Protein genomics analysis method and system for discovering new coding regions

Abstract

The invention provides a protein genomics analysis method for discovering a new coding region, which comprises the following steps: respectively carrying out DNA six-frame translation on a genome sequence and RNA three-frame translation on a transcriptome sequence, and merging corresponding translation results to construct a customized database; searching the mass spectrum data through a multi-step database search strategy to obtain a new peptide fragment identification result; mapping the identified new peptide fragment to a potential expression region of a genome through a sequence tag indexing method to generate a CDS candidate; filtering the CDS candidates by using a gene scoring model based on deep learning for the first type of species; and filtering the CDS candidates by using a filter based on the multi-stage CDS mapping group aiming at the second type of species so as to reconstruct a complete gene coding region structure. The invention also provides a proteogenomics analysis system for discovering the new coding region. Therefore, the coverage degree and the accuracy of new coding area discovery can be greatly improved, and meanwhile, the calculation efficiency is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics technology, and in particular to a protein genomics analysis method and system for discovering new coding regions. Background Technology

[0002] In recent years, proteogenomics has developed rapidly as an interdisciplinary field integrating proteomics, transcriptomics, and genomics. While genomics provides a fundamental blueprint of genetic information, its annotations are often incomplete or inaccurate, failing to fully reflect the exact protein products generated after transcription and translation. Traditional tools can infer gene structure based on annotation information from closely related species or transcriptome data, but such predictions are usually uncertain. In contrast, proteogenomics employs a novel strategy that overcomes the limitations of prediction-based methods; it utilizes protein-level evidence from mass spectrometry analysis to infer new coding sequences (CDSs), becoming the gold standard for verifying new coding events. This method has enormous potential for improving the accuracy of gene annotation, confirming translation products, and advancing our understanding of complex biological processes. Mass spectrometry technology is rapidly developing, and next-generation instruments can provide deeper proteomic coverage and generate various high-quality whole-proteome datasets. However, due to the lack of effective analytical tools, the wealth of valuable information contained in these datasets remains underutilized. Currently, proteogenomics has been successfully applied to genome reannotation, identification of short open reading frames (sORFs, ≤300 nucleotides), and identification of non-classical HLA-binding peptides.

[0003] Similar to traditional proteomics, the core steps of proteogenomics still rely on mass spectrometry-based peptide and protein identification, but there are three key differences.

[0004] First, the construction of customized databases. Databases for proteogenomic analysis should not only include reference proteins, but should also cover as many potential novel events as possible (such as proteins derived from previously annotated non-coding regions).

[0005] Second, large-scale database searches. Compared to traditional proteomics searches of reference databases, customized databases expand in size by tens of times. Furthermore, it requires the integration of information from post-translational modifications, alternative splicing, and mutations, further expanding the search space. This puts pressure on search engines, significantly reducing the accuracy, sensitivity, and speed of identification.

[0006] Third, new coding sequences can be inferred from novel peptides, thus supporting downstream biological analyses. Although new peptides are identified, the prevalence of repetitive sequences and codon degeneracy in the genome means that a single new peptide can correspond to multiple new coding sequences. This problem is particularly pronounced in eukaryotes, where widespread splicing further complicates CDS inference. For example, over 98% of the regions in the human genome do not encode proteins, leading to ambiguity in the interpretation of gene expression.

[0007] While proteogenomics has shown great potential in gene annotation and functional analysis, its technical processes still face challenges such as decreased sensitivity due to the expanding search space, increased computational resource requirements, and ambiguous interpretation of results. There is an urgent need to optimize algorithms and strategies to improve analytical efficiency and reliability. In particular, the lack of comprehensive, end-to-end analytical tools makes related analyses in biological laboratories difficult, hinders the thorough extraction of data value, and results in a waste of human and economic resources.

[0008] Currently, proteogenomics data analysis faces multiple technical bottlenecks, directly impacting its accuracy, coverage, and computational efficiency. Regarding custom database generation, existing methods such as Augustus, Balrog, and SpliceAI primarily rely on coding region prediction. This strategy has inherent flaws, systematically missing short open reading frames (sORFs) and multi-frame translation events, which are precisely the key targets for detection in proteogenomics. This limitation stems from the inherent design philosophy of traditional tools based on conserved gene structure models, making it difficult to effectively identify atypical translation events.

[0009] To address the database coverage problem, existing technologies have developed hybrid strategies, such as six-frame genome translation combined with three-frame RNA-seq translation. While this approach can improve genome coverage, it also introduces a large number of artificial sequences, resulting in a massive database containing excessive redundant information. This not only reduces search accuracy but also significantly increases the computational burden. The root cause of this problem lies in the lack of effective means for refining the database size in existing methods.

[0010] In the database search phase, existing technologies employ iterative strategies such as dynamic database optimization and hierarchical spectrogram analysis to mitigate the performance degradation caused by large databases. However, these methods often require a trade-off between sensitivity and computational resource consumption, making it difficult to achieve both simultaneously. This is mainly because existing search strategies fail to achieve targeted optimization of the search space and lack the ability to collaboratively analyze multidimensional information such as mutations and alternative splicing.

[0011] In the process of inferring new CDS from novel peptides, existing tools such as JUMPg, PGTools, and NextSearch, while capable of matching peptides to the genome or pre-constructed spliceosomes, have significant shortcomings in handling redundant matches, and the method of pre-constructing spliceosomes is prone to missing atypical splicing phenomena. The GAPE tool directly discards peptides that can match multiple positions, while IPAW requires high-confidence common peptides to support the report of novel spliced ​​peptides. While this conservative strategy can reduce false positives, it can also lead to the omission of true results.

[0012] Furthermore, existing technologies suffer from insufficient integration across the entire process. Currently, proteogenomic analysis tools are mostly fragmented, independent modules, lacking a complete end-to-end solution. More importantly, the industry has not yet established systematic evaluation standards for the quality of CDS reports, which severely affects the comparability and reproducibility of different research results. Summary of the Invention

[0013] To address the problems existing in the prior art, this invention proposes a protein genomics analysis method and system for discovering new coding regions, which can significantly improve the coverage and accuracy of new coding region discovery, while also significantly improving computational efficiency.

[0014] To achieve the aforementioned technical effects, on the one hand, this invention provides a protein genomics analysis method for discovering new coding regions, comprising the following steps:

[0015] The genome sequence was translated into DNA six-frames and the transcriptome sequence into RNA three-frames, and the translation results were merged to construct a custom database.

[0016] A multi-step database search strategy is used to search the mass spectrometry data to obtain new peptide identification results. Specifically, the multi-step database search strategy is as follows: based on the mass spectrometry data, a reference database is searched to obtain reliable peptides; unresolved spectra are used to search the customized database to retain the proteins corresponding to the identified peptides to form a refined database; and the mass spectrometry data is then used to search the refined database to identify new peptides.

[0017] The identified novel peptides are mapped to potential expression regions of the genome using a sequence tag indexing method to generate CDS candidates;

[0018] For the first category of species, a gene scoring model based on deep learning is used to filter the CDS candidates; wherein, the first category of species are those in which splicing generally does not occur.

[0019] For the second type of species, a filter based on a multi-stage CDS mapping group is used to filter the CDS candidates in order to reconstruct the complete gene coding region structure; wherein, the second type of species is a species in which splicing is common.

[0020] Furthermore, the step of searching a reference database based on the mass spectrometry data to obtain reliable peptides specifically involves:

[0021] The reference database was searched using all mass spectrometry data, and identification results with a false discovery rate of no more than 1% at the peptide level were considered reliable peptides; the unresolved spectra were mass spectra that were not identified in the search.

[0022] Furthermore, the step of using the mass spectrometry data to search the refined database to identify new peptides specifically includes:

[0023] The refined database is searched using the entire dataset of the mass spectrometry data to obtain peptide identification results;

[0024] Based on whether the peptide identification results are in the reference database, they are distinguished as new peptides and old peptides, and the FDR estimation is performed on the two types of peptides, while retaining results with a false discovery rate of less than 1%.

[0025] Furthermore, the step of mapping the identified novel peptides to potential expression regions of the genome using a sequence tag indexing method to generate CDS candidates specifically includes:

[0026] S31. The identified new peptides are mapped to the genome according to the priority of ordinary mapping, splicing mapping, and mutation mapping, using the sequence tag indexing method;

[0027] S32. Based on the mapping position, extend to the 5' end and the 3' end to generate CDS candidates.

[0028] Furthermore, the step of extending from the mapping position to the 5' and 3' ends to generate CDS candidates also includes:

[0029] For the first type of species, starting from the upstream peptide mapping position and extending towards the 5' end, the starter is searched according to the preset start codon priority.

[0030] Furthermore, the deep learning-based gene scoring model is a multilayer perceptron (MLP) model; the filtering of the CDS candidates using the deep learning-based gene scoring model for the first species includes:

[0031] S41. Extract the identification results of the reliable peptide, map them to the genome, and complete the inference of the starter;

[0032] S42. Extract training samples from the genome mapping results; wherein, the positive samples of the training samples are mapping results supported by genome annotation, and the negative samples include mapping results without genome annotation support and mapping results of decoy peptides.

[0033] S43. Extract CDS length, peptide score, amino acid coverage, and initiator type features from the training samples to train the multilayer perceptron (MLP) model.

[0034] S44. Use the trained multilayer perceptron (MLP) model to score the CDS candidates and filter them based on the scores.

[0035] Furthermore, for the second type of species, filtering the CDS candidates using a filter based on a multi-stage CDS mapping group to reconstruct the complete gene coding region structure includes:

[0036] S51. The new peptide is mapped back to the protein translated from RNA based on the sequence tag indexing method;

[0037] S52. Construct a first index table set, where the key is the number of the RNA-translated protein and the value is a Gödel code list covering the peptide segments of that protein.

[0038] S53. Construct a second index table set, where the key is the Gödel code of the new peptide and the value is a list of protein IDs covering the new peptide on the DNA strand.

[0039] S54. Based on the first index table and the second index table, determine whether the DNA strand potentially contains a region expressing RNA protein.

[0040] S55. Use a depth-first search algorithm to verify whether the CDS candidate satisfies the consistency between the order of peptide segments from the N-terminus to the C-terminus on the RNA protein and the order from the 5' end to the 3' end on the DNA strand, and add the target candidate CDS that satisfies the order consistency verification to the final result list.

[0041] S56. Gradually reduce the judgment threshold ratio and repeat the above steps S54 to S55 until the candidate combinations obtained in each step are added to the final result list.

[0042] Furthermore, the step of determining whether a DNA strand potentially contains a region expressing an RNA protein based on the first index table set and the second index table set specifically includes:

[0043] Read the peptide codes covered by the RNA-translated protein from the first index table set;

[0044] Calculate whether the proportion of the peptide encoding that exists in the second index table exceeds the determination threshold proportion; if so, determine that the DNA chain potentially contains a region expressing RNA protein.

[0045] On the other hand, the present invention also provides a protein genomics analysis system for implementing the method described above, comprising:

[0046] One or more processors;

[0047] The memory stores computer-executable instructions that, when executed by the processor, are configured to perform the steps of the method.

[0048] The proteogenomic analysis method for discovering new coding regions provided by this invention effectively solves the technical problems existing in the prior art by innovatively integrating multi-frame translation strategies, developing open multi-step iterative search strategies, designing a deep learning-based gene scoring model, and a filtering algorithm based on multi-stage CDS mapping groups, thus providing a more reliable and efficient analysis tool for proteogenomic research. Attached Figure Description

[0049] Figure 1 This is a flowchart of the steps of a protein genomics analysis method for discovering new coding regions provided in an embodiment of the present invention;

[0050] Figure 2 A flowchart illustrating the steps involved in generating the CDS candidate using a protein genomics analysis method provided in an embodiment of the present invention;

[0051] Figure 3 This is a flowchart illustrating the steps of analyzing and processing a first-class species using the protein genomics analysis method provided in an embodiment of the present invention.

[0052] Figure 4 The flowchart illustrates the steps involved in analyzing and processing a second type of species using the protein genomics analysis method provided in an embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0054] It should be noted that references to "an embodiment," "embodiment," "example embodiment," etc., in this specification refer to the described embodiment including specific features, structures, or characteristics, but not every embodiment must include these specific features, structures, or characteristics. Furthermore, such expressions do not refer to the same embodiment. Moreover, when describing specific features, structures, or characteristics in conjunction with embodiments, whether or not explicitly described, it is indicated that incorporating such features, structures, or characteristics into other embodiments is within the knowledge of those skilled in the art.

[0055] Furthermore, certain terms are used in the specification and subsequent claims to refer to specific components or parts. Those skilled in the art will understand that manufacturers may use different names or terms to refer to the same component or part. This specification and subsequent claims do not distinguish components or parts by differences in name, but rather by differences in function. The terms "comprising" and "including" used throughout the specification and subsequent claims are open-ended and should be interpreted as "including but not limited to." Additionally, the term "connection" here includes any direct and indirect electrical connection means. Indirect electrical connection means include connections made through other means.

[0056] In researching protein genomics analysis methods, five key challenges were identified in existing technologies:

[0057] First, existing custom database construction methods suffer from incomplete coverage, failing to fully encompass all potential new peptides; this incompleteness of the upstream search space directly leads to the permanent loss of real peptides in downstream analysis.

[0058] Second, although it is theoretically possible to build a more comprehensive customized database, this would introduce a large number of artificial sequences, causing the search space to expand dramatically, resulting in a simultaneous decrease in the sensitivity, accuracy and speed of peptide identification.

[0059] Third, after a new peptide is identified, due to the ubiquitous repetitive sequences and codon degeneracy in the genome, the new peptide often corresponds to multiple candidate coding regions, making it difficult to accurately determine the true coding region.

[0060] Fourth, when splicing and mutation phenomena are considered, the potential search space grows exponentially, and computational speed becomes a serious bottleneck.

[0061] Fifth, there is a lack of systematic evaluation standards for the reporting quality of CDS results inferred from new peptides.

[0062] To address five key technical challenges in existing proteogenomics analysis—incomplete coverage of customized databases, performance degradation due to search space expansion, ambiguous localization of new coding regions, low computational efficiency, and lack of systematic evaluation standards—this invention constructs a proteogenomics analysis method for discovering new coding regions. This invention stems from targeted breakthroughs in addressing these challenges: First, by integrating DNA six-frame translation and RNA three-frame translation strategies, a customized database with maximum coverage of potential coding sequences is constructed, avoiding the omission of true peptides from the outset. Second, to cope with database expansion, an innovative iterative search strategy is designed, and the DBReducer tool based on sequence tag indexes is introduced for intelligent compression, significantly improving efficiency while ensuring comprehensive search coverage. For the problem of ambiguous coding region localization, a species-specific solution is proposed: developing a deep learning-based gene scoring model for simple species and designing a filter based on a multi-stage CDS mapping set for complex species. Furthermore, the computational speed bottleneck is successfully overcome through efficient Gödel coding indexing technology and a multi-level parallel strategy. Finally, a systematic CDS report simulation evaluation scheme is established.

[0063] To enable those skilled in the art to reproduce and implement the present invention, the various steps of the proteogenomic analysis method for discovering new coding regions will be described in detail below with reference to the accompanying drawings and specific embodiments. These embodiments are specific illustrations of the present invention, and their description is helpful in understanding the invention, but should not be construed as limiting the invention.

[0064] Figure 1 This invention illustrates a protein genomics analysis method for discovering new coding regions, comprising the following steps:

[0065] S101: Performs six-frame translation of DNA sequences and three-frame translation of RNA sequences from transcriptome sequences, and merges the translation results to construct a custom database. Specifically, the genomic and transcriptome sequence data can be processed through a preset database construction module to build the custom database.

[0066] The fusion strategy of DNA six-frame translation and RNA three-frame translation employed in this embodiment overcomes the limitations of traditional gene prediction tools, achieving complete coverage of the customized database. This approach maximizes the preservation of potential coding sequences while introducing complex phenomena such as splicing and mutation through transcriptomics, thus preventing the loss of correct results at the upstream of data analysis. Furthermore, evaluations on labeled datasets show that, compared to the customized databases constructed using comparison software such as Augustus, Balrog, SpliceAI, and pAnno, the coverage of labeled peptides is improved by an average of 7.6 percentage points.

[0067] S102: A multi-step database search strategy is used to search the mass spectrometry data to obtain new peptide identification results; wherein, the multi-step database search strategy is specifically as follows: based on the mass spectrometry data, a reference database is searched to obtain reliable peptides; unresolved spectra are used to search the customized database to retain the proteins corresponding to the identified peptides to form a refined database; and then the mass spectrometry data is used to search the refined database to identify new peptides.

[0068] For mass spectrometry data, this embodiment can use a multi-step database search strategy based on the Open-pFind kernel to search a customized database and obtain new peptide identification results.

[0069] Furthermore, reliable peptides are obtained by searching a reference database based on the mass spectrometry data. Specifically, the reference database is searched using all mass spectrometry data, and identification results with a false discovery rate of no more than 1% at the peptide level are considered reliable peptides. The unresolved spectra are mass spectra that were not identified in the search.

[0070] The step of using the mass spectrometry data to search the refined database to identify new peptides specifically includes: searching the refined database using the entire dataset of the mass spectrometry data to obtain peptide identification results; classifying peptides into new peptides and old peptides based on whether the peptide identification results are in the reference database, and performing FDR estimation on the two types of peptides, while retaining results with a false discovery rate of less than 1%.

[0071] Specifically, step S102 in this embodiment includes the following sub-steps:

[0072] S21, use the entire dataset to search the reference database, and consider results with a peptide-level FDR (False Discovery Rate) of less than 1% as reliable results. This step can be done using the open search mode of Open-pFind.

[0073] S22, using the unresolved spectra from step S21, search a customized database, retaining the proteins corresponding to the identified peptides, to form a new refined database. This step can be implemented using DBReducer.

[0074] S23. Use the entire dataset to search a refined database. For peptide identification results, classify them into two categories based on whether they can be found in the reference database: old peptides and new peptides. Perform FDR estimation on the two categories of peptides and retain the results within 1%. This step can be implemented based on the open search mode of Open-pFind.

[0075] This embodiment employs a unique iterative search strategy and database reduction technique. First, a preliminary search is performed using a validated reference database. Then, a secondary analysis is conducted on the unresolved spectra, minimizing the amount of mass spectrometry data directly accessed from a large, custom-designed database. In the unresolved spectrum analysis step, the DBReducer tool is introduced. This tool, based on sequence tag indexing technology, achieves intelligent database compression, resulting in a more accurate database. Finally, the entire mass spectrometry dataset is used to search the reduced database, significantly improving computational efficiency while ensuring comprehensive search coverage.

[0076] S103: The identified novel peptides are mapped to potential expression regions of the genome using a sequence tag indexing method to generate CDS candidates.

[0077] See Figure 2 In an optional implementation, step S103 includes:

[0078] S31. The identified novel peptides are mapped to the genome using sequence tag indexing, prioritizing normal mapping, splicing mapping, and mutation mapping. That is, all novel peptides are mapped to potential genomic expression sources using sequence tag indexing, prioritizing normal mapping, splicing mapping, and mutation mapping.

[0079] S32. Based on the mapping position, extend to the 5' end and the 3' end to generate CDS candidates.

[0080] Step S32 further includes: for the first type of species, starting from the upstream peptide mapping position and extending towards the 5' end, searching for a start codon according to a preset start codon priority. If the species under study is a simple species with almost no splicing, assuming the CDS is a complete protein expression region, then starting from the upstream peptide mapping position and extending towards the 5' end, searching for a start codon with a priority of ATG, GTG, CTG, TTG, AAT, and ATA.

[0081] This embodiment addresses the problem of searching large databases by designing an iterative search strategy combined with DBReducer database compression technology. This technology employs a two-stage search strategy: first searching a reference database, then searching a customized database using unresolved spectra. Combined with intelligent database compression based on sequence tag indexing, this significantly improves computational efficiency while ensuring comprehensive search coverage, effectively solving the performance degradation problem caused by database expansion. Furthermore, evaluations on labeled datasets show that the peptide identification performance of the method employed in this embodiment is significantly superior to other evaluation software (MaxQuant, Open-pFind, MS-GF+, MSFragger, and Comet). Especially on a human dataset with a 53-fold increase in database size, this invention maintains significantly higher recall rates of 98.5% and 96.0%, respectively, while the average recall rates of other search engines are 73.3% and 65.3%, respectively.

[0082] S104: For the first category of species, a deep learning-based gene scoring model is used to filter CDS candidates; where the first category of species are those in which splicing generally does not occur.

[0083] See Figure 3 In an optional implementation, the deep learning-based gene scoring model is a multilayer perceptron (MLP) model; step S104 specifically includes:

[0084] S41. Extract the identification results of the reliable peptides, map them to the genome, and complete the inference of the starter; that is, extract the peptide identification results (including target and decoy peptides) from step S21, map them to the genome, and complete the inference of the starter.

[0085] S42. Extract training samples from the genome mapping results; wherein, the positive samples of the training samples are mapping results supported by genome annotation, and the negative samples include mapping results without genome annotation support and mapping results of decoy peptides.

[0086] S43. Extract CDS length, peptide score, amino acid coverage, and initiator type features from the training samples to train the multilayer perceptron (MLP) model; that is, input the extracted feature data into the MLP-based gene scoring model to train it by distinguishing between positive and negative samples.

[0087] S44. The trained Multilayer Perceptron (MLP) model is used to score the CDS candidates, and the scores are used for filtering. Specifically, the CDS candidates obtained in step S32 are used with the same feature extraction method as in step S43, based on the trained scoring model, to obtain a score for the CDS candidates, and unreliable results are filtered out based on the score.

[0088] S105: For the second type of species, a filter based on a multi-stage CDS mapping group is used to filter the CDS candidates (specifically, low-confidence CDS candidates are filtered) to reconstruct the complete gene coding region structure; wherein, the second type of species are species in which splicing is common.

[0089] See Figure 4 In specific implementation, step S105 includes:

[0090] S51. The new peptide is mapped back to the protein translated from RNA based on the sequence tag indexing method.

[0091] S52. Construct a first index table set, where the key is the number of the RNA-translated protein and the value is a list of Gödel codes covering the peptide segments of the protein; that is, for each item in the first index table set, the key is the number of the RNA-translated protein and the value is a list recording the Gödel codes covering the peptide segments covered by the protein.

[0092] S53. Construct a second index table set, where the key is the Gödel code of the new peptide and the value is a list of protein numbers covering the new peptide on the DNA strand; specifically, construct the second index table set, with one table for each DNA strand; for each item, the key is the Gödel code of the new peptide and the value is a list recording the numbers of the DNA translation proteins covering the peptide on this DNA strand.

[0093] S54. Based on the first index table and the second index table, determine whether the DNA strand potentially contains a region expressing RNA protein.

[0094] Further, step S54 specifically includes: reading the peptide coding covered by the RNA translation protein from the first index table set; calculating whether the proportion of the peptide coding present in the second index table set exceeds the determination threshold proportion; if so, determining that the DNA strand potentially contains a region expressing RNA protein. Specifically, based on the two index table sets, determining whether DNA strand A potentially contains a region expressing RNA protein B; reading the peptide coding covered by protein B from index table set one, and determining whether the index table of DNA strand A can cover peptides covered by protein B exceeding the threshold proportion.

[0095] S55. Use a depth-first search algorithm to verify whether the CDS candidate satisfies the consistency between the order of peptide segments from the N-terminus to the C-terminus on the RNA protein and the order from the 5' end to the 3' end on the DNA strand, and add the target candidate CDS that satisfies the order consistency verification to the final result list.

[0096] S56. Gradually reduce the judgment threshold ratio and repeat the above steps S54 to S55 until the candidate combinations obtained in each step are added to the final result list.

[0097] To address the challenge of coding region localization, this embodiment proposes a species-specific solution. Species with virtually no splicing are defined as simple species, while those with widespread splicing are defined as complex species. For simple species, a deep learning-based gene scoring model is designed to assess the reliability of the reported results using gene-level features. For complex species, a filter based on a "multi-stage CDS mapping group" is innovatively designed using transcriptome data to exclude random matches. This dual-track strategy significantly improves the accuracy of coding region inference. Evaluations on labeled datasets show that the recall rate of the CDS inference results from this embodiment is improved by 14.7 percentage points compared to other methods, while the doubt rate is reduced by 20.9 percentage points.

[0098] To address the computational bottleneck, this invention provides an efficient sequence tag indexing technology. By encoding protein subsequences with Gödel and storing them in an index table, the time complexity of querying "the first protein number covering a certain amino acid sequence" is reduced to 0. This successfully overcomes the bottleneck of drastically reduced computational efficiency when considering splicing and mutation, making large-scale proteogenomic analysis possible.

[0099] Compared with existing technologies, this invention has significant technical advantages and practical application value. First, in terms of analytical performance, the method provided by this invention maintains high sensitivity and accuracy even with a 50-fold increase in database size, achieving accurate inference of new coding regions with only about a 3% increase in processing time, demonstrating unprecedented resolution and processing speed. Second, in terms of practical application effects, the method discovered 2108 new events validated by BLASTp in pear genome annotation, including 10466 splice sites and 3664 mutation sites, identifying 1.73 times more new proteins than traditional methods. In lung cancer sample analysis, the number of non-classical HLA-binding peptides identified by this method is more than 50 times that of existing methods, with over 70% demonstrating binding affinity validated by NetMHC. This indicates that this invention can comprehensively detect splicing, mutations, and new protein isoforms, supporting various downstream applications and showing broad application prospects in fields such as immunopeptidomics. These advantages make the method provided by this invention a new benchmark for proteogenomics analysis, providing reliable technical support for in-depth multi-omics research.

[0100] In another embodiment of the present invention, a protein genomics analysis system is also provided for implementing the method as described in the above embodiments, comprising:

[0101] One or more processors;

[0102] The memory stores computer-executable instructions that, when executed by the processor, are configured to perform the steps of the proteogenomic analysis method for discovering new coding regions.

[0103] The protein genomics analysis system described in this embodiment is a hardware and software combination implementation of the protein genomics analysis method for discovering new coding regions described in the previous embodiments. The core uses a collaborative architecture of "processor + memory" to transform the method steps into executable computer operations, ensuring the stable and efficient implementation of the detection method.

[0104] The system described in this embodiment is composed of various functional components. The specific functions of each functional component can be found in the description of the above embodiment, and will not be repeated here.

[0105] In summary, the proteogenomic analysis method and system for discovering new coding regions provided by this invention achieves complete coverage of a customized database through multi-framework translation fusion technology; it is the first to integrate iterative search and intelligent database compression strategies to design a large-scale database search process, solving the problem of search space expansion; it is the first to design a gene scoring model for simple species and a filter based on a "multi-stage CDS mapping group" for complex species, solving the problem of inference accuracy from new peptides to new coding regions; it overcomes the computational speed bottleneck by employing efficient indexing technology and multi-level parallelization technology; and it proposes a systematic evaluation scheme for CDS report quality in the field of proteogenomics. These innovations together constitute a comprehensive, accurate, and efficient new system for proteogenomic analysis, providing reliable technical support for related research.

[0106] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0107] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of the present invention is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0108] Of course, the present invention may have other various embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and modifications according to the present invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A protein genomics analysis method for discovering new coding regions, characterized in that, Including the following steps: The genome sequence was translated into DNA six-frames and the transcriptome sequence into RNA three-frames, and the translation results were merged to construct a custom database. A multi-step database search strategy is used to search the mass spectrometry data to obtain new peptide identification results. Specifically, the multi-step database search strategy is as follows: based on the mass spectrometry data, a reference database is searched to obtain reliable peptides; unresolved spectra are used to search the customized database to retain the proteins corresponding to the identified peptides to form a refined database; and the mass spectrometry data is then used to search the refined database to identify new peptides. The identified novel peptides are mapped to potential expression regions of the genome using a sequence tag indexing method to generate CDS candidates; For the first category of species, a gene scoring model based on deep learning is used to filter the CDS candidates; wherein, the first category of species are those in which splicing generally does not occur. For the second type of species, a filter based on a multi-stage CDS mapping group is used to filter the CDS candidates in order to reconstruct the complete gene coding region structure; wherein, the second type of species is a species in which splicing is common.

2. The method according to claim 1, characterized in that, The process of obtaining reliable peptides by searching a reference database based on the mass spectrometry data specifically involves: The reference database was searched using all mass spectrometry data, and identification results with a false discovery rate of no more than 1% at the peptide level were considered reliable peptides; the unresolved spectra were mass spectra that were not identified in the search.

3. The method according to claim 1, characterized in that, The step of using the mass spectrometry data to search the refined database to identify new peptides specifically includes: The refined database is searched using the entire dataset of the mass spectrometry data to obtain peptide identification results; Based on whether the peptide identification results are in the reference database, they are distinguished as new peptides and old peptides, and the FDR estimation is performed on the two types of peptides, while retaining results with a false discovery rate of less than 1%.

4. The method according to claim 1, characterized in that, The step of mapping the identified novel peptides to potential expression regions of the genome using a sequence tag indexing method to generate CDS candidates specifically includes: S31. The identified new peptides are mapped to the genome according to the priority of normal mapping, splicing mapping, and mutation mapping, using the sequence tag indexing method; S32. Based on the mapping position, extend to the 5' end and the 3' end to generate CDS candidates.

5. The method according to claim 4, characterized in that, The step of extending from the mapping position to the 5' and 3' ends to generate CDS candidates also includes: For the first type of species, starting from the upstream peptide mapping position and extending towards the 5' end, the starter is searched according to the preset start codon priority.

6. The method according to claim 1, characterized in that, The deep learning-based gene scoring model is a multilayer perceptron (MLP) model; the filtering of the CDS candidates using the deep learning-based gene scoring model for the first species includes: S41. Extract the identification results of the reliable peptide, map them to the genome, and complete the inference of the starter; S42. Extract training samples from the genome mapping results; wherein, the positive samples of the training samples are mapping results supported by genome annotation, and the negative samples include mapping results without genome annotation support and mapping results of decoy peptides. S43. Extract CDS length, peptide score, amino acid coverage, and initiator type features from the training samples to train the multilayer perceptron (MLP) model. S44. Use the trained multilayer perceptron (MLP) model to score the CDS candidates and filter them based on the scores.

7. The method according to claim 1, characterized in that, For the second type of species, the method of filtering the CDS candidates using a filter based on a multi-stage CDS mapping group to reconstruct the complete gene coding region structure includes: S51. The new peptide is mapped back to the protein translated from RNA based on the sequence tag indexing method; S52. Construct a first index table set, where the key is the number of the RNA-translated protein and the value is a Gödel code list covering the peptide segments of that protein. S53. Construct a second index table set, where the key is the Gödel code of the new peptide and the value is a list of protein IDs covering the new peptide on the DNA strand. S54. Based on the first index table and the second index table, determine whether the DNA strand potentially contains a region expressing RNA protein. S55. Use a depth-first search algorithm to verify whether the CDS candidate satisfies the consistency between the order of peptide segments from the N-terminus to the C-terminus on the RNA protein and the order from the 5' end to the 3' end on the DNA strand, and add the target candidate CDS that satisfies the order consistency verification to the final result list. S56. Gradually reduce the judgment threshold ratio and repeat the above steps S54 to S55 until the candidate combinations obtained in each step are added to the final result list.

8. The method according to claim 7, characterized in that, The determination of whether a DNA strand potentially contains a region expressing RNA protein, based on the first and second index tables, specifically includes: Read the peptide codes covered by the RNA-translated protein from the first index table set; Calculate whether the proportion of the peptide encoding that exists in the second index table exceeds the determination threshold proportion; if so, determine that the DNA strand potentially contains a region expressing RNA protein.

9. A protein genomics analysis system for implementing the method as described in any one of claims 1 to 8, characterized in that, include: One or more processors; The memory stores computer-executable instructions that, when executed by the processor, are configured to perform the steps of the method.

Images