments with uncertain quality shifts can be represented as fuzzy tags with quality tolerance.

[0037] In one embodiment, a tag index is established. The tag index is used to quickly map local sequence tags to candidate regions in the target database. If multiple tags are obtained from the de novo results, the union, intersection, or weighted merging of the recall results from different tags can be performed to obtain candidate database regions.

[0038] In one embodiment, an ion index is established. This ion index is used to correlate theoretical fragment ion masses with candidate peptides or fragments. Inverted tables or hierarchical mass buckets can be constructed based on theoretical b / y ions, neutral loss ions, ions with different charge states, and modification-related shift ions to support rapid matching of experimental fragment ion peaks.

[0039] In one embodiment, de novo information, tag indexes, and ion indexes are fused to generate candidate peptides. Specifically, the tag index is first used to recall candidate database regions that match the local sequence of the spectrum, and then the ion index is used to perform rapid fragment ion matching on the recalled regions to obtain a set of candidate peptides. Subsequently, modification assignment and modification localization are performed on the candidate peptide set according to a set of modification rules to generate open-modification candidate peptide spectrum matching results.

[0040] In one embodiment, an initial score is performed on open-modification candidate peptides. This initial score may include quality error scoring, fragment matching scoring, continuous ion coverage scoring, intensity interpretation scoring, tag consistency scoring, de novo support scoring, modification localization scoring, and ion index recall scoring. The top few candidate PSMs for each spectrum can be retained based on the initial score for subsequent rescoring.

[0041] In one embodiment, candidate PSMs are input into a dedicated Percolator rescoring module. This dedicated Percolator rescoring module uses a feature set designed for open modification search of immune peptides, including but not limited to candidate peptide length, non-enzyme-digested sequence features, number of open modification candidates, modification quality offset, modification localization score, tag support, denovo consistency, ion coverage, precursor quality error, fragmentation quality error, candidate generation source, and target / decoy markers. The model outputs a rescoring score for each candidate PSM.

[0042] In one embodiment, FDR (Free Detection and Reduction) is controlled at the PSM, peptide, or modification site level according to a target-decoy strategy, and the results of open modification identification of immune peptides that meet the threshold requirements are output. The output results can be further used for modification profile statistics, HLA binding prediction, antigen presentation analysis, or downstream biological interpretation.

[0043] (ii) Alternative implementation methods Without departing from the purpose of this invention, the Tag index may employ tags of different lengths, fuzzy tags that allow mismatches, local sequence indexes based on graph structures, or other local segment retrieval structures.

[0044] Without departing from the purpose of this invention, the ion index may employ a theoretical spectrum inverted index, a hierarchical ion bucket index, a fast ion retrieval structure based on feature hashing, or a dedicated index structure for specific ion types.

[0045] Without departing from the purpose of this invention, de novo auxiliary information may be de novo sequence fragments, denovo confidence scores, local quality paths, local amino acid set constraints, or combinations thereof.

[0046] Without departing from the purpose of this invention, the dedicated Percolator rescoring module can also be replaced with other supervised learning models, semi-supervised learning models, ranking learning models or ensemble learning models, but its input feature system should be adapted to the search results of open modification of immune peptides.

[0047] Without departing from the purpose of this invention, the order of the candidate generation process can be adjusted. For example, a coarse recall can be performed first using the ion index, followed by a fine screening using tag and de novo information; alternatively, a multi-path parallel recall can be used followed by a unified summary score.

[0048] Without departing from the purpose of this invention, FDR control can be implemented at the PSM level, peptide level, modification level, or modification site level, depending on the application requirements.

Claims

1. An open search method based on OpenHLA panmodified immune peptides, characterized in that, include: Acquire and preprocess immunopeptidomics mass spectrometry data; construct a target database, a decoy database, and a set of modification rules; perform de novo analysis on MS / MS spectra to obtain candidate sequence fragments, local tag sequences, amino acid mass pathways, or reliable position fragments; construct or extract tag information based on the de novo analysis results and establish a tag index; establish an ion index, which is used to locate potential candidate peptides, candidate fragments, or modification position regions based on experimental fragment ion peaks; fuse the de novo analysis results, the tag index, and the ion index to generate open modification candidate peptide spectrum matching results; perform initial scoring on the open modification candidate peptide spectrum matching results; perform rescoring using the Percolator rescoring module specifically adapted for immunopeptide open modification search; and perform FDR control based on a target-decoy strategy to output immunopeptide open modification identification results.

2. The method according to claim 1, characterized in that, The preprocessing includes one or more of the following: spectrum denoising, peak screening, precursor quality correction, fragment ion quality correction, spectrum quality control, peak intensity normalization, and format standardization.

3. The method according to claim 1, characterized in that, The set of modification rules includes multiple variable modifications, and these multiple variable modifications are organized according to constraints such as amino acid site, peptide endpoint, protein endpoint, mass offset range, chemical category, or combination of modifications.

4. The method according to claim 1, characterized in that, The tag information includes continuous amino acid fragments, local sequence patterns with quality tolerance, modification-tolerant sequence tags, or combinations thereof.

5. The method according to claim 1, characterized in that, The Tag index is used to recall candidate database regions from the target database that are consistent with or approximately consistent with the local sequence of the spectrum.

6. The method according to claim 1, characterized in that, The ion index is constructed based on theoretical fragment ions, which include one or more of b ions, y ions, neutral loss ions, ions with different charge states, and modification-related shift ions.

7. The method according to claim 1, characterized in that, The generation of open-modification candidate peptide spectrum matching results includes: recalling candidate regions using de novo analysis results and tag indexes; performing rapid matching, scoring, or screening of candidate regions using ion indexes; and performing modification assignment, modification localization, and theoretical spectrum calculation on candidate peptides under the constraints of the modification rule set.

8. The method according to claim 1, characterized in that, The initial score is determined based on one or more of the following: precursor quality error, number of fragment ion matches, fragment ion intensity explanatory power, continuous ion coverage length, modification localization rationality, modification quality offset consistency, de novo consistency, tag matching consistency, ion index recall intensity, candidate peptide length, candidate peptide charge state, and candidate generation source.

9. The method according to claim 1, characterized in that, The input features of the Percolator rescoring module, which is specifically adapted for searching open modifications of immune peptides, include one or more of the following: number of open modification candidates, modification location features, non-enzyme digestion sequence features, de novo support, tag support, ion coverage, precursor quality error, fragment quality error, number of modifications, modification category, target / decoy distribution related features, and candidate generation source features.

10. The method according to claim 1, characterized in that, The FDR control is performed at the PSM level, peptide level, protein source level, modification level, or modification site level.

Images