Method and system for automatic identification of lipid double bond position based on ozone dissociation mass spectrometry data

Abstract

This invention discloses an automatic identification method and system for lipid double bond positions based on ozone dissociation mass spectrometry data. The method includes: S1 constructing an in-silico ozone lipid fragment database; S2 preprocessing the raw mass spectrometry data and screening to obtain candidate features and auxiliary features; S3 retrieving ozone fragment features from the auxiliary features based on retention time and mass difference constraints; S4 constructing a directed bipartite graph of candidate features and fragment features; S5 scoring edges through multi-dimensional similarity calculation and removing redundancy by attribute combination; S6 comprehensively scoring the candidate lipid structures and outputting the double bond position identification results. This invention achieves high-throughput, low-false-positive automatic identification of double bond positions in multiple lipids and large-scale lipids through broad database coverage, multi-dimensional matching scoring, and redundancy removal strategies. It has strong structural resolution, is compatible with common mass spectrometry data formats, can be integrated with standard workflows, and is suitable for structural lipidomics research.

Description

Technical Field

[0001] This invention relates to the fields of structural lipidomics and mass spectrometry data analysis, specifically to a method and system for automatic identification of lipid double bond positions based on ozone dissociation mass spectrometry data. Background Technology

[0002] Lipids, as important biomolecules in living organisms, exhibit structural diversity (especially carbon-carbon double bond positional isomerism) that is closely related to physiological functions. Accurately determining the double bond positions of lipids is one of the core objectives of structural lipidomics research. Currently, various technical approaches have been reported for structural determination of the C=C position in lipids, including the Pastenò–Büchi reaction, ultraviolet photodissociation (UVPD), electron-related dissociation (EAD), and traditional ozone-induced dissociation (OzID).

[0003] However, existing technologies have significant shortcomings: on the one hand, the aforementioned methods mostly rely on specialized hardware modifications or complex experimental conditions, limiting their versatility; on the other hand, existing publicly available software tools primarily focus on the identification of lipid head groups and acyl chains, lacking automated structural analysis capabilities for large-scale ozone dissociation data. Specifically, existing database-based structural lipidomics tools have the following deficiencies: 1) They support a limited range of lipid types, lacking fragment libraries covering multiple lipids and large-scale C=C position isomers; 2) They cannot effectively combine MS obtained from conventional ionization modes. 1 / MS 2 Information on aldehyde / Criegee fragments under ozone mode leads to insufficient structural analysis dimensions; 3) The lack of a systematic fragment matching and scoring algorithm makes it difficult to distinguish multiple structural isomers under the same m / z, which easily results in a high false positive rate.

[0004] Therefore, there is an urgent need to develop an algorithm and supporting software that can automatically identify the C=C double bond positions of lipids from ozone dissociation mass spectrometry data, so as to achieve accurate and high-throughput analysis of multi-lipid and large-scale structural lipidomics data. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an automatic identification method and system for lipid double bond positions based on ozone dissociation mass spectrometry data. By constructing a large-scale in-silico ozone dissociation lipid fragment database and designing a multi-dimensional matching and comprehensive scoring algorithm, high-throughput, low-false-positive automated identification of multiple lipids and multiple C=C position isomers can be achieved.

[0006] The technical solution provided by this invention is as follows: This invention provides an automatic identification method for lipid double bond positions based on ozone dissociation mass spectrometry data, characterized by the following steps: S1: Construct an in-silico ozone lipid fragment database, which contains theoretical information on the lipid precursor ion and the characteristic fragments generated by its corresponding ozone dissociation. The theoretical information includes at least the theoretical mass number, the theoretical mass difference between the precursor ion and the fragment, and the corresponding carbon-carbon double bond position. S2: Preprocess the raw mass spectrometry data to obtain the MS¹ peak table and corresponding MS. 2 Spectrum, based on MS 2 Candidate features and auxiliary features are obtained through spectral library matching and screening. The candidate features are MSs with matching scores higher than a first preset threshold. 1 Peak, the auxiliary feature is the MS that does not meet the candidate feature selection criteria. 1 peak; S3: Based on preset constraints, retrieve ozone fragment features associated with the candidate features from the auxiliary features; S4: Using the candidate features and ozone fragment features as two node sets, and the relationship between the two as edges, construct a directed bipartite graph. The attributes of each edge include at least the corresponding lipid structure information, carbon-carbon double bond position and fragment type. S5: Calculate the segment similarity score corresponding to each edge in the directed bipartite graph. The score is constructed based on matching indicators of at least two dimensions. Redundancy is removed from edges corresponding to the same attribute combination, and the edge with the best score is retained. S6: Based on the redundancy-free directed bipartite graph, the lipid structure corresponding to each candidate feature is comprehensively scored, and the identification result of the lipid carbon-carbon double bond position is output according to the scoring result.

[0007] Further, in step S1, the construction process of the in-silico ozone lipid fragment database includes: S11: Obtain a lipid list containing lipid category, acyl chain length and unsaturation information from the lipid database, and retain lipid structure information with carbon-carbon double bond position annotations; S12: For each lipid structure, the carbon-carbon double bond positions in its acyl chain are analyzed, and for each double bond position, the ozone dissociation process is simulated to generate aldehyde fragment ions and / or Criegee fragment ions. S13: Record the theoretical mass number of the lipid precursor ion, the theoretical mass number of the characteristic fragment, the theoretical mass difference between the precursor ion and the fragment, the corresponding carbon-carbon double bond position and fragment type, and serialize and store them to form the database.

[0008] Furthermore, in step S1, the database covers at least 12 types of lipids, including Cer, SM, PA, PE, PC, PG, PI, PS, BMP, DG, and TG, with a total number of records reaching 220,000.

[0009] Further, in step S2, the preprocessing includes quality detection, peak extraction, feature decomposition, and retention time alignment; in the selection of candidate features, the first preset threshold ranges from 0.7 to 0.9; the retention time RT, precursor m / z, and MS of each candidate feature are recorded. 2 Spectral vectors and preliminary identification of lipid types Further, in step S3, the preset constraints include: Constraint 1: The absolute value of the retention time difference between candidate features and auxiliary features is less than the second preset threshold ΔRT; Constraint 2: The absolute value of the mass difference between the candidate feature and the auxiliary feature and the theoretical mass difference between the corresponding parent ion and fragment in the database is less than the third preset threshold Δm / z.

[0010] Furthermore, in step S5, the multi-dimensional matching metrics include retention time similarity, quality difference similarity, and MS (Mean Search Scale). 2 At least two of the spectral similarities; the segment similarity score is a weighted linear combination of the matching indicators of each dimension.

[0011] Furthermore, the retention time similarity is constructed as a monotonically decreasing function based on the relationship between the retention time difference and a second preset threshold ΔRT; the quality difference similarity is constructed as a monotonically decreasing function based on the relationship between the quality difference deviation and a third preset threshold Δm / z; the MS 2 Spectral similarity is calculated using weighted cosine similarity or correlation coefficient.

[0012] Furthermore, in step S5, the same attribute combination is a combination of "lipid structure information + carbon-carbon double bond position + fragment type"; the redundancy removal process is to retain the edge with the highest fragment similarity score under the same attribute combination.

[0013] Further, in step S6, the comprehensive score is the sum of the segment similarity scores of all edges corresponding to the candidate feature, or the value after normalization of the number of segments; the output result also includes lipid structure abbreviation, comprehensive score, and optional relative abundance of each isoform.

[0014] This invention also provides an automatic lipid double bond position identification system based on ozone dissociation mass spectrometry data, employing the aforementioned automatic lipid double bond position identification method, characterized in that it includes: A database building module for constructing an in-silico ozone lipid fragment database; The data preprocessing module is used to preprocess the raw mass spectrometry data to obtain MS data. 1 Peak table and corresponding MS 2 Spectrum, based on MS 2Candidate features and auxiliary features are obtained through spectral library matching and screening; The fragment retrieval module is used to retrieve ozone fragment features associated with the candidate features from the auxiliary features under preset constraints. The graph construction module is used to construct a directed bipartite graph with the candidate features and ozone fragment features as two node sets and the relationship between the two as edges. A redundancy removal module is used to calculate the segment similarity score corresponding to each edge in the directed bipartite graph. The scoring output module is used to comprehensively score the lipid structure corresponding to each candidate feature based on the redundancy-free directed bipartite graph, and output the identification result of the lipid carbon-carbon double bond position according to the scoring result.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention utilizes in-silico ozone dissociation simulation to construct an ozone fragment database containing approximately 220,000 records, covering more than 12 lipids. Compared to existing public databases, this database offers more comprehensive coverage and can meet the needs of multi-lipid analysis. This invention combines conventional and ozone model spectral information to design retention time, quality difference, and MS (magnetic flux density) parameters. 2 The spectral three-dimensional scoring system, combined with a fragment redundancy removal strategy, effectively reduces the mismatch rate. In tests of standards and cell samples, it can correctly determine structures with known C=C positions (such as...). Figure 3 As shown in the figure, stable measurement results with a coefficient of variation of less than 10% were obtained in HeLa cells. This invention, through a quantitative deconvolution algorithm, can distinguish lipid isoforms at different C=C positions at the same m / z, accurately estimate the relative abundance of each isoform, and provide technical support for the discovery of novel structural lipids. Based on the universal mzML / MGF data format, this invention is compatible with common high-resolution mass spectrometry platforms and can interface with the standard LC-OzID workflow, achieving one-click data processing and significantly improving analytical efficiency. Attached Figure Description

[0016] Figure 1 This is a flowchart of the automatic identification method for lipid double bond positions based on ozone dissociation mass spectrometry data in an embodiment of the present invention; Figure 2 This is a framework diagram of the lipid double bond position automatic identification system based on ozone dissociation mass spectrometry data in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the analytical principle of lipid double bond position identification tests in biological samples in an embodiment of the present invention; Figure 4 This is a mass spectrometry spectrum with double bond characteristics generated by the ozone decomposition method of the corresponding PC 20:4 standard in the embodiments of the present invention.

[0017] The attached figures are labeled as follows: 1- Database construction module, 2- Data preprocessing module, 3- Fragment retrieval module, 4- Graph construction module, 5- Redundancy removal module, 6- Scoring output module, 100- Automatic lipid double bond position identification system. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments.

[0019] For ease of explanation, the following terminology is used in this invention: 1. Candidate feature: refers to a feature selected through regular MS (Multi-Site Analysis). 2 MS spectroscopic libraries (such as LipidBlast) were used for preliminary identification and to assign lipid species and molecular formula information. 1 Peak. The set consisting of all candidate features is denoted as U.

[0020] 2. Anchor ion: refers to the MS candidate feature whose matching score with the spectral library is higher than a preset threshold (e.g., 0.8). 1 The peak serves as an "anchor" for subsequent ozone fragment associations.

[0021] 3. Auxiliary features: These refer to MS features that have not been preliminarily identified by the spectral library. 1 The peak is used as a potential ozone splitting fragment signal. The set of all auxiliary features is denoted as F.

[0022] 4. Ozone fragment feature: refers to the MS signal corresponding to the aldehyde fragment ion or Criegee fragment ion generated by ozone dissociation. 1 Peak. The set V is denoted as all auxiliary features selected as candidate segments.

[0023] 5. Lipid structure abbreviation: refers to information including lipid type, acyl chain composition and C=C position, such as "PC 16:0_18:1(Δ-9)".

[0024] 6. In-silico Ozone Lipid Fragment Database: This refers to a collection of information such as the theoretical m / z, Δm / z, and C=C positions of lipid precursor ions and their corresponding aldehyde / Criegee fragments obtained by calculating and simulating ozone fragmentation of lipid structures.

[0025] See Figure 1This invention provides an automatic identification method for lipid double bond positions based on ozone dissociation mass spectrometry data, comprising the following steps: S1: Constructing an in-silico ozone lipid fragment database The database contains theoretical information on lipid precursor ions and the characteristic fragments generated by their corresponding ozone dissociation. The theoretical information includes at least the theoretical mass number (m / z), the theoretical mass difference between the precursor ion and the fragment (Δm / z), and the corresponding carbon-carbon double bond position (Δ-position).

[0026] The specific construction process is as follows: S11: Obtain a list of lipids containing information on lipid class, acyl chain length and unsaturation from public lipid databases (such as LIPID MAPS), and retain lipid structure abbreviations with explicit C=C positions (such as "PC 16:0_18:1(Δ-9)"). S12: For each lipid structure abbreviation, analyze each C=C position in all its acyl chains. For each C=C position, according to the classical ozone dissociation rule, simulate the breaking process at the double bond to generate a pair of theoretical fragments composed of aldehyde fragment ions and Criegee fragment ions. For each pair of theoretical segments, calculate and record the following information: Theoretical m / z of the parent ion; Aldehyde fragment ion theory m / z; Criegee fragment ion theory m / z; The theoretical mass difference Δm / z between the parent ion and the fragment; The corresponding C=C position (Δ position, such as Δ-6, Δ-9, etc.); Fragment type (aldehyde or Criegee).

[0027] S13: Record the theoretical m / z of the parent ion, the theoretical m / z of the aldehyde fragment ion, the theoretical m / z of the Criegee fragment ion, the theoretical Δm / z of the parent ion and fragment, the corresponding C=C position, and the fragment type (aldehyde or Criegee) for each lipid. Serialize and store this information as a searchable data structure (such as a JSON file or database table) to form an in-silico ozone lipid fragment database. This database preferably covers more than 12 lipid types, including Cer, SM, PA, PE, PC, PG, PI, PS, BMP, DG, and TG, with a total of approximately 220,000 records.

[0028] S2: Preprocess raw mass spectrometry data and screen candidate and auxiliary features. S21: Convert the raw mass spectrometry data acquired from the high-resolution mass spectrometer into a standard format (e.g., mzML), import it into mass spectrometry data processing software for preprocessing, including mass detection, peak extraction, feature decomposition, and retention time alignment, to obtain the MS data. 1 Peak table and corresponding MS 2 Spectrum files (such as MGF format); S22: Using MS 2 Spectral libraries (such as LipidBlast) for the above MS 2 The spectra are matched, and the similarity score (range 0~1) between each MS¹ peak and the library spectrum is calculated. S23: Select MSs with similarity scores higher than a first preset threshold (preferably 0.8). 1 Peaks are defined as candidate features (set U), and their retention time (RT), precursor m / z, and MS are recorded. 2 Spectral vectors and preliminary lipid type information; the remaining MS that were not successfully matched or whose scores were below the first preset threshold. 1 Peaks are used as auxiliary features (set F).

[0029] S3: Retrieve ozone fragment features from auxiliary features For each candidate feature u∈U, retrieve potential ozone fragment features v∈V from the auxiliary feature set F. The retrieval must satisfy the following two constraints: Constraint 1 (Retention Time Constraint): The absolute value of the difference between the retention times of candidate features and auxiliary features |RT_u RT_v| is less than the second preset threshold ΔRT (preferably 0.15 min); Constraint 2 (Quality Difference Constraint): The difference in quality number (m / z_u) between candidate features and auxiliary features. The absolute value of the deviation between m / z_v and the theoretical Δm / z obtained from the database constructed from S1 |(m / z_u) m / z_v) Δm / z_theoretical| is less than the third preset threshold Δm / z (preferably 10 ppm); The auxiliary feature v that satisfies the above constraints is included in the ozone fragment feature set V and a candidate association relationship is established with the candidate feature u. At the same time, the lipid structure abbreviation, C=C position and fragment type corresponding to the association relationship are recorded.

[0030] S4: Construct a directed bipartite graph of candidate features and fragment features. Using candidate feature set U and ozone fragment feature set V as two node sets, construct a directed bipartite graph G=(U, V,E, A), where: U represents the candidate feature node set, and V represents the ozone fragment feature node set; E is a set of edges, where each edge e(u, v) indicates that there is an ozone fragment matching association between candidate feature u and ozone fragment feature v; A is the set of edge attributes. The attributes of each edge e(u, v) include at least the corresponding lipid structure abbreviation L (e.g., PC16:0_18:1(Δ-9)), C=C position Δ-pos (e.g., Δ-6, Δ-9, etc.), and fragment type (aldehyde or Criegee). The construction of this directed bipartite graph must satisfy the following conditions: each edge e(u, v) conforms to the constraints of S3; the same candidate feature u can connect to multiple fragment features v; the same fragment feature v can connect to multiple candidate features u.

[0031] S5: Segment Similarity Calculation and Redundancy Removal S51: For each edge e(u, v) in the directed bipartite graph, calculate the total fragment similarity score Score_frag(u, v), which is obtained by a weighted combination of multi-dimensional matching indicators, preferably including: Retention time similarity Score_RT(u, v): based on |RT_u The relationship between RT_v| and ΔRT is used to construct a monotonically decreasing function, when |RT_u When RT_v|≥ΔRT, Score_RT=0; when |RT_u When RT_v approaches 0, Score_RT approaches 1; Quality difference similarity Score_mz(u, v): Based on the relationship between quality difference deviation and Δm / z, a monotonically decreasing function is constructed. When the deviation is ≥ Δm / z, Score_mz = 0. The smaller the deviation, the closer Score_mz is to 1. MS 2 Spectral similarity Score_spec(u, v): This calculates the mean spectral similarity between candidate feature u and fragment feature v. 2 The spectra are aligned along the m / z axis, and the peak intensity vectors are normalized and weighted. The scores between 0 and 1 are calculated using weighted cosine similarity or correlation coefficient. The formula for calculating the total segment similarity score is: Score_frag(u, v) = w_RT·Score_RT(u, v) + w_mz·Score_mz(u, v) + w_spec·Score_spec(u, v) Among them, w_RT, w_mz, and w_spec are weighting coefficients, preferably set to 0.1, 0.1, and 0.8 to highlight MS. 2 The contribution of spectral similarity.

[0032] S52: Redundancy removal: For the same attribute combination of "lipid structure L+C=C position Δ-pos+fragment type", in all the corresponding edge sets, only the edge with the highest Score_frag(u, v) is retained, and the rest of the edges and the corresponding fragment feature nodes are deleted to obtain the pruned directed bipartite graph G'=(U, V', E', A').

[0033] S6: Overall Scoring and Result Output S61: For each candidate feature u∈U, in the pruned graph G', summarize the Score_frag(u, v) of all edges connected to u, and normalize it according to the number of fragments N_frag(u) to obtain the comprehensive score Score_L(u) of the candidate feature under the corresponding lipid structure L. S62: For each candidate feature u, a candidate structure set is formed for the multiple lipid structures L. {L_1, L_2, …, L_k} and their combined scores {Score_L1(u), …, Score_Lk(u)}: Candidate structures with a comprehensive score lower than the fourth preset threshold (preferably 0.1) are removed. The remaining candidate structures are sorted from high to low according to Score_L(u), and the top N structures with the highest scores are retained (preferably N=3). S63: When the same MS 1 When a peak is assigned to multiple C=C position isomers, the MS is determined based on the relative magnitude of the Score_L(u) of each isomer. 1 The observed intensity of the peak is allocated to different structures according to a weight ratio to achieve quantitative deconvolution; S64: Output recognition results, including but not limited to: each MS 1 The peak corresponds to the lipid structure abbreviation, the position and type of the C=C double bond, the overall score Score_L(u), and the relative abundance of each isomer (optional).

[0034] The final output includes, but is not limited to: Each MS 1 The abbreviation of the lipid structure corresponding to the peak; The position and type of the C=C double bond in each lipid (e.g., Δ-6, Δ-9); The corresponding overall score is Score_L(u); If deconvolution is performed, the relative abundance of each isomer is also included.

[0035] like Figure 2 As shown, the present invention also provides an automatic lipid double bond position identification system 100 based on ozone dissociation mass spectrometry data, which employs the above-mentioned automatic lipid double bond position identification method, including: Database construction module 1 is used to build an in-silico ozone lipid fragment database; Data preprocessing module 2 is used to preprocess the raw mass spectrometry data to obtain MS data. 1 Peak table and corresponding MS 2 Spectrum, based on MS 2 Candidate features and auxiliary features are obtained through spectral library matching and screening; Fragment retrieval module 3 is used to retrieve ozone fragment features associated with the candidate features from the auxiliary features under preset constraints; Graph construction module 4 is used to construct a directed bipartite graph with the candidate features and ozone fragment features as two node sets and the relationship between the two as edges. Redundancy removal module 5 is used to calculate the segment similarity score corresponding to each edge in the directed bipartite graph; The scoring output module 6 is used to comprehensively score the lipid structure corresponding to each candidate feature based on the redundancy-free directed bipartite graph, and output the identification result of the lipid carbon-carbon double bond position according to the scoring result.

[0036] Example This embodiment provides an automatic lipid double bond position identification method based on the above method. The method uses an automatic lipid double bond position identification system 100 based on ozone dissociation mass spectrometry data, and specifically includes the following operations: 1. Collect sample data Lipid standards with known C=C positions, PC 20:4(5,8,11,14)_20:4(5,8,11,14), were prepared into 0.5 μM solutions. Data were acquired using a liquid chromatography-high resolution mass spectrometry system (such as Q Exactive Plus) in two modes: Standard mode: Nitrogen is used as the sheath gas; Ozone mode: Ozone is used as the sheath gas; The acquisition parameters are set to: MS 1 70,000 resolution, MS 2 Resolution 17,500, scan range m / z 200~1800.

[0037] A database of in-silico ozone lipid fragments is constructed using database construction module 1.

[0038] 2. Mass spectrometry data preprocessing is performed using data preprocessing module 2. The raw mass spectrometry data were converted to mzML format and imported into MZmine3 software for preprocessing, including quality detection, peak extraction, feature decomposition, and retention time alignment, to obtain the MS data. 1 Peak table and MS in MGF format2 Spectral files; MS spectra were analyzed using the LipidBlast spectral library. 2 Spectrum matching was performed, and MS samples with a similarity score ≥ 0.8 were identified. 1 The peak is defined as the candidate feature set U, and the remaining features are the auxiliary feature set F. The RT, precursor m / z, and MS of the candidate features are recorded. 2 Spectral vectors and preliminary identification of lipid types.

[0039] 3. Retrieve theoretical fragment information using fragment retrieval module 3. For each candidate feature u∈U, retrieve fragment features v from the auxiliary feature set F that satisfy the following conditions. To ensure physical consistency between the precursor ion features and ozone pyrolysis fragments in both chromatographic and mass dimensions, only fragment features that simultaneously satisfy the following conditions are retained: (1) Retention time difference |RT_u RT_v|<0.15 min, this threshold is set according to the lipid chromatography peak width in the experiment and is used to limit the co-elution characteristics; (2) Quality difference deviation |(m / z_u m / z_v) Δm / z_theoretical|<10 ppm, where The threshold is set based on the theoretical mass difference calculated from the lipid structure and its ozone decomposition rules, and is determined according to the mass accuracy of the high-resolution mass spectrometer.

[0040] The in-silico ozone lipid fragment database used in this embodiment contains 222,460 records, covering 12 lipids including Cer, SM, PA, PE, PC, PG, PI, PS, BMP, DG, and TG. The database is stored in JSON format, and the corresponding theoretical ozone cleavage fragment information can be quickly retrieved based on the lipid structure abbreviation or the m / z value of the parent ion.

[0041] 4. Construct a directed bipartite graph using Graph Construction Module 4. Incorporate the segments v that meet the conditions into the feature set V, and construct a directed bipartite graph. G=(U, V, E, A), in Represents the set of candidate features. Represents the set of fragment features. This represents the set of edges connecting candidate features and fragment features. A set of attributes representing an edge. Each edge For a specific lipid structure hypothesis, its attributes include lipid structure abbreviation (e.g., "PC 20:4(5,8,11,14)_20:4(5,8,11,14)"), carbon-carbon double bond position (Δ-5, Δ-8, Δ-11, Δ-14), and ozone fragmentation type (aldehyde ion or Criegee ion).

[0042] 5. Perform segment similarity calculation and redundancy removal using redundancy removal module 5. For each edge in a directed bipartite graph The overall fragment similarity score, Score_frag, is calculated by weighting and summing retention time similarity, quality deviation similarity, and fragment spectrum similarity. The weights for each factor are set as follows: , and Score_RT is based on |RT_u RT_v| is calculated linearly relative to 0.15 min, Score_mz is calculated linearly relative to 10 ppm based on the mass deviation, and Score_spec evaluates the spectral similarity between the precursor ion and fragment features using a weighted cosine similarity method.

[0043] To avoid multiple interpretations of the same lipid-specific fragment, for edges with the same lipid structure abbreviation, the same C=C position, and the same fragment type (e.g., "PC 20:4(5,8,11,14)_20:4(5,8,11,14)+Δ-5+aldehyde group"), only the edge with the highest Score_frag is retained, and the rest are deleted, thus completing the redundancy removal and pruning of the directed bipartite graph.

[0044] 6. Perform comprehensive scoring and output results through scoring output module 6. In the pruned directed bipartite graph, for each candidate feature and the corresponding lipid structure hypothesis The Score_frag of all fragment edges that support the structure is summarized, and the number of fragments is normalized to obtain the comprehensive score Score_L(u,lipid) of the candidate feature under the lipid structure.

[0045] In this embodiment, candidate features The overall score for the lipid structure PC 20:4(5,8,11,14)_20:4(5,8,11,14) was 0.85, which is higher than the preset threshold of 0.1. This result is completely consistent with the known structure of the standard, verifying the accuracy of the method of this invention in lipid double bond position determination. The corresponding ozone decomposition characteristic fragment of the PC 20:4 standard is shown in Figure 4.

[0046] This invention utilizes in-silico ozone dissociation simulation to construct an ozone fragment database containing approximately 220,000 records, covering more than 12 lipids. Compared to existing public databases, this database offers more comprehensive coverage and can meet the needs of multi-lipid analysis. This invention combines conventional and ozone model spectral information to design a three-dimensional scoring system based on retention time, quality difference, and MS² spectral density. Combined with fragment redundancy removal strategies, this effectively reduces mismatch rates. In tests on standards and cell samples, it can correctly determine structures with known C=C positions (such as…). Figure 3 As shown in the figure, stable measurement results with a coefficient of variation of less than 10% were obtained in HeLa cells. This invention, through a quantitative deconvolution algorithm, can distinguish lipid isoforms at different C=C positions at the same m / z, accurately estimate the relative abundance of each isoform, and provide technical support for the discovery of novel structural lipids. Based on the universal mzML / MGF data format, this invention is compatible with common high-resolution mass spectrometry platforms and can interface with the standard LC-OzID workflow, achieving one-click data processing and significantly improving analytical efficiency.

[0047] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automatic identification method for lipid double bond positions based on ozone dissociation mass spectrometry data, characterized in that, Includes the following steps: S1: Construct an in-silico ozone lipid fragment database, which contains theoretical information on the lipid precursor ion and the characteristic fragments generated by its corresponding ozone dissociation. The theoretical information includes at least the theoretical mass number, the theoretical mass difference between the precursor ion and the fragment, and the corresponding carbon-carbon double bond position. S2: Preprocess the raw mass spectrometry data to obtain MS 1 Peak table and corresponding MS 2 Spectrum, based on MS 2 Candidate features and auxiliary features are obtained through spectral library matching and screening. The candidate features are MSs with matching scores higher than a first preset threshold. 1 Peak, the auxiliary feature is the MS that does not meet the candidate feature selection criteria. 1 peak; S3: Based on preset constraints, retrieve ozone fragment features associated with the candidate features from the auxiliary features; S4: Using the candidate features and ozone fragment features as two node sets, and the relationship between the two as edges, construct a directed bipartite graph. The attributes of each edge include at least the corresponding lipid structure information, carbon-carbon double bond position and fragment type. S5: Calculate the segment similarity score corresponding to each edge in the directed bipartite graph. The score is constructed based on matching indicators of at least two dimensions. Redundancy is removed from edges corresponding to the same attribute combination, and the edge with the best score is retained. S6: Based on the redundancy-free directed bipartite graph, the lipid structure corresponding to each candidate feature is comprehensively scored, and the identification result of the lipid carbon-carbon double bond position is output according to the scoring result.

2. The automatic lipid double bond position identification method according to claim 1, characterized in that, In step S1, the construction process of the in-silico ozone lipid fragment database includes: S11: Obtain a lipid list containing lipid category, acyl chain length and unsaturation information from the lipid database, and retain lipid structure information with carbon-carbon double bond position annotations; S12: For each lipid structure, the carbon-carbon double bond positions in its acyl chain are analyzed, and for each double bond position, the ozone dissociation process is simulated to generate aldehyde fragment ions and / or Criegee fragment ions. S13: Record the theoretical mass number of the lipid precursor ion, the theoretical mass number of the characteristic fragment, the theoretical mass difference between the precursor ion and the fragment, the corresponding carbon-carbon double bond position and fragment type, and serialize and store them to form the database.

3. The method for automatic identification of lipid double bond positions according to claim 2, characterized in that, In step S1, the database covers at least 12 types of lipids, including Cer, SM, PA, PE, PC, PG, PI, PS, BMP, DG, and TG, with a total of 220,000 records.

4. The method for automatic identification of lipid double bond positions according to claim 1, characterized in that, In step S2, the preprocessing includes quality detection, peak extraction, feature decomposition, and retention time alignment; in the selection of candidate features, the first preset threshold ranges from 0.7 to 0.9; the retention time RT, precursor m / z, and MS of each candidate feature are recorded. 2 Spectral vectors and preliminary identification of lipid types.

5. The method for automatic identification of lipid double bond positions according to claim 4, characterized in that, In step S3, the preset constraints include: Constraint 1: The absolute value of the retention time difference between candidate features and auxiliary features is less than the second preset threshold ΔRT; Constraint 2: The absolute value of the mass difference between the candidate feature and the auxiliary feature and the theoretical mass difference between the corresponding parent ion and fragment in the database is less than the third preset threshold Δm / z.

6. The method for automatic identification of lipid double bond positions according to claim 5, characterized in that, In step S5, the multi-dimensional matching metrics include retention time similarity, quality difference similarity, and MS (Mean Search Scale). 2 At least two of the spectral similarities; the segment similarity score is a weighted linear combination of the matching indicators of each dimension.

7. The method for automatic identification of lipid double bond positions according to claim 6, characterized in that, The retention time similarity is constructed as a monotonically decreasing function based on the relationship between the retention time difference and a second preset threshold ΔRT; the quality difference similarity is constructed as a monotonically decreasing function based on the relationship between the quality difference deviation and a third preset threshold Δm / z; the MS 2 Spectral similarity is calculated using weighted cosine similarity or correlation coefficient.

8. The method for automatic identification of lipid double bond positions according to any one of claims 5-7, characterized in that, In step S5, the same attribute combination is a combination of "lipid structure information + carbon-carbon double bond position + fragment type"; the redundancy removal process is to retain the edge with the highest fragment similarity score under the same attribute combination.

9. The method for automatic identification of lipid double bond positions according to claim 1, characterized in that, In step S6, the comprehensive score is the sum of the fragment similarity scores of all edges corresponding to the candidate feature, or the value after normalization of the number of fragments; the output results also include lipid structure abbreviation, comprehensive score, and optional relative abundance of each isoform.

10. An automatic lipid double bond position identification system based on ozone dissociation mass spectrometry data, employing the automatic lipid double bond position identification method as described in any one of claims 1-9, characterized in that, include: A database building module for constructing an in-silico ozone lipid fragment database; The data preprocessing module is used to preprocess the raw mass spectrometry data to obtain the MS¹ peak table and corresponding MS. 2 Spectrum, based on MS 2 Candidate features and auxiliary features are obtained through spectral library matching and screening; The fragment retrieval module is used to retrieve ozone fragment features associated with the candidate features from the auxiliary features under preset constraints. The graph construction module is used to construct a directed bipartite graph with the candidate features and ozone fragment features as two node sets and the relationship between the two as edges. A redundancy removal module is used to calculate the segment similarity score corresponding to each edge in the directed bipartite graph. The scoring output module is used to comprehensively score the lipid structure corresponding to each candidate feature based on the redundancy-free directed bipartite graph, and output the identification result of the lipid carbon-carbon double bond position according to the scoring result.

Images