Mass spectrum-based maternal metabolite structure labeling method, system and application

Abstract

The invention discloses a mass spectrum-based maternal metabolite structure labeling method, which comprises the following steps of: inputting a structure SMILES character string of a maternal metabolite, and outputting a metabolite structure and a reaction type of the metabolite structure; calculating the mass-to-charge ratios of all output metabolites in different adduct ion forms, and performing correlation matching and annotation on the mass-to-charge ratios and a pre-constructed metabolite peak table to obtain candidate metabolites and candidate metabolite adduct ion forms; performing multi-dimensional scoring on the obtained candidate metabolites; screening the metabolite list according to a set filtering rule in combination with scores of multiple dimensions; and outputting annotation information and structure information of the metabolites in the peak table to a file, wherein the annotation information comprises a reaction type, an adduct ion form, a multi-dimensional score and a structure. The invention further discloses a labeling system and application for implementing the labeling method.

Description

Technical Field

[0001] The present invention belongs to the fields of biomedicine and biochemistry, and relates to a method, system and application for structural annotation of parent metabolites based on mass spectrometry. Background Art

[0002] Metabolites are the products of chemical reactions within organisms. They participate in biological processes such as signal transduction, communication, energy production, and metabolism. They can also serve as drugs, biomarkers, and nutrients to meet human needs. Understanding the roles and functions of metabolites is crucial for understanding many fields of biology, biomedicine, and biochemistry. However, the metabolic processes of metabolites within organisms are complex and highly variable, posing significant challenges to their analysis. In the biomedical field in particular, metabolite analysis plays an irreplaceable and important role in disease diagnosis, drug development, personalized medicine, disease mechanism research, and biomarker discovery, providing a scientific basis for precision medicine and new drug development.

[0003] In the field of drug metabolism research, understanding the Phase I and Phase II reactions of drugs is crucial for gaining a deeper understanding of their metabolic pathways, activity changes, metabolites, and safety assessments. Phase I and Phase II reactions are the two main phases of drug metabolism in the body. Phase I reactions are the initial stage of drug metabolism and typically involve oxidation, reduction, and hydrolysis. Through the action of enzyme systems (such as cytochrome P450 enzymes), drug molecules introduce or expose functional groups, making them more susceptible to modification or attachment of other molecules in subsequent metabolic reactions. Phase II reactions of drug metabolism include glucosidation, sulfation, and glycation. In Phase II reactions, metabolites produced by Phase I reactions may undergo covalent chemical reactions with small molecules in the body (such as glucose, sulfuric acid, and acetate). This improves the water solubility and excretion of drug metabolites, making them more easily eliminated by the excretory system. A deeper understanding of Phase I and Phase II reactions is crucial for the analysis of drug metabolites.

[0004] The identification and elucidation of metabolites is a complex and arduous task, involving the analysis of thousands of small molecules within a living organism and their interactions. Due to the diverse and structurally complex nature of metabolites, the elucidation process often requires the use of mass spectrometry-chromatography. Chromatography is used to separate metabolites from complex biological samples into distinct components to reduce sample complexity. High-resolution and high-sensitivity mass spectrometry is then used to investigate the fragmentation processes and to characterize and quantify metabolites. Metabolite elucidation requires mass spectrometry data processing, including noise background removal, deconvolution, peak extraction, peak alignment, and secondary extraction, combined with specialized software for visualization and analysis. Current metabolite identification relies on secondary databases of metabolite standards. However, due to the incompleteness of these databases and the complex diversity of metabolites, the identification of unknown metabolites is extremely difficult, necessitating the development of new algorithms and tools to assist in identification. Currently, some machine learning algorithms use existing standard structures and their secondary fragments as training sets to train models, thereby using compound structures to predict secondary fragments to mitigate database incompleteness. However, the accuracy and robustness of these models remain to be improved. However, it is unreliable to identify metabolites solely by secondary fragment similarity. Metabolite identification needs to consider multiple dimensions and information sources, such as parent ion peaks, retention times, and the relationship between reactants and products, to ensure accuracy and reliability.

[0005] In the fields of biochemistry and drug metabolism, the analysis and verification of metabolites face many difficulties. On the one hand, there are a large number of unknown metabolites and metabolic reactions. There may be unknown metabolic pathways in the organism, leading to the formation of unknown metabolites. These metabolic pathways may be caused by specific environmental conditions, changes in gene expression, or enzyme isoforms, making them difficult to predict and analyze. Metabolites can be generated through a variety of metabolic reactions, covering a wide range of chemical structures, which increases the difficulty of analysis and identification. On the other hand, they are limited by in vitro models and technologies. Although in vitro metabolic models can simulate some metabolic reactions, they often cannot cover all possible situations in the organism, resulting in some metabolites that cannot be accurately analyzed and predicted in in vitro models. Summary of the Invention

[0006] In order to address the deficiencies in the prior art, the purpose of the present invention is to provide a method, system and application for structural annotation of parent metabolites based on mass spectrometry; the present invention aims to address the limitations of traditional structural analysis methods of parent metabolites and provide an efficient and accurate analysis solution combined with mass spectrometry for identifying and inferring the chemical structure of complex metabolites.

[0007] The present invention uses a certain parent metabolite as a substrate, uses SyGMa software to calculate the possible theoretical metabolite structures, combines the mass spectrometry data of in vitro reactions, and establishes a strict multidimensional metabolite scoring system and rule-based filtering criteria for metabolite selection and annotation, reducing the difficulty of analyzing and identifying artificial metabolites.

[0008] The present invention proposes a mass spectrometry-based structural annotation method for parent metabolites. Starting from the structure of the parent metabolite, the SyGMa software is used to deduce the structure of its possible metabolites. Combined with mass spectrometry data under in vivo or in vitro reaction conditions, the metabolites are scored and ranked based on five dimensions: isotope distribution score, parent similarity score, fragment similarity score, retention time score, and cumulative occurrence probability score of metabolic reactions. Metabolite rule filtering is used to screen out more reasonable metabolite structures and metabolic reaction combinations with a high probability of occurrence.

[0009] like Figure 8 As shown, the mass spectrometry-based structural annotation method of the parent metabolite in the present invention includes the following steps:

[0010] Prerequisites: The mass spectrometry data of the parent metabolite in an in vivo or in vitro reaction system and a blank control have been obtained in advance, and professional mass spectrometry analysis software has been used to process the data to obtain the peak table, isotope annotation, and secondary fragmentation information of the metabolites.

[0011] Step 1: Using the structural SMILES (Simplified Molecular Input Line Entry System) string of the parent metabolite as input (SMILES is a string representation method used to represent chemical molecules, used to describe the structure and composition of molecules, usually composed of a series of characters and symbols to represent atoms, bonds and other molecular features), SyGMa software is used to deduce the topological structure of all metabolic reaction products, and finally output the metabolite structure and the type of reaction that occurred.

[0012] Step 2: Calculate the mass-to-charge ratios of all metabolites output in step 1 in different adduct ion forms, perform correlation matching with the pre-constructed metabolite peak table, and annotate the possible candidate products and their adduct ion forms in the metabolite peak table.

[0013] In step 2, the association matching refers to matching all theoretical adduct ion mass-to-charge ratios of metabolites with the measured mass-to-charge ratios in the metabolite peak table within an error of 5 ppm, and adding annotations of the matched metabolite-related information to the metabolite peaks in the metabolite peak table.

[0014] The process of mass-to-charge ratio calculation and peak table association matching annotation in step 2 is as follows Figure 1 shown.

[0015] Step 3: Use a multi-dimensional scoring module to score the candidate metabolites, including isotope distribution score, parent-metabolite similarity score, fragment similarity score, retention time score and metabolic reaction cumulative occurrence probability score.

[0016] 1) Isotope distribution scoring: The theoretical n-position isotope distribution is calculated based on the molecular formula of the candidate metabolite using the Python package molmass, and the measured m-position isotope peak intensity is obtained from the metabolite peak table; where n refers to the number of isotopes in a group (M, M+1, M+2, ...), and m refers to the number of isotope peaks actually detected. The isotope matching process is as follows: Figure 2 As shown. Using the normalized measured m-position isotope intensity E={e0,e1,…,e m} and theoretical n-position isotope intensity T={t0,t1,…,t n}Calculate the isotope distribution score. The isotope score is calculated as follows:

[0017]

[0018] Comparing the corresponding isotope score values, the closer the isotope score is to 1, the more consistent the metabolite molecular composition is with the natural distribution, which to a certain extent reflects the accuracy of metabolite characterization.

[0019] 2) Parent-metabolite similarity score: Calculate the similarity between the parent secondary spectrum and the candidate metabolite secondary spectrum. First, align the secondary spectrum peaks according to the preset mass-to-charge ratio error (the mass-to-charge ratio error is related to the accuracy of the mass spectrometer, and the error of high-resolution instruments is usually set to 5ppm). The two secondary spectra are represented by the signal intensity vector A = {a1,…,a n} and B={b1,…,b n}, align the misaligned peaks with 0 and calculate the entropy distance:

[0020] Parent-metabolite similarity score

[0021] The entropy E of the secondary spectrum S S The intensity of all ions in the secondary spectrum I i To calculate, E A represents the entropy of the parent secondary spectrum A, E B represents the entropy of the product secondary spectrum B, E AB It represents the entropy after mixing the two spectra A and B. The schematic diagram of the principle is as follows Figure 3As shown, in one embodiment, the mixing ratio is 1: 1. The corresponding parent-metabolite similarity scores are compared. The parent-metabolite similarity can be understood as calculating the distance between two high-dimensional vectors. The closer the distance (i.e., the closer the parent-metabolite similarity score is to 1), the higher the similarity.

[0022] 3) Fragment similarity score: MetFrag software is used to calculate the fragment similarity score between the theoretical fragments of the candidate metabolite itself and the fragments measured by mass spectrometry. The higher the score, the greater the similarity between the measured fragments and the theoretical fragments of the product, the higher the interpretability of the fragments, and the more reliable the annotated metabolite.

[0023] 4) Retention time score: Use the trained LightGBM model to predict the retention time of the parent or metabolite with the SMILES of the parent or metabolite as input, and use the error Err of the corrected product predicted retention time relative to the true retention time of the product during mass spectrometry detection as the retention time score. If the true retention time of the parent is known, parent , predicted retention time t parent , product predicted retention time t product and the product's actual retention time T product , then the predicted retention time of the product is corrected according to the time offset ratio of the parent, and the predicted retention time of the product after correction is t' product The calculation formula is:

[0024]

[0025] Calculate the error of the corrected product predicted retention time relative to the product's true retention time. The retention time score is calculated as follows:

[0026] Retention time scoring or The closer the retention time score is to 0, the smaller the error between the predicted retention time and the actual retention time, and the more reliable the annotated metabolite.

[0027] In step 4) of step 3, the self-built retention time prediction model has a working principle diagram as shown below: Figure 4As shown in the figure. The retention time prediction model uses a gradient boosting-based LightGBM model. The structures and retention times of 28,745 standard compound samples are used as a dataset. The Java library CDK (Chemistry Development Kit) is used to convert the compound molecular structures into numerical molecular descriptors for model training. The hyperparameter search space is set, with the learning rate, number of trees, maximum tree depth, number of leaves, and minimum number of leaf nodes ranging from 0.01 to 0.2, 50 to 500, 5 to 50, 31 to 255, and 1 to 25, respectively. Random search and grid search combined with five-fold cross-validation were used to optimize the model's hyperparameters, ultimately achieving an accuracy of 91.61% within a retention time error of one minute. The hyperparameters are a learning rate of 0.1, a number of trees of 245, a maximum tree depth of 25, a number of leaves of 40, and a minimum number of leaf nodes of 18. This model can predict the retention time of molecules using the compound's molecular structure SMILES as input.

[0028] 5) Cumulative probability score of metabolic reactions: For a one-step reaction, the cumulative probability score of metabolic reactions is the probability of the corresponding metabolic reaction occurring. For a two-step reaction, if the probability of metabolic reaction A occurring is P(A) and the probability of metabolic reaction B occurring is P(B), then the probability of reactions A and B occurring simultaneously is:

[0029] P(A∩B)=P(A)·P(B)

[0030] The probability of reactions A and B occurring simultaneously is the cumulative probability score (P(A∩B)). For multi-step reactions, the cumulative probability score is the product of the probabilities of each reaction. A higher cumulative probability score indicates a higher probability of the reaction occurring and a more reliable annotated metabolite.

[0031] Step 4: Combine the scores of the five dimensions in step 3 and filter the metabolite list according to the set filtering rules. The filtering rules include the five metabolite screening criteria summarized during the manual analysis process:

[0032] 1) Filter out metabolites that only have M+K or M+NH4 adduct ions;

[0033] 2) Exclude metabolites that have M-2H adduct ions but do not have two or more acidic groups (including sulfonic acid, carboxyl, and glucuronic acid) in their structure;

[0034] 3) Exclude the peaks of metabolites that are cleaved within the source, that is, for candidate products whose reaction types are sulfation, glucuronidation, and glutathione conjugation, retain the metabolites with the greatest reaction depth;

[0035] 4) Exclude metabolites with a difference between the measured and predicted retention times greater than 60 seconds;

[0036] 5) For metabolites of the same reaction type and with a retention time error of 0.1 minutes, the five scores in step 3 are combined and sorted according to the isotope distribution score, fragment similarity score, retention time score, parent-metabolite similarity score, and cumulative probability of occurrence of metabolic reactions. First, sort in descending order according to the isotope distribution score; for the same isotope distribution score value, sort in descending order according to the fragment similarity score; for the same fragment similarity score value, sort in ascending order according to the absolute value of the retention time score; for the same retention time score value, sort in descending order according to the parent-metabolite similarity score; for the same parent-metabolite similarity score value, sort in descending order according to the cumulative probability of occurrence of metabolic reactions. Finally, the top three candidate metabolites are retained.

[0037] Step 5: Metabolite Peak Table Structural Annotation and Information Export: Export the metabolite annotation and structural information from the peak table to a file. The annotation information includes reaction type, adduct ion form, multidimensional score, and structure (including molecular formula and SMILES representation). The structural information is in SDF format, including 2D MOL information of the molecular structure.

[0038] The present invention also provides a structure annotation system for implementing the above-mentioned product structure annotation method, wherein the structure annotation system includes a metabolite structure derivation and peak table association annotation module, a metabolite multi-dimensional scoring module, and a metabolite rule filtering and information export module. Figure 9 As shown;

[0039] The metabolite structure deduction and peak table association annotation module may further include a product structure deduction module and a peak table association annotation module;

[0040] The product structure deduction module is used to input the parent metabolite and output all theoretical metabolites of the parent and the reaction types that occur;

[0041] The peak table association annotation module is used to associate and match theoretical metabolite information with the metabolite peak table and annotate the adduct ion form;

[0042] The metabolite multidimensional scoring module may further include an isotope distribution scoring module, a parent-metabolite similarity scoring module, a fragment similarity scoring module, a retention time scoring module, and a metabolic reaction cumulative occurrence probability scoring module;

[0043] The isotope distribution scoring module is used to calculate the isotope distribution score of the metabolite and evaluate the degree of conformity of the elemental composition of the metabolite molecule with the natural distribution;

[0044] The parent-metabolite similarity scoring module is used to calculate the secondary spectrum similarity between the candidate metabolite and the parent;

[0045] The fragment similarity scoring module is used to calculate the matching similarity score between the measured secondary fragments of the metabolites and the theoretical secondary fragments;

[0046] The retention time scoring module is used to predict the retention time of metabolites and parent compounds based on the self-developed machine learning model LightGBM model, and calculate the retention time score based on the measured product and parent retention times;

[0047] The metabolic reaction cumulative occurrence probability scoring module is used to calculate the probability score of the cumulative occurrence of one or more metabolic reactions;

[0048] The metabolite rule filtering and information export module may further include a metabolite rule filtering module and a metabolite structure annotation and information export module;

[0049] The metabolite rule filtering module filters unreasonable metabolites based on the five metabolite filtering rules summarized in the manual analysis process and combines multi-dimensional scoring;

[0050] The metabolite structure annotation and information export module is used to output information such as the structure, reaction type, adduct ion form, multidimensional score and SMILES of the metabolite.

[0051] The present invention also discloses the application of the above-mentioned annotation method or annotation system in metabolite analysis.

[0052] The beneficial effects of the present invention include: the present invention has independently developed a machine learning model for retention time prediction, and uses SMILES to predict the retention time of metabolites. The multidimensional scoring strategy for metabolite identification comprehensively considers multiple key dimensions, including isotope distribution, parent-metabolite similarity, fragment similarity, retention time score, and cumulative occurrence probability score of metabolic reactions. Combined with multiple product filtering rules summarized in manual analysis, metabolites are automatically filtered, which reduces the difficulty of artificial metabolite structure analysis and identification, and outputs reliable metabolite structure, reaction type and multidimensional scoring information, wherein the reaction type helps to understand the reaction process of the parent in the organism. In short, the present invention provides an efficient and accurate metabolite structure annotation method, which can provide a reliable auxiliary tool in complex metabolite analysis tasks, and is expected to promote research and application in the fields of biomedicine and drug metabolism. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 Schematic diagram showing the principle of associating metabolites with peak tables.

[0055] Figure 2 Schematic diagram showing the principle of matching theoretical and measured isotope distributions.

[0056] Figure 3 Schematic diagram showing the calculation principle of maternal similarity.

[0057] Figure 4 Schematic diagram showing the principle of the metabolite retention time prediction model.

[0058] Figure 5 Schematic diagram showing the molecular structure of an exemplary parent compound in a specific embodiment.

[0059] Figure 6 Schematic diagram showing the secondary mass spectrometric comparison between metabolite A and its parent.

[0060] Figure 7 Molecular structure diagram showing the analysis results of some metabolites.

[0061] Figure 8 A schematic diagram showing the process of the mass spectrometry-based structural annotation method of parent metabolites of the present invention.

[0062] Figure 9 A schematic diagram showing the implementation of the above-mentioned mass spectrometry-based parent metabolite structure annotation system of the present invention. DETAILED DESCRIPTION

[0063] The present invention is further described in detail with reference to the following specific examples and accompanying drawings. The processes, conditions, experimental methods, etc. for implementing the present invention, except for those specifically mentioned below, are common knowledge and common common sense in the art and are not particularly limited by the present invention.

[0064] The present invention proposes a method for annotating the structure of parent metabolites based on mass spectrometry. The method comprises the following steps: inputting a structural SMILES string of the parent metabolite, outputting all possible metabolite structures and the types of reactions that occur therein; calculating the mass-to-charge ratios of all output metabolites in different adduct ion forms, correlating and matching them with a pre-constructed metabolite peak table, and annotating them to obtain candidate metabolites and their adduct ion forms; performing multi-dimensional scoring on the obtained candidate metabolites; combining the scores of the multiple dimensions, filtering the metabolite list according to set filtering rules; and outputting the metabolite annotation information and structural information in the peak table to a file, wherein the annotation information includes reaction type, adduct ion form, multi-dimensional score, and structure. The present invention also proposes an annotation system and application for implementing the above annotation method.

[0065] Implementation Case 1

[0066] (1) Taking the metabolite analysis of the parent Polyporusterone A as an example, the SMILES string of the parent Polyporusterone A is CC(C)[C@@H](C)C[C@@H](O)[C@](C)(O)[C@H]1CC[C@@]2(O)C3=CC(=O)[C@@H]4C[C@@H](O)[C@@H](O)C[C@]4(C)[C@H]3CC[C@]12C, and its molecular structure is as follows Figure 5 Using SyGMa software, the maximum reaction depth was set to 3, the maximum and minimum number of phase I reactions were set to 2 and 0, respectively, and a total of 3940 theoretical metabolites were derived from the parent polyporusterone A (some examples of the results are shown in Table 1).

[0067] Table 1 Some examples of theoretical metabolites and their reaction types

[0068]

[0069]

[0070] (2) Calculate the mass-to-charge ratios (m / z) of the adduct ions of all metabolites in step (1) based on the different adduct ion forms (Table 2). The calculation formula is:

[0071]

[0072] Where mass is the exact molecular weight of the molecule, nmol is the number of moles of M in the adduct ion, charge is the charge number, ΔM is the changed molecular weight, and M in the adduct ion is the metabolite. Figure 1As shown, an error range of 5 ppm is set to match the m / z in the metabolite peak table. Metabolite A is the hydroxylation product of the parent metabolite (reaction type is aliphatic hydroxylation (C of a methyl group connected to a tertiary carbon)), and is annotated to the [M+H]+ and [M+H-H2O]+ adduct ion peaks of metabolite A. Similarly, candidate structures and adduct ion annotations of all theoretical metabolites are added to the peak table.

[0073] Table 2 Examples of some adduct ions

[0074] Additive form Charge number moles Molecular weight model [M+Na]+ 1 1 22.989218 just [M+K]+ 1 1 38.963158 just [M+H]+ 1 1 1.007276 just [M+H-H2O]+ 1 1 -17.003838 just [M+NH4]+ 1 1 18.033823 just [MH]- 1 1 -1.007176 burden [M-H2O-H]- 1 1 -19.01829 burden [M+Cl]- 1 1 34.969402 burden [M+FA-H]- 1 1 44.998201 burden

[0075] (3) Multi-dimensional scoring is performed on the candidate metabolites in the metabolite peak table. First, the isotope distribution is scored. The theoretical isotope distribution is calculated based on the molecular formula. The experimental isotope distribution is calculated using the metabolite intensity in the same group of isotopes in the measured metabolite peak table. The score is calculated after normalization. Figure 2 As shown, taking the calculation of metabolite A as an example, the theoretical 16-position isotope distribution of A was calculated according to the molecular formula using the Python package molmass. The normalized isotope distribution ratio (isotope intensity) is T = {0.723586, 0.224889, 0.044170, 0.006491, 0.000776, 7.88e-05, 6.98e-06, 5. 48e-07, 3.86e-08, 2.46e-09, 1.43e-10, 7.63e-12, 3.72e-13, 1.64e-14, 2.55e-16, 1.22e-18}, the measured isotope intensity of position 2 is E = {15655291.2, 2685973.55}, after normalization it is {0.85355, 0.14645}. The score calculation formula is as follows:

[0076]

[0077] In this case, n = 16 is greater than m = 2, and the isotope distribution score calculated according to the formula is 0.87. Isotope distribution scores are calculated in the same way for other metabolites. If the measured isotope peak for a metabolite is not found in the peak table, the isotope distribution score is not provided.

[0078] (4) Figure 3 As shown, the parent secondary spectrum and product secondary spectrum are represented by their intensity vectors, and the secondary fragment peaks are aligned with a 10ppm error. The misaligned peaks are aligned with 0. First, according to formula E S =∑ i I i ln(I i ) calculates the entropy of the secondary spectrum, and then calculates the parent similarity score. Taking metabolite A as an example, the secondary mass spectrometry data of the product and the parent and their comparison are as follows Figure 6 As shown, the parent secondary mass spectrum entropy E is calculated based on the intensity data of the fragments in the figure. A =3.9078, product secondary mass spectrum entropy E B =2.9158, mass spectrum entropy of the parent and product secondary mixed spectrum E AB =3.9120, according to the formula The parent similarity of metabolite A was calculated to be 0.28. The parent similarity scores of other metabolites were calculated in the same way. If the secondary spectrum of a metabolite was not detected in the peak table, no parent similarity score was provided.

[0079] (5) The fragment similarity score was calculated using the MetFrag software, with the secondary fragments of metabolite A as input, and the calculated similarity score was 662.47. If the secondary spectrum of a metabolite was not detected in the peak table, no fragment similarity score was provided.

[0080] (6) Figure 4 As shown in the figure, the SMILES of the metabolites are used as the input of the retention time prediction model to predict the retention time of the metabolites. The predicted retention time of the metabolites is corrected proportionally according to the predicted retention time of the parent compound, and the relative retention time deviation is calculated. The SMILES of metabolite A is CC(C)[C@@H](CO)C[C@@H](O)[C@](C)(O)[C@H]1CC[C@@]2(O)C3=CC(=O)[C@@H]4C[C@@H](O)[C@@H](O)C[C@]4(C)[C@H]3CC[C@]12C. After inputting the model, the predicted retention time t product is 345.28, and the measured retention time of product A is known to be T product The parent's true retention time is 315.22. parent is 429.23, and the predicted retention time t parent is 489.58, according to the formula:

[0081]

[0082] Calculate the predicted retention time t' of the corrected product product is 302.72, according to the formula:

[0083] Retention time scoring or

[0084] Substituting this into the retention time score, Err, yields a value of 0.0412. This method is used to calculate the retention time scores for all metabolites in the peak table.

[0085] (7) The cumulative probability of the reaction was calculated using the probability of occurrence of the metabolic reaction type in SyGMa software. The reaction type of metabolite A is aliphatic hydroxylation (methyl C connected to tertiary carbon), which is a one-step reaction with a reaction probability of 0.03. When a multi-step reaction occurs, such as aliphatic hydroxylation (secondary carbon in a heterocycle) and dehydration (dehydroxylation side and sp2 hybridized carbon) to form metabolites, the cumulative probability score of the reaction is the product of the probability of each step of the reaction, which is 0.027*0.01397, equal to 0.000377. The larger the cumulative probability score of the reaction, the greater the probability of generating this metabolite.

[0086] (8) According to the five rules in the rule filtering module, the metabolites annotated in the peak table are filtered to remove unreasonable metabolites. The rules in the rule filtering module have been written into a script to automatically filter the peak table. The final metabolite peak table formed after filtering has a total of 27 metabolites of the parent compound. Some example results are shown in Table 3.

[0087] (9) Export the metabolite annotation information and structural information in the metabolite peak table to a file. The reaction types and scoring annotations of some metabolites are shown in Tables 3 and 4. The structural diagrams of some resolved metabolites are shown in Figure 7 shown.

[0088] Table 3 Analysis results of some metabolites Reaction type table

[0089]

[0090]

[0091] Table 4 Score table of some metabolite analysis results

[0092] serial number adduct ions Isotope distribution score Maternal similarity score Fragment similarity score Retention time scoring Possibility of cumulative reaction 1 [M+H-H2O]+,[M+H]+ 0.87 0.2832 662.4685 0.0413 0.03 2 [M+H-H2O]+,[M+H]+ 0.87 0.2832 662.4685 0.0457 0.097 3 [M+H-H2O]+,[M+H]+ 0.87 0.2832 589.5189 -0.0442 0.027 4 [M+H]+ 0.956 0.3933 635.3852 -0.0107 0.013 5 [M+H]+ 0.956 0.3933 623.8986 -0.0258 0.061 6 [M+H]+ 0.956 0.3933 570.2594 0.0316 0.027 7 [M+H-H2O]+ 0.9119 0.0488 321.9151 0.0762 0.027 8 [M+H-H2O]+ 0.9119 0.0488 310.1566 0.0601 0.013 9 [M+FA-H]- 0.9364 0 41.7146 -0.064 0.0004 10 [M+H]+ 0.9806 0.0488 498.1918 -0.0223 0.0004 11 [M+H-H2O]+ 0.8493 0.1935 264.0989 -0.0886 0.001 12 [M+H]+ 0.9145 0.1214 250.3037 -0.1316 0.0001 13 [M+Na]+,[M+H-H2O]+ 0.8907 0.1474 268.1923 -0.0113 0.101 14 [M+Na]+,[M+H-H2O]+ 0.8907 0.1474 268.1923 -0.0432 0.101 15 [M+Na]+,[M+H-H2O]+ 0.8907 0.1474 231.6307 -0.0353 0.101

[0093] The protection content of the present invention is not limited to the above embodiments. Without departing from the spirit and scope of the present invention, changes and advantages that can be thought of by those skilled in the art are included in the present invention and are protected by the appended claims.

Claims

1. A method for structural annotation of parent metabolites based on mass spectrometry, characterized in that: The marking method comprises the following steps: Step 1: Input the structural SMILES string of the parent metabolite and output the metabolite structure and the type of reaction that occurred; Step 2: Calculate the mass-to-charge ratios of all metabolites output in step 1 in different adduct ion forms, perform correlation matching and annotation with the pre-constructed metabolite peak table, and obtain candidate metabolites and the adduct ion forms of the candidate metabolites; Step 3: Perform multi-dimensional scoring on the candidate metabolites obtained in step 2; Step 4: Combine the scores of multiple dimensions in step 3 and filter the metabolite list according to the set filtering rules; Step 5: Output the metabolite annotation information and structure information in the peak table to a file, wherein the annotation information includes reaction type, adduct ion form, multidimensional score and structure.

2. The marking method according to claim 1, characterized in that: In step 1, the topological structure of the metabolic reaction product is derived using SyGMa software; and / or, In step 2, all theoretical adduct ion mass-to-charge ratios of metabolites are matched with the measured mass-to-charge ratios in the metabolite peak table within an error of 5 ppm, and the metabolite peaks in the metabolite peak table are annotated with the matched metabolite information; and / or, In step three, the multidimensional scoring refers to the isotope distribution score, parent-metabolite similarity score, fragment similarity score, retention time score and metabolic reaction cumulative occurrence probability score of the candidate metabolites.

3. The marking method according to claim 2, characterized in that: The isotope distribution scoring process is as follows: the theoretical n-position isotope distribution is calculated according to the molecular formula of the candidate metabolite, and the measured m-position isotope peak intensity is obtained from the metabolite peak table; wherein n refers to the number of a group of isotopes, and m refers to the number of isotope peaks actually detected; the normalized measured m-position isotope intensity E = {e0, e1, ..., e m } and the theoretical n-position isotope intensity T = {t0, t1, …, t n }Calculate the isotope distribution score; the isotope score is calculated as follows:

4. The marking method according to claim 2, characterized in that: The process of parent-metabolite similarity scoring is as follows: Calculate the similarity between the parent secondary spectrum and the candidate metabolite secondary spectrum: align the peaks of the parent secondary spectrum and the candidate metabolite secondary spectrum according to the preset mass-to-charge ratio error, and represent the two secondary spectra with signal intensity vectors A = {a1,…,a n } and B={b1,…,b n }, align the misaligned peaks with 0 and calculate the entropy distance: The entropy E of the secondary spectrum S S The intensity of all ions in the secondary spectrum I i To calculate, E A represents the entropy of the parent secondary spectrum A, E B represents the entropy of the product secondary spectrum B, E AB It represents the entropy of the mixture of two spectra A and B.

5. The marking method according to claim 2, characterized in that: The process of the fragment similarity score is as follows: MetFrag software is used to calculate the fragment similarity scores of the theoretical fragments of the candidate metabolite itself and the fragments measured by mass spectrometry.

6. The marking method according to claim 2, characterized in that: The retention time scoring process is as follows: using the pre-trained retention time regression model LightGBM, taking the SMILES of the parent or metabolite as input to predict the retention time of the parent or metabolite, and using the error Err of the corrected product predicted retention time relative to the true retention time as the retention time score; The predicted retention time of the product is corrected according to the time shift ratio of the parent, and the corrected predicted retention time of the product is shown in the following formula: Among them, T parent is the true retention time of the parent, t parent is the predicted retention time, t product Predict retention time, T for the product product is the actual retention time of the product; Calculate the error of the corrected product predicted retention time relative to the product's true retention time. The retention time score is calculated as: The learning rate of the retention time regression model is 0.1, the number of trees is 245, the maximum tree depth is 25, the number of leaves is 40, and the minimum number of leaf nodes is 18.

7. The marking method according to claim 2, characterized in that: The cumulative occurrence probability scores of the metabolic reactions include the following: For a one-step reaction, the cumulative probability score of the metabolic reaction is the corresponding probability of the metabolic reaction; for a multi-step reaction, the cumulative probability score of the metabolic reaction is the product of the probabilities of each reaction.

8. The marking method according to claim 1, characterized in that: In step 4, the filtering rules include: 1) Filter out metabolites that have only M+K or M+NH4 adduct ions; 2) Excluding metabolites that have an M-2H adduct ion form but do not have two or more acidic groups in their structure; the acidic groups include sulfonic acid, carboxyl and glucuronic acid; 3) Exclude the metabolite peaks of the source cleavage, that is, for the candidate products whose reaction types are sulfation, glucuronidation, and glutathione conjugation, retain the metabolites with the greatest reaction depth; 4) Exclude metabolites with a difference between the measured retention time and the predicted retention time greater than 60 seconds; 5) For metabolites of the same reaction type and with a retention time error of 0.1 min, combined with the multi-dimensional scoring in step 3, they are ranked in descending order according to the isotope distribution score, fragment similarity score, retention time score, parent-metabolite similarity score, and cumulative occurrence probability score of the metabolic reaction, and the top three candidate metabolites are retained.

9. A structure annotation system for implementing the product structure annotation method according to any one of claims 1 to 8, characterized in that: The structure annotation system includes a metabolite structure derivation and peak table association annotation module, a metabolite multi-dimensional scoring module, and a metabolite rule filtering and information export module; The metabolite structure derivation and peak table association annotation module further includes a product structure derivation module and a peak table association annotation module; The product structure derivation module is used to input the parent metabolite and output all theoretical metabolites of the parent and the types of reactions that occur; The peak table association annotation module is used to associate and match theoretical metabolite information with the metabolite peak table and annotate the adduct ion form; The metabolite multidimensional scoring module further includes an isotope distribution scoring module, a parent-metabolite similarity scoring module, a fragment similarity scoring module, a retention time scoring module, and a metabolic reaction cumulative occurrence probability scoring module; The isotope distribution scoring module is used to calculate the isotope distribution score of the metabolites and evaluate the degree of conformity between the elemental composition of the metabolite molecules and the natural distribution; The parent-metabolite similarity scoring module is used to calculate the secondary spectra similarity between the candidate metabolites and the parent; The fragment similarity scoring module is used to calculate the matching similarity scores between the measured secondary fragments of the metabolites and the theoretical secondary fragments; The retention time scoring module is used to predict the retention time of metabolites and the parent, and calculate the retention time score in combination with the measured product and parent retention times; The metabolic reaction cumulative occurrence probability scoring module is used to calculate the probability score of the cumulative occurrence of one or more metabolic reactions; The metabolite rule filtering and information export module further includes a metabolite rule filtering module and a metabolite structure annotation and information export module; The metabolite rule filtering module filters unreasonable metabolites based on metabolite filtering rules and multi-dimensional scores; The metabolite structure annotation and information export module is used to output the structure, reaction type, adduct ion form, multidimensional score and SMILES information of the metabolites.

10. Use of the metabolite structure annotation method according to any one of claims 1 to 8, or the structure annotation system according to claim 9 in metabolite analysis.

Images