Identification method of immune peptide group based on epitope conservation

Abstract

The present invention relates to the field of bioinformatics, and specifically, to an immune peptide group identification method based on epitope conservation. The immune peptide group identification method based on epitope conservation provided by the present invention solves the problem of too large search space caused by the "enzyme-free" characteristics of the immune peptide group, and the low sensitivity of mass spectrometry search identification caused by this, by adaptively learning the conservation of amino acid sites in the immune peptide group sequence. By incorporating the results of one round of search into the above-mentioned specific immune peptide group database, and conducting a second round of search, quality control is performed based on the classic target bait library strategy, solving the problem that the existing technical solutions generally lack reliable FDR.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to an immunopeptide group identification method based on epitope conservation. Background Art

[0002] The collection of short peptides presented to the cell surface by class I and class II human leukocyte antigen (HLA) molecules is called the immune peptidomime, which is of great significance to basic scientific research and translational medicine. For basic scientific research, a comprehensive analysis of the immune peptidomime helps to understand the key components of the immune system. For translational medicine, an accurate understanding of the immune peptidomime can directly promote and guide the development of a new generation of vaccines and cellular immunotherapies for autoimmune diseases, infectious diseases and cancer.

[0003] The discovery of immune peptide groups is mainly based on high-throughput mass spectrometry identification technology. The software and tools currently used for immune peptide group mass spectrometry data analysis were originally developed for traditional top-down proteomics. These software and tools complete mass spectrometry data identification by matching the mass spectrometry secondary spectrum with the reference protein sequence. However, the immune peptide group is quite different from the traditional proteome. In the preparation of traditional proteome samples, trypsin is required for enzymatic digestion. When the software performs algorithm matching, it only needs to add the peptides generated by the electronic digestion of the reference protein sequence by trypsin to the search space (Search Space), which has limited possibilities. After matching and scoring, a significance test can be performed to identify high-confidence peptides. However, the short peptide sequences in the immune peptide group samples are derived from the non-specific enzymatic digestion of the intracellular proteasome. All peptides of appropriate length contained in the reference protein sequence need to be added to the search space, resulting in excessive expansion of the search space. Too large a search space increases the number of candidates for peptide spectrum matches (PSM), reduces the score discrimination between peptide spectrum matches, interferes with the statistical effect of quality control, overestimates the false discovery rate (FDR), and thus reduces the sensitivity of the identification results. In addition, more and more new production mechanisms and sources of immune peptide groups have been elucidated, for example, immune peptide groups may originate from non-coding regions, may undergo proteasome splicing, etc. This further exacerbates the expansion of the search space, further reduces the sensitivity of identification, and even makes search identification difficult.

[0004] Therefore, there is an urgent need to develop a new and highly accurate immune peptide group identification method. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems in the related art to a certain extent. To this end, the present invention provides an immune peptide group identification method based on epitope conservation, which solves the problem of too large search space caused by the "enzyme-free" characteristics of the immune peptide group and the low sensitivity of mass spectrometry search identification caused by it through adaptive heuristic learning of the conservation of amino acid sites in the immune peptide group sequence. By incorporating the results of one round of search into the above-mentioned specific immune peptide group database, and conducting a second round of search, quality control is performed based on the classic target bait library strategy to solve the problem that the existing technical solutions generally lack reliable FDR.

[0006] To this end, the first aspect of the present invention provides a method for identifying an immune polypeptide group. According to an embodiment of the present invention, the method comprises:

[0007] (1) performing a search and identification on the immune peptide group mass spectrometry data from the sample to be tested and a reference protein library, and selecting high-confidence peptide segments to form a search library set 1, wherein the reference protein library contains multiple known protein sequences;

[0008] (2) constructing a machine learning model based on the search library set 1 as a training set, wherein the machine learning model is used to predict that the peptide segment belongs to a predetermined protein type;

[0009] (3) dividing the protein sequence in the reference protein library into peptide sequences to form a candidate polypeptide library, using the machine learning model to predict and score the candidate polypeptide library, filtering the candidate polypeptide library according to the score, and filtering the peptides obtained as sub-library 1;

[0010] (4) merging the search library set 1 with the sub-library 1 to obtain a specific polypeptide library;

[0011] (5) performing a secondary search and identification on at least a portion of the immune peptide group mass spectrometry data and the specific peptide library to obtain a final peptide segment, wherein the final peptide segment belongs to a predetermined protein type.

[0012] The immune peptide group identification method provided by the present invention uses mass spectrometry identification to obtain the amino acid conservation of the immune peptide group, narrows the reference protein database, performs a second round of database search and selectively merges the search results to obtain a high-reliability and high-sensitivity identification peptide segment.

[0013] The immune peptide group identification method according to the embodiment of the present invention may also have at least one of the following additional technical features:

[0014] According to an embodiment of the present invention, the method further includes:

[0015] The peptides in the search library set 1 are clustered by dimensionality reduction. After clustering, the optimal cluster number result is selected, outlier peptides are eliminated, and other clustered peptides are used as training sets to train the machine learning model. According to an embodiment of the present invention, the method also includes repeating steps (2) to (5) on the basis of the secondary search, and performing multiple searches until the identification amount reaches saturation.

[0016] According to an embodiment of the present invention, the dimensionality reduction clustering algorithm is selected from GibbsCluster, PCA, t-SNE, and UMAP dimensionality reduction clustering algorithms.

[0017] According to an embodiment of the present invention, the feature dimensions of the machine learning include amino acid sequence, polypeptide mRNA expression level, polypeptide hydrophilicity, and thermal stability.

[0018] According to an embodiment of the present invention, the data source of the machine learning model also includes multiple batches of identification results and / or external data of the same type.

[0019] According to an embodiment of the present invention, the machine learning model is selected from Position-Specific ScoringMatrix, Artificial neural networks, Gated recurrent Neural networks, Allele-specific feed forward neural networks, and Deep neural network. According to an embodiment of the present invention, in step (3) of the method, the threshold for filtering the reference protein library using the machine learning model is based on the score percentage, background sequence score estimation, or sequence score distribution.

[0020] According to an embodiment of the present invention, the immune polypeptide group is selected from antigen epitope peptides, neoantigen immune polypeptide groups, tumor-associated antigen immune polypeptide groups, proteasome splicing peptide immune polypeptide groups, and non-coding region epitope immune polypeptide groups.

[0021] According to an embodiment of the present invention, the antigen epitope peptide includes an antigen epitope peptide MHC class I epitope peptide or an MHC class II epitope peptide.

[0022] According to an embodiment of the present invention, the reference protein library is selected from a public protein database or a protein database generated by post-sequencing assembly and translation.

[0023] According to an embodiment of the present invention, the public protein database is selected from Uniprot Human database, NCBI database, Uniprot database, IEDB database, IMGT database, Proteomics DB database, PIR database. According to an embodiment of the present invention, the database search software is selected from at least one of MSGFplus, pFind, MaxQuant, Mascot, PD, MSfragger, PEAKS, X! Tandem, and Comet.

[0024] According to an embodiment of the present invention, the method comprises:

[0025] (1) Perform a search and identification on the immune peptide group mass spectrometry data and the reference protein library, retain the credible peptides with FDR <= 1%, and obtain the search library set 1;

[0026] (2) performing GibbsCluster clustering on the peptides in the search library set 1. After clustering, selecting the optimal cluster number result, eliminating outlier peptides, and using other clustered peptides as training sets to train the PSSM model;

[0027] (3) dividing the reference protein library into peptide sequences to form a candidate peptide library, using the PSSM model to predict and score the candidate peptide library, filtering according to the scores of the peptides, retaining peptides with PSSM Score>=0.4, and recording the peptides obtained by filtering as sub-library 1;

[0028] (4) merging the search library set 1 with the sub-library 1 to obtain a specific peptide library;

[0029] (5) performing a secondary search and identification on at least a portion of the immune peptide group mass spectrometry data and the specific peptide library to obtain a final identified peptide segment.

[0030] According to an embodiment of the present invention, in step (1), when performing the primary library search and identification, the enzyme cleavage mode is set to non-specific enzyme cleavage, and the peptide sequence length is 8 to 15.

[0031] According to an embodiment of the present invention, the peptide sequences obtained by searching the library are eliminated from common contaminating peptides and bait peptide sequences, and then FDR filtering is performed.

[0032] According to an embodiment of the present invention, in step (2), the number of clusters is set to 1 to 6, the Motif length is set to 9, the maximum insertion length is 1, and the maximum deletion length is 6.

[0033] According to an embodiment of the present invention, the amino acid background frequency used in the PSSM model is generated according to the human whole protein data of the Uniprot database.

[0034] According to an embodiment of the present invention, in step (3), the reference protein library in (1) is divided into peptide sequences according to a step length of 8 to 15 amino acids, and a candidate polypeptide library is formed after removing redundancy.

[0035] According to an embodiment of the present invention, in step (5), when performing secondary library search and identification, the enzyme cleavage mode is set to non-specific enzyme cleavage, and the peptide sequence length is 8 to 15.

[0036] Optionally, the peptide sequences obtained from the secondary search are free of common contaminating peptides and decoy peptide sequences, and then FDR filtering is performed.

[0037] The second aspect of the present invention provides an immune peptide group identification system. According to an embodiment of the present invention, the system comprises:

[0038] A library search unit, wherein the library search unit is used to search and identify the immune peptide group mass spectrometry data in a reference protein library, screen peptide segments according to credibility, and obtain a library search set 1;

[0039] A machine learning unit, wherein the machine learning unit is used to perform dimensionality reduction clustering on the peptide segments in the search library set 1. After the clustering is completed, the optimal cluster number result is selected, outlier peptide segments are eliminated, and other clustered peptide segments are used as training sets to train the machine learning model;

[0040] A specific polypeptide library generation unit, wherein the specific polypeptide library generation unit is used to divide the reference protein library into peptide sequences to form a candidate polypeptide library, use the machine learning model to predict and score the candidate polypeptide library, filter according to the scores of the peptides, record the peptides obtained by filtering as sub-library 1, merge the search library set 1 with the character library 1, and obtain a specific polypeptide library;

[0041] A secondary library search unit is used to perform secondary library search and identification on the immune peptide group mass spectrum data and the specific peptide library to obtain the final identified peptide segment.

[0042] The immune peptide group identification system according to the embodiment of the present invention may also have at least one of the following additional technical features:

[0043] According to an embodiment of the present invention, the dimensionality reduction clustering algorithm is selected from GibbsCluster, PCA, t-SNE, and UMAP dimensionality reduction clustering algorithms.

[0044] According to an embodiment of the present invention, the feature dimensions of the machine learning include amino acid sequence, polypeptide mRNA expression level, polypeptide hydrophilicity, and thermal stability.

[0045] According to an embodiment of the present invention, the machine learning model is selected from Position-Specific ScoringMatrix, Artificial neural networks, Gated recurrent Neural networks, Allele-specific feed forward neural networks, and Deep neural network.

[0046] According to an embodiment of the present invention, the immune polypeptide group is selected from antigen epitope peptides, neoantigen immune polypeptide groups, tumor-associated antigen immune polypeptide groups, proteasome splicing peptide immune polypeptide groups, and non-coding region epitope immune polypeptide groups.

[0047] According to an embodiment of the present invention, the antigen epitope peptide includes an antigen epitope peptide MHC class I epitope peptide or an MHC class II epitope peptide.

[0048] According to an embodiment of the present invention, the reference protein library is selected from a public protein database or a protein database generated by post-sequencing assembly and translation.

[0049] According to an embodiment of the present invention, the public protein database is selected from Uniprot Human database, NCBI database, Uniprot database, IEDB database, IMGT database, Proteomics DB database, PIR database. According to an embodiment of the present invention, the database search software is selected from at least one of MSGFplus, pFind, MaxQuant, Mascot, PD, MSfragger, PEAKS, X! Tandem, and Comet.

[0050] The third aspect of the present invention provides an electronic device for identifying an immune polypeptide group. According to an embodiment of the present invention, the electronic device for identifying an immune polypeptide group includes a memory and a processor;

[0051] The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the immune polypeptide group identification method described in the first aspect of the present invention.

[0052] A fourth aspect of the present invention provides a computer-readable storage medium. According to an embodiment of the present invention, the computer-readable storage medium stores a computer program, which, when executed by a processor, implements the immune polypeptide group identification method described in the first aspect of the present invention.

[0053] Many studies have shown that excessive search space will have an adverse effect on mass spectrometry identification results, and a considerable proportion of correct identification results will be discarded by FDR error filtering. For this reason, a variety of solutions have been proposed in the field of immunopeptide groups.

[0054] Solution 1: Quality control adjustment solution. Directly adjust the FDR threshold of the quality control of the immune peptide group mass spectrometry library search process, such as relaxing the FDR threshold to 10%, or directly abandoning the FDR quality control, so as to reduce or avoid peptides that are mistakenly filtered out by the quality control process.

[0055] Solution 2: SpectMHC (Murphy JP, Konda P, Kowalewski DJ, et al. MHC-I ligand discovery using targeted database searches of mass spectrometry data: implications for T-cell immunotherapies [J]. Journal of proteome research, 2017, 16 (4): 1806-1816). By applying the pre-trained HLA epitope affinity prediction software netMHCpan-3.0, the search space is screened and filtered to construct a specific targeted database with a smaller search space. The targeted database is then used for mass spectrometry search identification to improve the sensitivity of identification. The average number of newly identified epitope peptides per single experiment increased by 60%.

[0056] Scheme 3: MS-Rescue (Andreatta M, Nicastri A, Peng X, et al. MS-Rescue: A Computational Pipeline to Increase the Quality and Yield of Immunopeptidomics Experiments [J]. Proteomics, 2019, 19 (4): 1800357). The high-confidence peptides (Peptide Level FDR <= 1%) obtained by standard search library identification were used as training sets, and the NNAlign model was used to learn their sequence features. The model was then used to filter low-confidence peptides (Peptide Level FDR> 1%), and these "rescued" peptides were merged with high-confidence peptides as the final identification results. On average, the number of newly identified epitope peptides in a single experiment increased by 20%.

[0057] Solution 4: De novo solution. De novo mass spectrometry identification software is also used in immunopeptide group identification to avoid the problem of too large a search library, such as PEAKSStudio (Ma B, Zhang K, Hendrie C, et al. PEAKS: powerful software for peptide denovo sequencing by tandem mass spectrometry [J]. Rapid communications in mass spectrometry, 2003, 17 (20): 2337-2342) software developed by Bioinformatics Solutions Inc. The de novo strategy does not rely on the sequence information in the reference database, but directly infers the sequence information of the peptide segment based on the mass-to-charge ratio difference between the ion peaks in the spectrum, thereby avoiding the problem of too large a search library.

[0058] In general, the above technical solutions all have shortcomings, which are often due to the fact that these solutions change the traditional quality control strategy based on the target bait library or directly avoid the library search process, resulting in a lack of credibility in the identification results. Specifically, each solution has its own unique defects, which are detailed as follows.

[0059] Solution 1: Quality control adjustment solution. First, its scope of application is limited. For studies on the discovery of individual tumor-specific antigen epitopes, relaxing or abandoning FDR quality control can alleviate the problem caused by excessive search space. However, for studies that produce a large amount of highly accurate immune peptide group data, relaxing or abandoning FDR quality control will lead to a decrease in data accuracy, which may seriously mislead subsequent studies such as sequence characteristics. Secondly, the credibility of the peptide sequence obtained through quality control adjustment is low, and it is generally necessary to combine synthetic peptide mass spectrometry verification experiments to illustrate the authenticity of the identified peptides, which leads to increased actual identification costs and longer time.

[0060] Solution 2: Targeted library reduction solution based on prior affinity prediction model - SpectMHC. The targeted database generation process relies on the model trained by external prior data, and there are several potential problems. First, the identification results may introduce the bias of prior training data with the model prediction process, which is not conducive to discovering the true sequence characteristics. Secondly, the sources and data quality of different typing data in the external prior data vary greatly, and the credibility of affinity prediction varies greatly, which ultimately leads to the credibility of the identification results being unstable. In addition, the external prior data is derived from affinity experimental verification data. Although having a certain affinity is a necessary condition for antigen presentation, there is no strict causal relationship between the two, so the use of a prediction model constructed based on in vitro affinity experiments to screen the targeted database for antigen presentation may cause database bias.

[0061] Solution 3: MS-Rescue, a solution for rescuing low-confidence peptides based on sequence conservation. The peptides rescued by this solution are filtered only by the scores of the NNAlign model, without the traditional target decoy strategy (TDS) quality control, and therefore do not have a measurable FDR, and the overall credibility of its identification results is low. In addition, this solution does not provide a clear solution for how to integrate the rescued low-confidence peptides with the original high-confidence peptides.

[0062] Solution 4: De novo analysis solution. Lacking the classic TDS quality control process, it is impossible to provide a reliable FDR and the accuracy is difficult to guarantee. Moreover, de novo analysis is difficult to calculate and has many possibilities. The identification results rarely overlap with the sequences obtained by traditional search strategies (Li S., DeCourcy A., Tang H. (2018) Constrained De Novo Sequencing of neo-Epitope Peptides Using Tandem Mass Spectrometry. In: Raphael B. (eds) Research in Computational Molecular Biology. RECOMB 2018. Lecture Notes in Computer Science, vol 10812. Springer, Cham), and the obtained sequences have poor interpretability. Moreover, mainstream de novo mass spectrometry identification software represented by PEAKS Studio is mostly commercial software, which cannot be obtained for free and has high application costs.

[0063] The immunopeptide group identification method based on epitope conservation provided by the present invention solves the above problems. The key technical point of the present invention is to adaptively and heuristically obtain the amino acid conservation characteristics of the immunopeptide group, thereby narrowing the reference protein database, performing a second round of database search and selectively merging the search results to obtain high-reliability and high-sensitivity identification peptide segments.

[0064] The present invention intends to utilize the conservation of amino acid sites in the immune polypeptide group sequence, and train the machine learning model with the immune polypeptides obtained by searching the standard reference protein library, and use the model to filter the reference protein library to obtain a specific peptide library with a significantly reduced search space, and merge it with the results of the first round of library search, and then search the library to obtain the final identification result after screening. The so-called "adaptive heuristic" means that the present invention does not require any prior knowledge of the conservative characteristics of the immune polypeptide group, but instead conducts instant learning of the immune polypeptides produced in each experiment, obtains features and uses them to improve the library search algorithm and improve the identification results.

[0065] The beneficial effects that the present invention can bring are as follows:

[0066] 1) High sensitivity: The present invention can effectively reduce the search space and increase the discrimination between the spectrum peptide matching scores, thereby increasing the difference in the score distribution between the target library and the bait library in the classic target bait library quality control strategy, improving the statistical effectiveness of quality control, and identifying and obtaining more immune peptide group peptides.

[0067] 2) High accuracy: When performing the second round of library search, the present invention directly incorporates the peptide list of the first round of library search into the database, avoiding the quality control failure problem caused by artificial merging of peptides or spectra after the second round of library search. At the same time, the main source of new peptides identified in the second round of the present invention is the spectra that were not identified or incompletely identified in the first round, so there are very few contradictions in the result merging, ensuring the high accuracy of the present invention.

[0068] 3) Wide adaptability: The present invention's study of the conservation of amino acid sites in immune peptide groups is completely based on the data produced by the current experiment, which ensures the wide applicability of the present invention and avoids the limitation of some similar software that cannot be applied due to the lack of prior data.

[0069] 4) Low cost: The present invention does not rely on experimental methods and has no reagent and consumables costs. And because the database used for the second round of database search is extremely small, the memory consumption is low, and the time taken is short, the analysis can be completed on an office computer without the need to purchase a high-performance computer or computing platform. Compared with the improvement of experimental technology or commercial software analysis, the present invention has a very low cost for improving the identification of immune peptide groups.

[0070] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0072] Figure 1 A schematic diagram showing the technical solution of the immune polypeptide group identification method provided by the present invention;

[0073] Figure 2 The results of evaluating the number of identified peptides after applying the present invention (MIDS, motif-guided immunopeptidomic database search) in mass spectrometry sample analysis of 8 groups of tumor cell lines are shown;

[0074] Figure 3The results of the identification accuracy evaluation after applying the present invention (MIDS) in the mass spectrometry sample analysis of 8 groups of tumor cell lines are shown;

[0075] Figure 4 The figure shows the consistency of the spectrum matching when two rounds of identification of K562-A1101 were performed using the identification method of the present invention;

[0076] Figure 5 An experimental secondary spectrum of a newly identified peptide using the immune peptide group identification method provided by the present invention is shown, as well as a theoretical spectrum predicted by the third-party software pDeep. DETAILED DESCRIPTION

[0077] The present invention intends to utilize the amino acid site conservation of the immune peptide group sequence, apply the immune peptide obtained by searching the standard reference protein library to train the machine learning model, and apply the model to filter the reference protein library to obtain a specific peptide library with a significantly reduced search space, and merge it with the first round of library search peptide results, and then search the library again, and obtain the final identification result after screening. The specific implementation plan of the present invention includes the following four steps. The implementation plan can be referred to in the attached Figure 1 :

[0078] Step 1: Standard library search and obtain high-confidence identification peptides. Import the mass spectrometry Raw file and the human protein library (such as Uniprot Human database) Fasta file into the mass spectrometry analysis software for library search and identification, set the enzyme digestion mode to non-specific digestion, the peptide sequence length to 8-15, and other parameters according to the experimental and computer information. The peptide sequences obtained from the library search exclude common contaminating peptides and bait peptide sequences, and then filter according to FDR, retaining high-confidence peptides with FDR <= 1%, recorded as library search set 1.

[0079] Step 2: Learning the conservation of immune peptide group sites. Cluster the high-confidence peptides using GibbsCluster, set the number of clusters to 1-6, set the Motif length to 9, the maximum insertion length to 1, the maximum deletion length to 6, and the initial seed number and iteration number to the specific situation. After clustering, select the optimal cluster number result, remove the outlier peptides, and use the other clustered peptides as training sets to train the position-specific scoring matrix (PSSM) model. The amino acid background frequency used in the model is generated according to the human whole protein data of the Uniprot database.

[0080] Step 3: Generation of specific peptide library. The whole human protein library used for standard library search and identification in step 1 is divided into peptide sequences according to the step length of 8-15 amino acids, and the candidate peptide library is formed after removing redundancy. The position-specific scoring matrix model trained in step 2 is used to predict and score the candidate peptide library, and the peptides are filtered according to the scores obtained. Peptides with PSSM Score>=0.4 are retained, and a Fasta file is generated, recorded as sub-library 1. Subsequently, the high-confidence identified peptides obtained in a round of library search in step 1 are also generated into a Fasta file, recorded as sub-library 2. Sub-library 1 and sub-library 2 are merged to form a specific peptide library.

[0081] Step 4: Secondary library search. Import the mass spectrometry Raw file in step 1 and the specific peptide library Fasta file in step 3 into the mass spectrometry analysis software for library search and identification. Set the enzyme cleavage mode to non-specific enzyme cleavage, the peptide sequence length to 8-15, and other parameters according to the experimental and computer information. The peptide sequences obtained from the library search exclude common contaminating peptides and bait peptide sequences, and then filter according to FDR. The peptides with FDR <= 1% are retained as the final identified peptides. The parameter details described in the above steps, such as peptide length, are only applicable to HLA class I peptide identification and should be adjusted as appropriate in practice. In the following descriptions and charts, "library reduction strategy" is the abbreviation for the above technical solution.

[0082] The method of selecting high-confidence peptides as training sets in the present invention is not limited to FDR<=1%, and can also be screened by identification score, identification sequence ranking percentage, or other FDR values, as long as high-confidence peptides can be obtained.

[0083] The terms used in the present invention are explained below.

[0084] "Contaminating peptides" are common contaminants introduced during the sample preparation and detection process of mass spectrometry experiments, such as keratin, bovine serum albumin, protease, etc.

[0085] “High-confidence peptides” refer to peptides with high spectral quality, strong signals, and many repetitions.

[0086] "Decoy peptides" are non-natural protein peptides artificially designed in the classic Target decoy strategy to complete FDR calculation.

[0087] "Optimal clustering" shows how many clusters with different amino acid characteristics the clustering software believes there are in the input peptides. For example, when the optimal clustering is 5, it means that the clustering software believes there are 5 clusters with different amino acid characteristics in the input peptides, and all peptides in these 5 clusters are used in the training set.

[0088] "External data" refers to public data and commercially purchased data, which can be used as one of the data sources for training machine learning models.

[0089] The "bait library" is a "fake" protein library constructed in proteomic data analysis to calculate FDR. It is a very common and standardized strategy and a necessary search database for FDR quality control.

[0090] GibbsCluster is an unsupervised clustering software. It will find groups of peptides with similar amino acid site characteristics and cluster them together. As the clustering process proceeds, a small number of peptides that cannot belong to any group are marked as outliers. The criterion for determining the optimal number of clusters is the maximum KLD. Figure 1 As shown, the immune peptide group identification method provided by the present invention uses mass spectrometry identification to obtain the amino acid conservation of the immune peptide group, narrows the reference protein database, performs a second round of database search and selectively merges the search results to obtain high-reliability and high-sensitivity identification peptide segments.

[0091] The embodiments of the present invention are described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. If no specific techniques or conditions are specified in the embodiments, the techniques or conditions described in the literature in this area or the product specifications are used. The reagents or instruments used that do not specify the manufacturer are all conventional products that can be obtained commercially.

[0092] Example 1 Identification of the Mass Spectrometry Data of the Immune Peptide Group of the K562 Cell Line Stably Transfected with the HLA-A11:01 Gene

[0093] Due to some software limitations, the analysis needs to be performed alternately under Win and Linux systems. The detailed steps are as follows:

[0094] (1) Raw data format conversion: Import the original file K562-A1101_D180912_S1.raw of the downloaded data into MSConvertGUI.exe for format conversion. Set the output format to mgf, the binary encoding precision to 64 bits, the peakPicking msLevel to 1-2, and the other parameters to the software default. The file is named K562-A1101_D180912_S1.mgf, which is the original spectrum file. (Win)

[0095] (2) Preparation of reference protein library and bait library: Select all protein entries of Humo sapiens in the Uniprot online database (https: / / www.uniprot.org / ), download and save as uniprot.fasta file. In addition, download the contaminants.fasta file of the commonly used contaminant protein library identified by mass spectrometry from the MaxQuant official website. Merge the two fasta files to obtain uniprot.contaminants.fasta, which is the reference protein library for the first round of library search (Linux). Use the mimic tool provided in MSGFPlus to generate the bait library for subsequent FDR quality control. Use the BuildSA tool provided in MSGFPlus to generate indexes for the reference protein library and the bait library to speed up the library search process. (Linux)

[0096] (3) The first round of library search: MSGFPlus is used to search the reference protein library and the bait library respectively, and the result is the library search result file in the mzid file format; (Linux)

[0097] Use the MzIDToTsv tool in MSGFPlus to convert the above search results into TSV format; (Linux)

[0098] Use the msgf2pin tool in Percolator to convert the above TSV file into the pin file required for quality control; (Linux)

[0099] Use Percolator to perform quality control analysis on pin files and calculate peptide Q value to facilitate subsequent FDR quality control; (Linux)

[0100] The FilterPercolatorResult.pl script was used to filter the quality control results at 1% FDR to obtain high-confidence peptides in the first round of identification. As of this point, a total of 2,393 high-confidence epitope peptides with a length of 9-11 amino acids were obtained after redundancy removal. (Linux)

[0101] (4) Clustering to remove contamination: GibbsCluster-2.0 was used to perform cluster analysis on the high-confidence peptides identified in the first round, and the clustering result file was obtained in txt format; (Linux)

[0102] Observe the distribution characteristics of the amino acid sequence from the clustering results, select the appropriate cluster number results, remove the outlier peptides, and obtain the training set peptide list. (Linux)

[0103] (5) Model training: PSSM model training is used to learn the sequence amino acid features and obtain the amino acid position-specific score matrix. (Linux)

[0104] (6) Narrowing the reference protein library: All protein sequences in the reference protein library are divided into short peptide sequences of 9-11 amino acids and deduplicated to obtain the segmented peptides required for PSSM model prediction; (Linux)

[0105] Use the previously trained model to predict the short peptide sequence after segmentation to obtain the PSSM score of each peptide segment; (Linux)

[0106] The peptides with PSSM score exceeding 0.4 in the prediction results were extracted and converted into fasta format, and merged with the list of 2393 high-confidence peptides obtained in the first round of library search to form the reference protein library for the second round of library search. (Linux)

[0107] (7) Second round of database search: The database search was performed using pFind with the following settings: Reference database: K562-A1101_D180912.pssm0.4.fasta; Spectrum file: K562-A1101_D180912_S1.mgf; Enzyme setting: NoEnzymeU_C; Enzyme specificity setting: Non-Specific; Peptide false discovery rate (FDR): 1%; Protein false discovery rate (FDR): 100%; Sequence match length limit: 9-11aa; Initial allowed mass deviation of leading ion: 20ppm; Maximum fragment mass deviation: 20ppm; Open search: True. The identification results are output and saved in the "all_result.txt" file. Peptides matching the bait library and the contamination library are removed. The remaining peptide sequences are the identification results of MHC-I restricted epitopes and are saved as K562-A1101_D180912.SecondRun.tsv. (Win)

[0108] Example 2 Evaluation of the number of identified peptides after applying MIDS in mass spectrometry analysis of tumor cell lines

[0109] As attached Figure 2As shown, the number of identified peptides after applying the present invention (MIDS) in the mass spectrometry sample analysis of 8 groups of tumor cell lines was evaluated. Compared with the identification results of the standard strategy of MSGFplus (Kim S, Pevzner P A. MS-GF+ makes progress towards auniversal database search tool for proteomics [J]. Nature communications, 2014, 5: 5277), the library reduction strategy of the present invention increased the number of identifications by 30% to 100%.

[0110] Example 3 Evaluation of the accuracy of peptide identification after applying MIDS in mass spectrometry analysis of tumor cell lines

[0111] As attached Figure 3 As shown, the identification accuracy evaluation after applying the present invention (MIDS) in the mass spectrometry sample analysis of 8 groups of tumor cell lines. The accuracy used is measured by the pass rate of affinity prediction, and the prediction software is netMHCpan4.0 (Jurtz V, Paul S, Andreatta M, et al. NetMHCpan-4.0: improved peptide-MHC class I interaction predictions integrating eluted ligand and peptide binding affinity data [J]. The Journal of Immunology, 2017, 199 (9): 3360-3368). As can be seen from the figure, compared with the MSGFplus standard identification results, the present invention can maintain the same or even slightly improve the identification accuracy, with an improvement value of 0.01 to 0.07.

[0112] Example 4 Spectral matching consistency during two rounds of identification of K562-A1101

[0113] Figure 4 The figure shows the consistency of spectrum matching when two rounds of identification of K562-A1101 were performed using the identification method of the present invention. Figure 4 It shows that about 65% of the spectra were not identified in the first round of library search, but were successfully identified in the second round of library search ("unique after library reduction"), and only less than 2% of the spectra were completely inconsistent with the results of the two rounds of library searches ("inconsistent").

[0114] Example 5 Secondary Spectrum Verification of Newly Identified Peptides

[0115] As attached Figure 5As shown, an experimental secondary spectrum of a newly identified peptide obtained by the identification method of the present invention is shown, and a theoretical spectrum predicted by a third-party software pDeep (Zhou XX, Zeng WF, Chi H, et al. pdeep: Predicting MS / MS spectra of peptides with deep learning [J]. Analytical chemistry, 2017, 89 (23): 12690-12697) is shown. The two show a high degree of consistency, which proves the credibility of the new peptides identified by this method. The spectrum drawing software used is PDV (Li K, Vaudel M, Zhang B, et al. PDV: an integrative proteomics data viewer [J]. Bioinformatics, 2018, 35 (7): 1249-1251).

[0116] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0117] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for identifying an immune polypeptide group, characterized in that: include: (1) performing a search and identification on the immune peptide group mass spectrometry data from the sample to be tested and the reference protein library, and selecting the peptide segments with a FDR <= 1% confidence level to form a search library set 1, wherein the reference protein library contains multiple known protein sequences; (2) constructing a machine learning model based on the search library set 1 as a training set, wherein the machine learning model is used to predict that the peptide segment belongs to a predetermined protein type; (3) dividing the protein sequence in the reference protein library into peptide sequences to form a candidate polypeptide library, using the machine learning model to predict and score the candidate polypeptide library, filtering the candidate polypeptide library according to the obtained scores, and filtering the peptides obtained as sub-library 1; (4) merging the search library set 1 with the sub-library 1 to obtain a specific polypeptide library; (5) performing a secondary search and identification on at least a portion of the immune peptide group mass spectrometry data and the specific peptide library to obtain a final peptide segment, wherein the final peptide segment belongs to a predetermined protein type.

2. The method according to claim 1, characterized in that The method further comprises: The peptides in the search library set 1 are clustered by dimensionality reduction. After clustering, the optimal cluster number result is selected, outlier peptides are eliminated, and other clustered peptides are used as training sets to train the machine learning model.

3. The method according to claim 1, characterized in that The method further includes repeating steps (2) to (5) on the basis of the secondary library search, and performing multiple library searches until the identification amount reaches saturation.

4. The method according to claim 1, characterized in that: The feature dimensions of the machine learning include amino acid sequence, polypeptide mRNA expression level, polypeptide hydrophilicity, and thermal stability.

5. The method according to claim 1, characterized in that The data source of the machine learning model also includes multiple batches of identification results of the same type and / or external data.

6. The method according to claim 1, characterized in that In step (3), the threshold for filtering the reference protein library using the machine learning model is based on a score percentage, a background sequence score estimate, or a sequence score distribution.

7. The method according to claim 1, characterized in that The method comprises: (1) Perform a search and identification on the immune peptide group mass spectrometry data and the reference protein library, retain the credible peptides with FDR <= 1%, and obtain the search library set 1; (2) performing GibbsCluster clustering on the peptides in the search library set 1. After clustering, selecting the optimal cluster number result, eliminating outlier peptides, and using other clustered peptides as training sets to train the PSSM model; (3) dividing the reference protein library into peptide sequences to form a candidate peptide library, using the PSSM model to predict and score the candidate peptide library, filtering according to the scores of the peptides, retaining peptides with PSSM Score>=0.4, and recording the peptides obtained by filtering as sub-library 1; (4) merging the search library set 1 with the sub-library 1 to obtain a specific peptide library; (5) performing a secondary search and identification on at least a portion of the immune peptide group mass spectrometry data and the specific peptide library to obtain a final identified peptide segment.

8. The method according to claim 7, characterized in that In step (1), when performing the primary library search and identification, the enzyme cleavage mode is set to non-specific enzyme cleavage.

9. The method according to claim 7, characterized in that: The peptide sequences obtained from the search were removed from common contaminating peptides and bait peptide sequences, and then FDR filtering was performed.

10. The method according to claim 7, characterized in that The peptide sequences obtained from the secondary search were removed from common contaminating peptides and bait peptide sequences, and then FDR filtering was performed.

11. An immune peptide group identification system, characterized in that: include: A library search unit, wherein the library search unit is used to search and identify the immune peptide group mass spectrometry data in a reference protein library, screen peptide segments according to credibility, and obtain a library search set 1; A machine learning unit, wherein the machine learning unit is used to perform dimensionality reduction clustering on the peptide segments in the search library set 1. After the clustering is completed, the optimal cluster number result is selected, outlier peptide segments are eliminated, and other clustered peptide segments are used as training sets to train the machine learning model; A specific polypeptide library generation unit, wherein the specific polypeptide library generation unit is used to divide the reference protein library into peptide sequences to form a candidate polypeptide library, use the machine learning model to predict and score the candidate polypeptide library, filter according to the scores of the peptides, record the peptides obtained by filtering as sub-library 1, merge the search library set 1 with the character library 1, and obtain a specific polypeptide library; A secondary library search unit is used to perform secondary library search and identification on the immune peptide group mass spectrum data and the specific peptide library to obtain the final identified peptide segment.

12. The system according to claim 11, characterized in that The immune polypeptide group is selected from antigen epitope peptides, neoantigen immune polypeptide groups, tumor-associated antigen immune polypeptide groups, proteasome splicing peptide immune polypeptide groups, and non-coding region epitope immune polypeptide groups.

13. The system according to claim 12, characterized in that The antigenic epitope peptide includes an antigenic epitope peptide MHC class I epitope peptide or an MHC class II epitope peptide.

14. The system according to claim 11, characterized in that The feature dimensions of the machine learning include amino acid sequence, polypeptide mRNA expression level, polypeptide hydrophilicity, and thermal stability.

15. An electronic device for identifying immune peptide groups, characterized in that: Including memory and processor; The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the immune polypeptide group identification method as claimed in any one of claims 1 to 10.

16. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the immune polypeptide group identification method as claimed in any one of claims 1 to 10 is implemented.

Images