Method for identifying protein cross-linking mass spectrometry peptide fragment pair
By extracting the long-range dependency of crosslinking spectra from the Transformer encoder and retrieving it using a vector database, the problems of low accuracy and large search space in peptide crosslinking identification are solved, achieving high-precision peptide crosslinking identification and fast matching.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- COMPUTER INNOVATION TECH RES INST OF ZHEJIANG UNIV
- Filing Date
- 2025-12-19
- Publication Date
- 2026-04-10
AI Technical Summary
Existing technologies suffer from problems such as excessively large search space, low recognition accuracy, and unstable localization of crosslinking sites when dealing with polypeptide crosslinking.
By extracting the long-range dependence of crosslinking spectra from the Transformer encoder and retrieving the types and crosslinking sites of multiple peptides using a vector database, a peptide vector database is constructed. Spectral characterization sequences are generated and similarity searches are performed to determine the target crosslinked peptides and crosslinking sites.
It significantly improves the accuracy of peptide crosslink recognition, reduces computational complexity, is applicable to various crosslinking forms, has good scalability and versatility, and can quickly match in large-scale protein databases.
Smart Images
Figure CN121838883A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the field of computer-aided biological data analysis, and particularly relates to a method for recognizing peptide pairs in protein cross-linking mass spectrometry. BACKGROUND
[0002] Chemical cross-linking mass spectrometry (XL-MS) is a powerful biochemical analysis technique that uses chemical cross-linking reagents to covalently link amino acid residues (usually lysines) within or between proteins or protein complexes that are close enough in space. This method can capture the conformation and interaction of proteins in living cells or in a natural state, which is difficult to achieve by traditional structural biology techniques such as X-ray crystallography and nuclear magnetic resonance (NMR). The arm length of the cross-linking reagent (usually between 10 and 15 Å) provides valuable distance constraint information for protein structure analysis.
[0003] Traditional cross-linking mass spectrometry mainly involves the cross-linking of two peptides (α / β), but in the presence of multi-functional cross-linking reagents (such as tri-functional, tetra-functional cross-linking reagents) or complex protein networks, cross-linking events involving three or more peptides often occur. Compared to dipeptide cross-linking, multi-peptide cross-linking has the following challenges: Exponential increase in spectral complexity: After fragmentation of multiple peptides, single-chain ions, double-chain ions, triple-chain ions, and even more combinations exist simultaneously, resulting in an exponential increase in the number of fragment peaks.
[0004] Greater variety of ion types: In addition to conventional b / y ions, there are also multi-peptide cross-linking ions, stub ions, dead-end ions, loop-link ions, and mixed ions that partially retain cross-linking, making fragment interpretation much more difficult.
[0005] Candidate search space expansion: The complexity of candidate number for n peptide cross-linking is O(N n ), which is much higher than that of dipeptide cross-linking, making traditional mass-based exhaustive search + scoring methods almost infeasible in large-scale proteomics.
[0006] Significant long-range dependence: Multi-peptide fragments often span a large range in the spectrum, and the global dependence between peaks in the spectrum must be captured in order to correctly identify the cross-linking combination and site.
[0007] Increased noise and interference: As the peak density increases, it becomes more difficult to distinguish isotopic interference, neutral loss peaks, and low-abundance fragment signals.
[0008] Existing methods mainly focus on the recognition of dipeptide cross-linking, and when faced with multi-peptide cross-linking, there are problems such as excessively large search space, low recognition accuracy, and unstable cross-linking site localization. SUMMARY
[0009] In order to solve the problems in the background art, the present application provides a method for identifying protein cross-linking mass spectrometry peptide pairs, which solves the technical problems of large search space, low identification accuracy and unstable cross-linking site positioning in the prior art when facing polypeptide cross-linking. The present application extracts the long-range dependence of the cross-linking spectrum in the Transformer encoder, and obtains the types and cross-linking sites of multiple peptides through vector database retrieval.
[0010] The technical scheme adopted by the present application is: A method for identifying protein cross-linking mass spectrometry peptide pairs: S1, constructing a peptide vector database: generating a candidate cross-linking peptide set based on a target protein database, and encoding each candidate cross-linking peptide into a corresponding cross-linking peptide embedding vector through a pre-trained peptide encoder, and storing it in the vector database.
[0011] S2, obtaining and preprocessing spectrum data: obtaining the original spectrum generated by the cross-linking mass spectrometry experiment, and preprocessing to obtain a peak list containing mass-to-charge ratio and intensity.
[0012] S3, generating a spectrum representation sequence: converting the peak list into a fixed-length spectrum representation sequence.
[0013] S4, generating a spectrum embedding vector: inputting the spectrum representation sequence into a pre-trained spectrum encoder for processing, and outputting the spectrum embedding vector corresponding to the original spectrum.
[0014] S5, retrieving candidate peptides: taking the spectrum embedding vector as a query vector, performing similarity retrieval in the vector database, and screening the top N most similar cross-linking peptide embedding vectors and their corresponding candidate cross-linking peptides.
[0015] S6, determining the target cross-linking peptide and the cross-linking site: performing ion matching scoring between each screened candidate cross-linking peptide and the original spectrum, and the candidate cross-linking peptide with the highest score is the target cross-linking peptide, and the cross-linking site of the target cross-linking peptide is obtained.
[0016] The candidate peptide set generated based on the target protein database is specifically: by simulating in vitro enzyme digestion or open cleavage, the target protein database is processed to generate the candidate cross-linking peptide set.
[0017] Each candidate cross-linking peptide is a dipeptide cross-linking pair, a tripeptide cross-linking combination or a tetrapeptide cross-linking combination.
[0018] In step S2, the preprocessing is specifically peak extraction, isotope removal, charge state deconvolution, noise suppression and normalization operation in sequence.
[0019] The step S3 is specifically: S31, uniformly divide the preset mass-to-charge ratio range into a plurality of continuous and equally spaced mass-to-charge ratio intervals; for each peak in the peak list, according to the mass-to-charge ratio of each peak, the intensity value corresponding to the peak is assigned to the corresponding mass-to-charge ratio interval; for each mass-to-charge ratio interval, the intensity values assigned by all peaks in the interval are summed to obtain the total intensity value of the single mass-to-charge ratio interval; the sequence formed by the total intensity values of all mass-to-charge ratio intervals arranged in order is taken as the spectrum intensity sequence.
[0020] S32, add one or more metadata tokens for characterizing the cross-linking experimental conditions at the starting position of the spectrum intensity sequence, thereby forming a first sequence containing metadata tokens.
[0021] The metadata token includes a cross-linking agent type for indicating the type of cross-linking agent and / or a molecular mass for indicating the relative molecular mass of the cross-linking agent.
[0022] S33, compare the length of the first sequence with the preset length; if the length of the first sequence is less than the preset length, add padding tokens at the end of the first sequence until the length of the first sequence is equal to the preset length; if the length of the first sequence is greater than the preset length, truncate the part exceeding the preset length from the end; thereby obtaining a fixed-length sequence of the preset length.
[0023] S34, add a sequence start token at the starting end of the obtained fixed-length sequence and add a sequence end token at the end of the sequence, thereby obtaining the final fixed-length spectrum characterization sequence.
[0024] In the step S31, if the length of the initial sequence is less than the preset length, padding tokens are used for padding to make all spectrum characterization sequences have the same length.
[0025] The pre-trained peptide segment encoder and the pre-trained spectrum encoder are pre-trained in the pre-training stage by a contrast learning method, and the training steps include: D1, obtain a training data set, the training data set contains a plurality of training samples, each training sample includes an original spectrum generated by a cross-linking mass spectrometry experiment and a plurality of peptide segments corresponding to the original spectrum.
[0026] D2, for each original spectrum, pre-processing the original spectrum to obtain a corresponding peak list, then converting the peak list into a spectrum representation sequence, inputting the spectrum representation sequence into a spectrum encoder to obtain a spectrum embedding vector, taking the obtained spectrum embedding vector as an anchor point of a positive sample of the current original spectrum; using a peptide segment encoder to encode each peptide segment corresponding to the original spectrum into a peptide segment embedding vector, and aggregating all peptide segment embedding vectors to obtain an aggregated peptide segment embedding vector, and taking the aggregated peptide segment embedding vector as the positive sample of the original spectrum; from the same training batch, randomly selecting several other original spectrum corresponding aggregated peptide segment embedding vectors as negative samples of the current original spectrum, thereby obtaining a negative sample set of the current original spectrum.
[0027] D3, calculating a contrastive learning loss function, and updating the peptide segment encoder and the spectrum encoder according to the loss function.
[0028] The aggregation of all peptide segment embedding vectors is specifically: calculating the weighted average of all peptide segment embedding vectors, and taking the weighted average as the aggregated peptide segment embedding vector.
[0029] The contrastive learning loss function is set according to the following formula: Wherein, is the contrastive learning loss function; is a constant; is the anchor point of the positive sample; and are the positive sample and the negative sample respectively; is a cosine similarity function; is the negative sample set; is a temperature parameter.
[0030] The ion matching scoring between each candidate cross-linking peptide segment and the original spectrum is specifically: F1, for each candidate cross-linking peptide segment, based on its amino acid sequence and the chemical properties of the cross-linking agent, a theoretical fragment ion set including single-chain fragment ions and cross-linking fragment ions is generated.
[0031] F2, matching the measured peaks in the peak list corresponding to the original spectrum with the theoretical fragment ion set.
[0032] F3, based on the number of matched measured peaks, the total value of the intensity of all peaks and the preset score of the ion type matched by each peak, the matching score between the candidate cross-linking peptide segment and the original spectrum is calculated.
[0033] The beneficial effects of the present application are: 1. Significantly improve the recognition accuracy of polypeptide segment cross-linking: By modeling the spectrum sequence with a Transformer encoder, long-range dependencies between fragment peaks can be captured, effectively distinguishing different peptide fragment and cross-linking characteristic ions. Through contrastive learning, the possible cross-linked peptide segments are clustered, improving the recognition accuracy in complex spectrum of polypeptide segment cross-linking (such as tripeptide, tetrapeptide cross-linking).
[0034] 2. Greatly reduce the candidate search space: Use vector database for similarity retrieval, convert the originally exponential growth of polypeptide segment combination search problem into approximate nearest neighbor search (ANN) problem, greatly reduce the computational complexity, realize the fast matching on large-scale protein database.
[0035] 3. Good scalability and universality: The method of the present application is not only suitable for traditional dipeptide cross-linking, but also can handle tripeptide, tetrapeptide and other complex cross-linking forms, and has strong universality. At the same time, the retrieval framework of vector database supports fast updating and expansion, and is suitable for cross-linking mass spectrometry data analysis under different species and different experimental conditions. BRIEF DESCRIPTION OF DRAWINGS
[0036] Figure 1 is a flowchart of the method of the present application.
[0037] Figure 2 is a flowchart of the recognition process of the target cross-linked peptide segment. DETAILED DESCRIPTION
[0038] The present application will be described in more detail below in conjunction with the drawings and examples, but the present application is not limited thereto. Those skilled in the art can make several improvements and refinements without departing from the principles of the present application, and these improvements and refinements are also considered within the scope of protection of the present application. The contents not described in detail in the specification are the existing technology known to those skilled in the art.
[0039] As shown in Figure 1 , the method for recognizing protein cross-linking mass spectrometry peptide pairs of the present embodiment comprises the following steps: S1, Constructing a peptide vector database: generating a set of candidate cross-linking peptide segments based on a target protein database, and encoding each candidate cross-linking peptide segment into a corresponding cross-linking peptide segment embedding vector through a pre-trained peptide segment encoder, and storing it in the vector database; the pre-trained peptide segment encoder uses a pre-trained Transformer encoder.
[0040] Generating a set of candidate peptide segments based on a target protein database specifically refers to: processing the target protein database by simulating in vitro enzyme digestion or open cleavage to generate a set of candidate cross-linking peptide segments.
[0041] Each candidate cross-linking peptide segment is a dipeptide cross-linking pair, a tripeptide cross-linking combination, or a tetrapeptide cross-linking combination.
[0042] S2, obtaining and preprocessing spectrum data: obtaining the original spectrum generated by the cross-linking mass spectrometry experiment, and preprocessing to obtain a peak list containing the mass-to-charge ratio and intensity of the peaks and other attributes.
[0043] In step S2, the preprocessing specifically includes peak extraction, isotope removal, charge state deconvolution, noise suppression, and normalization operations in sequence.
[0044] S3, generating a spectrum representation sequence: converting the peak list into a spectrum representation sequence of a fixed length.
[0045] S31, sequencing: uniformly dividing a preset mass-to-charge ratio range into a plurality of consecutive and equally spaced mass-to-charge ratio intervals; for each peak in the peak list, assigning the intensity value corresponding to the peak to the corresponding mass-to-charge ratio interval according to the mass-to-charge ratio of each peak; for each mass-to-charge ratio interval, accumulating and summing the intensity values assigned by all peaks in the interval to obtain a total intensity value of the single mass-to-charge ratio interval; arranging the total intensity values of all mass-to-charge ratio intervals in sequence to form a spectrum intensity sequence; each total intensity value is a unit constituting the spectrum intensity sequence. Thus, an experimental spectrum is represented as a spectrum intensity sequence of a fixed length formed by the total intensity values of the mass-to-charge ratio intervals arranged in sequence.
[0046] In step S31, if the length of the initial sequence is less than the preset length, padding is performed using padding tokens to make all spectrum representation sequences have the same length.
[0047] S32, adding metadata: adding one or more metadata tokens for representing cross-linking experimental conditions at the starting position of the spectrum intensity sequence, thereby forming a first sequence containing the metadata tokens.
[0048] The metadata tokens include a cross-linking agent type for indicating the type of cross-linking agent and / or a molecular mass for indicating the relative molecular mass of the cross-linking agent.
[0049] S33, sequence length processing: comparing the length of the first sequence with the preset length; if the length of the first sequence is less than the preset length, adding padding tokens at the end of the first sequence until the length of the first sequence is equal to the preset length; if the length of the first sequence is greater than the preset length, truncating the part exceeding the preset length from the end; thereby obtaining a fixed-length sequence of the preset length.
[0050] S34, generating a final sequence: adding a sequence start token at the starting end of the obtained fixed-length sequence and adding a sequence end token at the end of the sequence, thereby obtaining a final spectrum representation sequence of a fixed length.
[0051] In this embodiment, the final spectral characterization sequence output is as follows: [BOS][LINKAGENT=DSS][WEIGHT]122[ / WEIGHT][MZ]100.1[ / MZ][INT]500[ / INT][MZ]200.2[ / MZ][INT]3000[ / INT]...[SEP]; Where [BOS] indicates the start of the spectral vector, LINKAGENT indicates the crosslinking agent type, [WEIGHT] indicates the relative molecular mass, [MZ] indicates the mass-to-charge ratio, [INT] indicates intensity information, and [SEP] indicates the end of feature encoding.
[0052] S4. Generate Spectral Embedding Vectors: Input the spectral representation sequence into a pre-trained spectral encoder for processing, and output the spectral embedding vectors corresponding to the original spectra. The pre-trained spectral encoder uses a pre-trained Transformer encoder.
[0053] In specific implementation, after receiving the spectral representation sequence, the second Transformer encoding model first generates a corresponding position code for each unit in the spectral representation sequence. The position code is used to represent the order information of the unit in the sequence. Then, the position code is combined with the initial feature representation of the unit to form an input vector containing position information. Finally, the input vector is encoded to output the spectral embedding vector corresponding to the original spectral image.
[0054] When performing positional encoding, even positions use a sine function: Use the cosine function for odd positions: Among them, PE (pos,2i) and PE (pos,2i+1) , are the scalar values of the position encoding vector at dimensions 2i and 2i+1, respectively; sin() and cos() are the sine and cosine functions, respectively; pos is the position index of the corresponding unit in the input sequence, d model This represents the vector dimension information, where i is the index of the column to which it is mapped.
[0055] like Figure 2 As shown, S5, retrieving candidate peptides: using spectral embedding vectors ( Figure 2 Using the spectral vector in the vector database as the query vector, a similarity search is performed to select the top N most similar cross-linked peptide embedding vectors. Figure 2 The peptide vector in the dataset and its corresponding candidate crosslinked peptides, where N is a preset positive integer.
[0056] In a specific embodiment, the similarity search is performed using approximate nearest neighbor search. The use of vector database for similarity search converts the originally exponential growth polypeptide segment combination search problem into an approximate nearest neighbor search (ANN) problem, greatly reducing the computational complexity and achieving fast matching on large-scale protein databases.
[0057] S6, determining the target cross-linking peptide segment and the cross-linking site: scoring ion matching between each candidate cross-linking peptide segment and the original spectrum, and the candidate cross-linking peptide segment with the highest score is the target cross-linking peptide segment, and the cross-linking site of the target cross-linking peptide segment is obtained.
[0058] The ion matching scoring between each candidate cross-linking peptide segment and the original spectrum is specifically: F1, for each candidate cross-linking peptide segment, based on its amino acid sequence and the chemical properties of the cross-linking agent, a theoretical fragment ion set including single-chain fragment ions and cross-linking fragment ions is generated.
[0059] F2, matching the measured peaks in the peak list corresponding to the original spectrum with the theoretical fragment ion set with mass tolerance.
[0060] F3, based on the number of matched measured peaks, the total value of the intensity of all peaks, and the preset score of the ion type matched by each peak, the matching score between the candidate cross-linking peptide segment and the original spectrum is calculated.
[0061] In a specific embodiment, the matching score between the candidate cross-linking peptide segment and the original spectrum is obtained by weighting and summing the number of measured peaks, the total value of the intensity of all peaks, and the preset weight of the matched ion type.
[0062] Specifically, , wherein is the number of matched measured peaks, and I is the total value of the intensity of all peaks. is the index; is the preset score corresponding to the ion type of the ith matched peak, , and are weight coefficients. Further, , and are obtained by constructing positive samples (high-confidence matching results), negative samples (“false” peptide segments obtained by matching after reversing or randomly scrambling the protein sequence), and training by support vector machine.
[0063] The pre-trained peptide segment encoder and the pre-trained spectrum encoder are pre-trained in the pre-training stage by a contrast learning method, and the training steps include: D1, obtain a training data set, the training data set contains multiple groups of training samples, each group of training samples includes an original spectrum generated by a cross-linking mass spectrometry experiment and multiple peptide segments corresponding to the original spectrum; the multiple peptide segments are cross-linked by a cross-linking agent in the original spectrum.
[0064] D2, for each original spectrum, pre-process the original spectrum to obtain a corresponding peak list, then convert the peak list into a spectrum representation sequence, input the spectrum representation sequence into a spectrum encoder to obtain a spectrum embedding vector, and take the obtained spectrum embedding vector as the positive sample anchor point of the current original spectrum; encode each peptide segment corresponding to the original spectrum into a peptide segment embedding vector using a peptide segment encoder, aggregate all peptide segment embedding vectors to obtain an aggregated peptide segment embedding vector, and take the aggregated peptide segment embedding vector as the positive sample of the original spectrum; randomly select several other original spectra (excluding the current original spectrum) corresponding to the aggregated peptide segment embedding vectors to obtain negative samples of the current original spectrum, thereby obtaining a negative sample set of the current original spectrum.
[0065] The aggregation of all peptide segment embedding vectors is specifically: calculating the weighted average of all peptide segment embedding vectors, and taking the weighted average as the aggregated peptide segment embedding vector.
[0066] The way of constructing positive and negative samples in this step is simple and efficient in calculation. The larger the batch is, the more negative samples are provided, and the better the learning effect is generally.
[0067] D3, calculate a contrastive learning loss function, and update the peptide segment encoder and the spectrum encoder according to the loss function. The loss function is used to increase the similarity between the positive sample anchor point and the corresponding positive sample, and to reduce the similarity between the positive sample anchor point and each negative sample.
[0068] The contrastive learning loss function is set according to the following formula: Wherein, is the contrastive learning loss function; is a constant; is the positive sample anchor point; and are the positive sample and the negative sample, respectively; is a cosine similarity function; is the negative sample set; is a temperature parameter.
[0069] The contrastive learning loss function is calculated based on the cosine similarity between the positive sample anchor point and the positive sample and the negative sample.
[0070] The method of the present application can not only be applied to traditional dipeptide cross-linking, but also can process tripeptide, tetrapeptide and other complex cross-linking forms, and has strong universality. Meanwhile, the retrieval framework of the vector database supports rapid updating and expansion, and is suitable for cross-linking mass spectrometry data analysis under different species and different experimental conditions.
[0071] The above examples are only preferred embodiments for fully illustrating the present application, and the protection scope of the present application is not limited thereto. Any equivalent replacement or transformation made by those skilled in the art on the basis of the present application is within the protection scope of the present application. The protection scope of the present application is subject to the claims.
Claims
1. A method for identifying peptide pairs in protein cross-linking mass spectrometry, characterized in that, Includes the following steps: S1. Construct a peptide vector database: Generate a set of candidate cross-linked peptides based on the target protein database, and encode each candidate cross-linked peptide into a corresponding cross-linked peptide embedding vector through a pre-trained peptide encoder, and store it in the vector database; S2. Acquire and preprocess spectral data: Acquire the raw spectrum generated by the cross-linking mass spectrometry experiment and preprocess it to obtain a list of peaks containing mass-to-charge ratio and intensity. S3. Generate a spectral characterization sequence: Convert the peak list into a fixed-length spectral characterization sequence; S4. Generate spectral embedding vector: Input the spectral representation sequence into the pre-trained spectral encoder for processing, and output the spectral embedding vector corresponding to the original spectrum; S5. Search for candidate peptides: Using the spectral embedding vector as the query vector, perform a similarity search in the vector database to select the top N most similar cross-linked peptide embedding vectors and their corresponding candidate cross-linked peptides. S6. Determine the target cross-linked peptide and cross-linking site: Perform ion matching and scoring between each selected candidate cross-linked peptide and the original spectrum. The candidate cross-linked peptide with the highest score is the target cross-linked peptide, and the cross-linking site of the target cross-linked peptide is obtained.
2. The method for identifying peptide pairs in protein cross-linking mass spectrometry according to claim 1, characterized in that, The specific steps of generating a candidate peptide set based on the target protein database are as follows: The target protein database is processed by simulating in vitro enzymatic cleavage or open cleavage to generate the candidate cross-linked peptide set.
3. The method for identifying peptide pairs in protein cross-linking mass spectrometry according to claim 1, characterized in that: Each candidate cross-linked peptide is a dipeptide cross-linking pair, a tripeptide cross-linking combination, or a tetrapeptide cross-linking combination.
4. The method for identifying peptide pairs in protein cross-linking mass spectrometry according to claim 1, characterized in that: In step S2, the preprocessing specifically includes peak extraction, isotope removal, charge state deconvolution, noise suppression, and normalization operations performed sequentially.
5. The method for identifying peptide pairs in protein cross-linking mass spectrometry according to claim 1, characterized in that, Step S3 specifically involves: S31. Divide the preset mass-to-charge ratio range evenly into multiple continuous, equally spaced mass-to-charge ratio intervals; for each peak in the peak list, allocate the intensity value corresponding to the peak to the corresponding mass-to-charge ratio interval according to the mass-to-charge ratio of each peak; for each mass-to-charge ratio interval, sum the intensity values allocated to all peaks in the interval to obtain the total intensity value of a single mass-to-charge ratio interval; the sequence of total intensity values of all mass-to-charge ratio intervals arranged in order is taken as the spectral intensity sequence; S32. Add one or more metadata tokens to the starting position of the spectral intensity sequence to characterize the crosslinking experimental conditions, thereby forming a first sequence containing metadata tokens; The metadata token includes the crosslinking agent type used to indicate the crosslinking agent type and / or the molecular mass used to indicate the relative molecular mass of the crosslinking agent; S33. Compare the length of the first sequence with a preset length; if the length of the first sequence is less than the preset length, add a padding mark to the end of the first sequence until the length of the first sequence is equal to the preset length. If the length of the first sequence is greater than the preset length, then the portion exceeding the preset length is truncated from the end; This results in a fixed-length sequence of a preset length; S34. Add a sequence start marker to the beginning of the obtained fixed-length sequence and add a sequence end marker to the end of the sequence to obtain the final fixed-length spectral characterization sequence.
6. The method for identifying peptide pairs in protein cross-linking mass spectrometry according to claim 5, characterized in that: In step S31, if the length of the initial sequence is less than the preset length, padding tokens are used to pad the sequence so that all spectral representation sequences have the same length.
7. The method for identifying peptide pairs in protein cross-linking mass spectrometry according to claim 1, characterized in that, The pre-trained peptide encoder and the pre-trained spectral encoder are pre-trained through a comparative learning method during the pre-training phase. The training steps include: D1. Obtain the training dataset, which contains multiple training samples. Each training sample includes an original spectrum generated by a cross-linking mass spectrometry experiment and multiple peptide segments corresponding to the original spectrum. D2. For each original spectrum, preprocess the original spectrum to obtain the corresponding peak list, then convert the peak list into a spectrum representation sequence, input the spectrum representation sequence into the spectrum encoder to obtain the spectrum embedding vector, and use the obtained spectrum embedding vector as the positive sample anchor point of the current original spectrum; use a peptide encoder to encode each peptide corresponding to the original spectrum into a peptide embedding vector, and aggregate all peptide embedding vectors to obtain an aggregated peptide embedding vector, and use the aggregated peptide embedding vector as the positive sample of the original spectrum; from the same training batch, randomly select several aggregated peptide embedding vectors corresponding to other original spectra and use them as negative samples of the current original spectrum, thus obtaining the negative sample set of the current original spectrum; D3. Calculate the contrastive learning loss function and update the peptide encoder and spectral encoder based on the loss function.
8. The method for identifying peptide pairs in protein cross-linking mass spectrometry according to claim 7, characterized in that, The aggregation of all peptide embedding vectors specifically involves: Calculate the weighted average of all peptide embedding vectors and use the weighted average as the aggregated peptide embedding vector.
9. The method for identifying peptide pairs in protein cross-linking mass spectrometry according to claim 7, characterized in that: The contrastive learning loss function is set according to the following formula: in, To learn the loss function by comparison; It is a constant; Anchor points for positive samples; and These are positive and negative samples, respectively. The cosine similarity function; For the negative sample set; This refers to the temperature parameter.
10. The method for identifying peptide pairs in protein cross-linking mass spectrometry according to claim 1, characterized in that, The process of performing ion matching and scoring between each candidate cross-linked peptide and the original spectrum is as follows: F1. For each candidate crosslinked peptide, based on its amino acid sequence and the chemical properties of the crosslinking agent, a theoretical fragment ion set including single-chain fragment ions and crosslinked fragment ions is generated. F2. Match the measured peaks in the peak list corresponding to the original spectrum with the theoretical fragment ion set; F3. Based on the number of matched measured peaks, the total intensity of all peaks, and the preset score of the ion type matched by each peak, calculate the matching score between the candidate crosslinked peptide and the original spectrum.