A method for processing metabolomics data based on a graph theory strategy

Abstract

The application provides a metabolomics data processing method based on a graph theory strategy, and belongs to the technical field of bioinformatics and analytical chemistry. First, metabolomics original peak data of multiple biological samples is acquired; quality cleaning is performed on the original peak data to obtain a peak set after cleaning, each peak being composed of mass-to-charge ratio, retention time and peak intensity to form multi-dimensional characteristic data; a peak matching graph is constructed based on the multi-dimensional characteristic data, graph nodes representing peaks, and graph edge weights reflecting the comprehensive similarity between peaks of any two different samples, the similarity being obtained by weighted fusion of normalized similarities of mass-to-charge ratio, retention time and peak intensity; based on the peak matching graph, an effective matching group is generated by depth-first search under the conditions of meeting basic similarity and chromatography-mass spectrometry continuity and high instrument resolution capability advanced constraints. The application significantly improves the accuracy and robustness of cross-sample metabolite peak matching, and effectively supports large-scale, multi-batch complex metabolomics data analysis.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of bioinformatics and analytical chemistry, and relates to a metabolomics data processing method based on graph theory strategies. Background Technology

[0002] Metabolomics utilizes high-throughput mass spectrometry to systematically detect small-molecule metabolites in organisms, playing a crucial role in disease detection and drug response assessment. Peak matching, a core data processing step, aims to accurately correlate chromatographic-mass spectrometric peaks originating from the same metabolite in different samples, a prerequisite for cross-sample metabolite quantification and comparative analysis. Currently, mainstream metabolomics analysis tools such as XCMS widely employ peak matching strategies based on window alignment or dynamic time warping. However, these methods face multiple technical bottlenecks when processing complex, large-scale metabolomics data:

[0003] (1) Insufficient utilization of multidimensional features. Existing methods usually process features such as mass-to-charge ratio, retention time and peak intensity independently, or only use simple tolerance windows for matching. They fail to effectively integrate the intrinsic relationship between mass-to-charge ratio, retention time and peak intensity, which leads to mismatch in areas with dense peaks.

[0004] (2) Retention time drift robustness is poor. Although local or global alignment algorithms are introduced, the global alignment assumption fails when the instrument is aged, the performance of the chromatographic column fluctuates or the batch effect is significant. Local alignment is also difficult to handle nonlinear drift, resulting in a large number of missed matches or false positive matches.

[0005] (3) Lack of instrument resolvability constraints. Existing matching strategies do not take into account the actual resolvability of the mass spectrometer, which may mistakenly regard adjacent peaks that the instrument cannot distinguish as the same metabolite, or conversely, misjudge resolvable peaks as unmatchable.

[0006] (4) Ignoring the influence of experimental metadata, the consistency of peaks can be significantly affected by different experimental batches, collection temperatures and other metadata. However, the current method does not incorporate such prior knowledge into the matching decision process, which reduces the matching reliability of cross-batch data.

[0007] (5) It is difficult to balance computational efficiency and accuracy. As the scale of metabolomics research continues to expand, the traditional XCMS peak matching method has high computational complexity and low operating efficiency when processing large-scale data. If the efficiency is improved by simplifying the matching strategy or relaxing the constraints, it often leads to a decrease in matching accuracy, making it difficult to meet the analysis requirements of matching accuracy and high computational efficiency.

[0008] Therefore, there is an urgent need for a novel peak matching method that can deeply integrate multidimensional features, combine instrument physical constraints, dynamically integrate experimental meta-information, and balance matching accuracy and computational efficiency, so as to improve the accuracy and efficiency of metabolomics data analysis. Summary of the Invention

[0009] This application provides a graph theory-based metabolomics data processing method, which aims to solve the technical problems in the prior art, such as low peak matching accuracy, insufficient consideration of retention time drift and instrument resolution constraints, insufficient utilization of multidimensional features, poor consistency of cross-batch data, and difficulty in balancing efficiency and matching accuracy in large-scale data processing.

[0010] The first aspect disclosed in this application provides a graph-based strategy for metabolomics data processing, including:

[0011] Obtain raw metabolomics peak data from multiple biological samples. The raw peak data includes the mass-to-charge ratio and retention time of each peak, as well as the peak intensity, peak area, and peak width in each sample.

[0012] The original peak data is cleaned to obtain a cleaned peak set. Each peak in the peak set consists of multidimensional feature data composed of its mass-to-charge ratio, retention time and peak intensity.

[0013] A peak matching graph is constructed based on the multidimensional feature data, where graph nodes represent each peak and are associated with its multidimensional feature data, and graph edge weights reflect the comprehensive similarity between peaks of any two different samples. The comprehensive similarity is obtained by normalizing the similarity based on mass-to-charge ratio, retention time, and peak intensity and then weighting and fusing them.

[0014] Based on the peak matching diagram, the matching relationship between peaks representing the same metabolite in different samples is determined, resulting in multiple effective matching groups. Each effective matching group contains a set of peaks from different samples that are associated with the same metabolite. The matching relationship satisfies basic similarity constraints as well as advanced matching constraints based on chromatographic-mass spectrometry continuity and instrument resolution.

[0015] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0016] (1) By constructing a comprehensive similarity based on multidimensional features of mass-to-charge ratio, retention time and peak intensity, and using this to generate the edge weights of the peak matching graph, the intrinsic relationship between mass-to-charge ratio, retention time and peak intensity is effectively integrated, which significantly improves the accuracy of matching.

[0017] (2) The introduction of advanced matching constraints, including retention time continuity, mass-to-charge ratio change rate continuity and mass spectrometer resolvability constraints, effectively suppresses mismatch and omission caused by instrument drift, peak density or resolution limitation, and enhances the robustness of the present invention under complex data.

[0018] (3) The graph edge weights are dynamically adjusted based on information such as experimental batch consistency and sample collection temperature, which improves the matching consistency and reproducibility of cross-batch metabolomics data;

[0019] (4) Greedy filtering is used to improve efficiency when processing large-scale data, and the Hungarian algorithm is used to meet the research needs of both computational efficiency and matching accuracy when analyzing rare metabolites and other high-precision requirements.

[0020] In summary, this application can effectively overcome several limitations of the prior art while ensuring high matching accuracy, providing a highly reliable and adaptable peak matching solution for metabolomics data analysis. Attached Figure Description

[0021] Figure 1 A schematic flowchart of a graph theory-based metabolomics data processing method provided in this application embodiment;

[0022] Figure 2 Peak matching diagrams provided for embodiments of this application;

[0023] Figure 3 The image shows the metabolite matching results after processing with the traditional XCMS peak matching method.

[0024] Figure 4 The image shows the metabolite matching results after metabolomics data processing based on graph theory strategies, as provided in the embodiments of this application. Detailed Implementation

[0025] This application provides a graph theory-based method for metabolomics data processing. The technical solutions described below will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the accompanying drawings, not all of them.

[0026] like Figure 1 As shown, this application provides a graph theory-based method for metabolomics data processing, which specifically includes the following steps:

[0027] S100: Obtain raw metabolomics peak data of multiple biological samples, the raw peak data including the mass-to-charge ratio and retention time of each peak, as well as the peak intensity, peak area and peak width of each peak in each sample.

[0028] Furthermore, before obtaining raw metabolomics peak data from multiple biological samples, the following steps are taken:

[0029] Obtain raw chromatographic-mass spectrometry data from multiple biological samples;

[0030] The original chromatographic-mass spectrometry data are preprocessed, including noise-reduced chromatograms and mass spectra, to obtain noise-reduced chromatographic-mass spectra data.

[0031] Based on the noise-reduced chromatographic-mass spectrometry data, peak detection and feature extraction are performed to generate an initial peak data table. The initial peak data table includes the mass-to-charge ratio, retention time, peak intensity, peak area, and peak width of each peak in each sample, forming raw metabolomics peak data for multiple biological samples.

[0032] Specifically, this embodiment involves acquiring and preliminarily processing raw chromatographic-mass spectrometry (GC-MS) data from multiple biological samples, aiming to lay a solid foundation for subsequent detailed analysis. This process mainly relies on XCMS, a mainstream metabolomics data analysis tool, to complete data extraction and preliminary purification. The specific steps are as follows: Raw data acquisition: A series of biological samples are detected using high-resolution GC-MS techniques (e.g., LC-MS, GC-MS, etc.) to obtain raw GCMS data for each sample. These data contain important information about the compounds within the sample, including but not limited to mass-to-charge ratio (m / z), retention time (RT), peak intensity, and peak area. Data extraction using XCMS software: After acquiring the raw data, XCMS is used to accurately extract this data. XCMS can effectively identify and extract key information from chromatograms and mass spectra, such as mass-to-charge ratio, retention time, and detailed peak information, including peak intensity, peak area, and peak width. Data preprocessing, performed using XCMS, primarily includes noise reduction. Techniques employed in this stage include: sliding window filtering: based on the characteristics of metabolomics data, the sliding window size is set to 5 to 10 data points, and parameter settings are optimized experimentally to effectively remove high-frequency noise from the data; or wavelet transform: a suitable wavelet basis function (such as the Daubechies wavelet) is selected for the metabolomics data, and the data is decomposed and reconstructed at different scales, effectively reducing noise while maintaining the integrity of important peak information. After these steps, high-quality denoised chromatograms and mass spectra are generated, preparing for further peak detection and feature extraction. After data denoising, XCMS is used to perform peak detection and feature extraction on the denoised chromatogram-mass spectrometry signals, generating a structured initial peak data table. This table uses each row to represent a detected metabolite feature (i.e., peak), and each column contains the general attributes of that feature and its quantitative information in various biological samples. General attributes refer to information describing the basic properties of each detected metabolite feature (i.e., peak). These attributes typically include, but are not limited to, the following: FeatureID: used to uniquely identify each detected peak; Mass-to-charge ratio: indicates the center position of the peak in the mass spectrometry dimension, which is one of the key parameters for identifying compounds; Retention time: refers to the time point at which the peak is eluted in the chromatographic column, which is one of the important parameters for identifying compounds; Peak width: reflects the broadening of the chromatographic peak, helping to judge the peak quality; Signal-to-noise ratio: used to measure the intensity of the signal relative to background noise, which helps to evaluate the peak quality. Quantitative information refers to the quantitative data of the peak in various biological samples.This type of information is mainly used to compare differences between different samples and may include the following: Peak intensity: representing the maximum or average signal intensity of the peak in a specific sample, usually used to reflect the relative abundance of the compound; Peak area: representing the total area under the peak, as another method to measure the abundance of the compound, generally considered more stable than peak intensity. For each sample, there will be one or more columns corresponding to the above quantitative indicators, thus forming an M×N matrix, where M is the number of peaks and N is the number of samples plus the number of characteristic attributes. In addition to the raw chromatographic-mass spectrometry signal, this embodiment also synchronously acquires experimental metadata related to each biological sample, including: Experimental batch identifier: used to distinguish sample groups prepared at different times or under different conditions, usually determined by a laboratory information management system (LIMS) or sample numbering rules; Ambient temperature at sample collection: recorded in the experimental log or electronic experimental record book (ELN), reflecting the consistency of sample processing conditions; Mass spectrometer resolution: as a core parameter of the instrument acquisition method, it is embedded in the metadata of the raw mass spectrometry data file (such as .mzML, .raw) and can be automatically read during data import.

[0033] The aforementioned metadata is used to dynamically adjust similarity weights or impose advanced constraints during peak matching. The acquisition method is a standard practice in the field of metabolomics and can be achieved without additional creative effort.

[0034] S200: Perform quality cleaning on the original peak data to obtain a cleaned peak set. Each peak in the peak set consists of multidimensional feature data composed of its mass-to-charge ratio, retention time, and peak intensity.

[0035] Furthermore, the original peak data is cleaned to obtain a cleaned peak set, including:

[0036] The quality cleaning includes removing abnormal peaks, which is achieved based on at least one characteristic parameter of retention time, peak area, and peak width, and is accomplished in the following manner:

[0037] Set the allowable range for each characteristic parameter, and take the peaks that exceed the allowable range as candidate abnormal peaks. Further confirm whether they are abnormal peaks through box plot analysis. If they are, remove them to obtain the cleaned peak set.

[0038] Specifically, after obtaining raw metabolomics peak data from multiple biological samples, this embodiment performs quality cleaning on the raw peak data to remove low-quality or unreliable peak signals. The specific steps of quality cleaning are as follows: First, for the four key characteristic parameters—peak area, peak height, retention time, and peak width—the 95% confidence intervals in all samples are calculated. These confidence intervals can be determined by assuming the data follows a normal distribution and using the mean ± 1.96 times the standard deviation; or by using a non-parametric method (such as the percentile method) to directly take the 2.5th percentile to the 97.5th percentile as the interval boundary. Peaks whose characteristic parameters exceed their corresponding 95% confidence intervals are marked as candidate anomalous peaks. Subsequently, to further verify their anomalousness, box plot analysis is performed on the candidate anomalous peaks for secondary verification. Specifically, each characteristic parameter is processed independently. For each characteristic parameter—peak area, peak height, retention time, peak width, etc.—the numerical sequence of that parameter is extracted from the entire peak set before cleaning, forming a one-dimensional dataset D = {x1, x2, ..., x...}. n} where n is the total number of peaks. Calculate the first quartile (Q1) and the third quartile (Q3), where Q1 is the value located at the 25th percentile of the sorted data (i.e., 25% of the data are ≤Q1); and Q3 is the value located at the 75th percentile of the sorted data (i.e., 75% of the data are ≤Q3). Calculate the interquartile range (IQR). Define the outlier detection boundary, lower bound = The upper bound is Q3 + 1.5 × IQR. To identify outlier peaks, for any candidate outlier peak, if its value on at least one characteristic parameter is below the lower bound or above the upper bound, then the peak is considered a statistical outlier and is removed.

[0039] S300: Construct a peak matching graph based on the multidimensional feature data, where graph nodes represent each peak and are associated with its multidimensional feature data, and graph edge weights reflect the comprehensive similarity between peaks of any two different samples. The comprehensive similarity is obtained by normalizing the similarity based on mass-to-charge ratio, retention time, and peak intensity and then weighting and fusing them.

[0040] Furthermore, a peak matching graph is constructed, including:

[0041] Each peak in the peak set is used as a graph node, and each peak is associated with its corresponding multidimensional feature data, including mass-to-charge ratio, retention time and peak intensity.

[0042] The space of the mass-to-charge ratio is divided into multiple intervals of preset width, and only peak pairs falling into the same or adjacent intervals are retained as candidate correlation pairs.

[0043] The peak intensity is preprocessed by performing a logarithmic transformation to obtain the preprocessed peak intensity value. The logarithmic transformation formula is as follows:

[0044] ;

[0045] Where I' is the preprocessed peak intensity value, and I is the peak intensity in the multidimensional feature data;

[0046] Calculate the overall similarity of each peak pair in the candidate association pairs, where each peak pair comes from peaks in different samples, using the following formula:

[0047] (1) Calculate the mass-to-charge ratio similarity:

[0048] ;

[0049] in, and Let be the mass-to-charge ratios of any two peaks. and These are the maximum and minimum mass-to-charge ratios for all peaks in the entire set of peaks;

[0050] (2) Calculate retention time similarity:

[0051] ;

[0052] in, and , where represents the retention time of any two peaks. and These are the maximum and minimum values ​​of all retention times in the entire set of peaks, respectively;

[0053] (3) Calculate the intensity similarity:

[0054] ;

[0055] in, and These are the peak intensity values ​​after logarithmic transformation of any two peaks. and These are the maximum and minimum values ​​of the peak intensity after logarithmic transformation in the entire peak set, respectively;

[0056] The three similarities mentioned above are weighted and fused to obtain the comprehensive similarity score, as shown in the following formula:

[0057] Sim=w1·Sim(m / z)+w2·Sim(tR)+w3·Sim(I');

[0058] Where Sim represents the overall similarity, w1, w2, and w3 are preset weight coefficients, and w1+w2+w3=1, with the preferred values ​​being w1=0.4, w2=0.35, and w3=0.25;

[0059] Using the comprehensive similarity as the graph edge weights, an initial peak matching graph is constructed;

[0060] The initial peak matching graph is subjected to graph optimization processing, including: removing self-looping edges and sparsifying edges with a degree less than 3, to obtain the optimized peak matching graph.

[0061] Specifically, after cleaning the original peak data, this embodiment further constructs a peak matching graph for cross-sample peak matching based on the multidimensional feature data in the cleaned peak set. The specific construction process is as follows: The graph nodes are constructed and associated with features, with each peak in the cleaned peak set serving as a node in the graph. Each node is associated with its corresponding multidimensional feature data, including three core dimensions: mass-to-charge ratio (m / z), retention time (t), and other parameters. RThe mass-to-charge ratio (m / z) space is divided into multiple continuous intervals of preset width. Only cross-sample peak pairs with mass-to-charge ratios falling within the same or adjacent intervals are retained as candidate correlation pairs. This effectively excludes peak pairs with excessively large mass-to-charge ratio differences, avoids meaningless similarity calculations, and retains the possibility of reasonable proximity matching due to minor instrument drift. In this embodiment, the entire mass-to-charge ratio (m / z) space is divided into multiple continuous intervals of preset width. The preset width is mainly based on the mass resolution or minimum resolvable mass-to-charge ratio difference of the mass spectrometer used. For example, on high-resolution mass spectrometry platforms, such as Orbitrap or Q-TOF, the instrument can usually distinguish ions with mass-to-charge ratio differences in the range of 0.005–0.02 Da. Therefore, to ensure that the peaks of the same real metabolite caused by minor drift still fall within the same or adjacent intervals, and to avoid different metabolites being incorrectly classified into the same interval, the preset width is preferably set to 0.01–0.05 Da. For example, using a Q-TOF mass spectrometer with a nominal resolution of 30,000 m / z 556, the smallest resolvable mass-to-charge ratio difference is approximately 0.015 Da. In this case, the interval width can be set to 0.02 Da. This ensures that peaks originating from the same metabolite are likely to fall within the same or adjacent intervals, while metabolites with different chemical structures are effectively isolated, thus suppressing mismatch risk while maintaining recall. Logarithmic transformation preprocessing of peak intensities is crucial. Original peak intensities are typically highly skewed (a few strong peaks dominate, with a large number of weak peaks densely packed). Directly using linear intensity values ​​would lead to strong peaks dominating similarity calculations. Therefore, a logarithmic transformation preprocessing (log(I+1) form) is applied to the intensity values ​​of all peaks to compress the dynamic range, allowing strong and weak peaks to contribute more evenly in similarity calculations. Combining similarity calculation with edge weight assignment, the similarity of each candidate association pair is calculated from three dimensions: mass-to-charge ratio, retention time, and preprocessed peak intensity. The similarity of each dimension is calculated using a normalization method, mapping the similarity of each dimension to a unified [0, 1] interval. Then, the similarities of the three dimensions are weighted and fused according to preset weight coefficients to obtain a comprehensive similarity value. This value reflects the overall closeness of the two peaks in the multidimensional feature space. The peak matching graph is optimized by performing the following operations on the initial peak matching graph: removing self-looping edges, i.e., deleting edges connecting the same node; sparsifying low-connectivity edges, removing the edge for any node with a certain degree (i.e., the number of connecting edges), resulting in the optimized peak matching graph, as shown below. Figure 2 As shown, this sparsification strategy effectively eliminates isolated or weakly supported suspicious connections, reduces noise interference, and retains reliable matching relationships with sufficient neighborhood support, making the graph structure more focused on high-confidence associations. Finally, this comprehensive similarity is used as the weight of the graph edge between corresponding two nodes to construct the initial peak matching graph. The higher the edge weight, the more likely the two peaks originate from the same metabolite.

[0062] In a preferred embodiment, for a peak with a mass-to-charge ratio of 150.23, a retention time of 8.5 minutes, and a peak intensity of 5000, a node is created and these attributes are accurately recorded. When calculating similarity, normalized similarity is used for peak A (mass-to-charge ratio m / z1 = 150.23) and peak B (mass-to-charge ratio m / z2 = 150.30), as well as for retention time. To more comprehensively consider the similarity between peaks, a comprehensive similarity function Sim = w1·Sim(m / z) + w2·Sim(tR) + w3·Sim(I') is constructed. The weights are determined through multiple experiments, with mass-to-charge ratio weight w1 = 0.4, retention time weight w2 = 0.35, and peak intensity weight w3 = 0.25. The similarity results are mapped to edge weights; if the comprehensive similarity between peak A and peak B is 0.75, then the edge weight connecting them is set to 0.75. Meanwhile, considering data quality and reliability, if the coefficient of variation of the intensity of a peak in multiple repeated experiments is greater than 10%, the weight of the edges between that peak and other peaks is appropriately reduced. A peak matching graph is constructed based on nodes and edges. If the research objective is to analyze the relationships between metabolites, an undirected graph is chosen; if the focus is on the matching order of peaks, a directed graph is chosen. During the construction process, graph optimization algorithms are used to remove self-loop edges, and edges connecting fewer than 3 nodes are sparsified to improve graph quality and the efficiency of subsequent algorithms.

[0063] Furthermore, constructing the initial peak matching plot also includes:

[0064] Before constructing the initial peak matching map, potential isotope peak pairs are identified based on the cleaned peak set: if the mass-to-charge ratio difference between two peaks falls within the range of [0.99, 1] Da, the retention time difference is less than the preset tolerance, and the intensity ratio conforms to the distribution law of natural isotopes, then the two are marked as the parent peak and the isotope peak.

[0065] When constructing the initial peak matching diagram, an edge connection is established between the corresponding nodes of the parent peak and the isotope peak, and the weight of the edge is a preset high confidence constant.

[0066] Furthermore, constructing the initial peak matching plot also includes:

[0067] Before constructing the initial peak matching map, based on the mass-to-charge ratio of each peak in the cleaned peak set, peaks that satisfy the multi-charge distribution characteristics are identified. Based on their isotopic spacing or charge sequence characteristics, the number of charges z they carry is inferred. If z ≥ 2, a multi-charge peak is obtained.

[0068] When constructing the initial peak matching map, for the identified multi-charge peaks, their mass-to-charge ratio is converted into a neutral mass M based on their charge number z. The conversion formula is as follows:

[0069] ;

[0070] Where m / z is the mass-to-charge ratio of the multi-charge peak, and z is the charge number. The mass of the proton is preferably 1.0078 Da;

[0071] For peaks not identified as having multiple charges, their charge number z = 1 is set, and their equivalent mass M is calculated using the same formula; when calculating the mass-to-charge ratio similarity between peaks, the equivalent mass M is used instead of the original mass-to-charge ratio.

[0072] Specifically, isotope peaks and multi-charge peaks are common but easily misjudged mass spectrometry phenomena. Before constructing a peak matching diagram based on the cleaned peak set, this embodiment further introduces two key processing steps: isotope peaks and multi-charge peaks, to improve the accuracy of matching. Identification and preprocessing of isotope peak pairs: For the cleaned peak set, the system identifies possible isotope peak pairs. If two peaks meet the following conditions: (1) The difference in mass-to-charge ratio falls between 0.99 and 1.00 (Da), which is within the typical ¹³C isotope mass shift range; (2) The retention time difference does not exceed the preset tolerance, with a preferred preset tolerance of 0.1 minutes, to ensure that the two have highly consistent chromatographic elution behavior; (3) The intensity ratio of the isotope peak to the parent peak conforms to the distribution law of natural isotopes. Preferably, the intensity of the single carbon isotope peak is between 0.8% and 15% of that of its parent peak, and this ratio remains relatively stable in different biological samples. Preferably, the relative standard deviation is less than 30%. For such peak pairs, a dedicated connection edge will be established between the corresponding two nodes in the subsequently constructed peak matching graph, and the weight of this edge will be set to a preset high-confidence constant, which is close to 1.0, preferably 0.95. This operation ensures that the isotopic relationship is explicitly preserved in the graph structure, avoiding misclassification as two independent metabolites during subsequent matching.

[0073] This embodiment further identifies multi-charge peaks. The identification criteria for peaks that may carry multiple charges include: (1) a regular mass-to-charge ratio spacing between peaks, such as Δ(m / z)≈1 / z; (2) the existence of peak clusters that conform to the characteristics of charge sequence, such as when z=2, the isotope peak spacing is about 0.5 Da. For peaks identified as multi-charged, based on their inferred charge number z ≥ 2, their original mass-to-charge ratio is converted into neutral molecular mass, i.e., the equivalent mass after deconvolution. For peaks not identified as multi-charged, their charge number is assumed to be 1, and their equivalent mass is calculated in the same way. When calculating the mass-to-charge ratio similarity between peaks in the subsequent calculation, the original mass-to-charge ratio is no longer used, but the equivalent mass is uniformly used as the comparison benchmark. This conversion eliminates the mass-to-charge ratio difference caused by different charge states, so that peaks generated by the same compound under different ionization states can be accurately matched on a unified mass scale, which significantly improves cross-sample consistency. Through the above two preprocessing steps, this embodiment has taken into account the two common mass spectrometry complexities of isotope relationship and charge state difference before constructing the peak matching map. This not only reduces missed matches caused by chemically identical peaks but different apparent parameters, but also avoids misclassifying isotopic or multi-charge peaks as different metabolites, thus laying a more reliable chemical foundation for subsequent high-precision peak matching based on graph theory.

[0074] Based on the peak matching diagram, the matching relationship between peaks representing the same metabolite in different samples is determined, resulting in multiple effective matching groups. Each effective matching group contains a set of peaks from different samples that are associated with the same metabolite. The matching relationship satisfies basic similarity constraints as well as advanced matching constraints based on chromatographic-mass spectrometry continuity and instrument resolution.

[0075] Furthermore, based on the peak matching diagram, the matching relationship between peaks representing the same metabolite in different samples is determined, including:

[0076] Obtain the optimized peak matching diagram and initialize the matching result table, which is used to record the formed valid matching groups;

[0077] The nodes in the optimized peak matching graph are traversed sequentially as seed nodes. For each unmatched seed node, a current matching cluster is constructed starting from it, and the current node is used as the current node. The cluster is expanded using a depth-first search strategy. During the expansion process, each unmatched neighbor of the current node in the peak matching graph is treated as a candidate neighbor node, and the following matching decision is performed:

[0078] (a) Determine whether the retention time interval between the candidate neighbor node and the current node does not exceed a preset threshold;

[0079] (b) Determine whether the rate of change of the mass-to-charge ratio between the candidate adjacent node and the current node does not exceed a preset rate of change threshold;

[0080] (c) Obtain the mass spectrometer resolution, determine the minimum resolvable mass-to-charge ratio difference at that resolution, and determine whether the difference between the candidate adjacent node and the current node is not less than the minimum resolvable mass-to-charge ratio difference.

[0081] If all of the above (a) to (c) are satisfied, the experimental batch identifiers of the candidate adjacent nodes and the current node are obtained. If they belong to the same experimental batch, the graph edge weight between the two nodes is increased. If they do not belong to the same experimental batch, the temperature when each sample is collected is obtained. If the temperature difference exceeds the preset temperature difference threshold, the graph edge weight between the two nodes is decreased.

[0082] Determine whether the mass-to-charge ratio difference, retention time difference, peak intensity, and dynamically adjusted graph edge weights between the candidate adjacent nodes and the current node meet the preset similarity constraints.

[0083] If the conditions are met, the candidate neighbor node is included in the current matching cluster, and the depth-first search is continued with the candidate neighbor node as the new current node.

[0084] If the current matching cluster cannot be further expanded, it is recorded as a valid matching group if the number of peaks it contains is not less than 3.

[0085] If a round of traversal fails to generate a new valid matching group, the similarity constraint is relaxed and the matching process is re-executed until the preset iteration termination condition is met.

[0086] After matching is complete, output the matching result table.

[0087] Furthermore, determining whether the mass-to-charge ratio difference, retention time difference, peak intensity, and dynamically adjusted graph edge weights between the candidate adjacent nodes and the current node satisfy preset similarity constraints includes:

[0088] Obtain the mass-to-charge ratio difference, retention time difference, peak intensity, and dynamically adjusted graph edge weights between the candidate adjacent nodes and the current node;

[0089] Determine whether the mass-to-charge ratio difference is less than a preset mass-to-charge ratio difference threshold;

[0090] Determine whether the retention time difference is less than a preset retention time difference threshold;

[0091] Determine whether the peak intensity is within a preset effective range;

[0092] Determine whether the dynamically adjusted edge weights of the graph are greater than the preset weight threshold;

[0093] If all the above-mentioned preset similarity constraints are met, the candidate adjacent nodes will be included in the current matching cluster.

[0094] Among them, increasing the graph edge weight between two nodes means adding a first offset to the graph edge weight between the candidate adjacent node and the current node.

[0095] Reduce the graph edge weight between two nodes, that is, reduce the second offset based on the graph edge weight between the candidate adjacent node and the current node;

[0096] The first offset and the second offset may be the same or different.

[0097] Specifically, this embodiment identifies peaks belonging to the same metabolite from different samples by analyzing peak matching graphs, thereby forming effective matching groups. This process considers not only basic similarity constraints, such as direct comparisons of graph edge weights, mass-to-charge ratios, retention times, and peak intensities, but also advanced matching constraints such as chromatographic-mass spectrometry continuity and instrument resolution to improve the accuracy and reliability of the matching. The specific implementation steps are as follows: Initialize matching, obtain the peak matching graph, and initialize a matching result table. This table records the final effective matching groups, each containing a set of peaks from different samples but associated with the same metabolite. Construct and expand matching clusters: For each node in the peak matching graph, i.e., each peak, if it has not yet been included in any matching cluster, it is considered a seed node, and the construction of a new matching cluster begins. A depth-first search strategy is adopted, starting from the seed node, attempting to add neighboring nodes with sufficient similarity to it to the current matching cluster. During the expansion process, a series of rigorous matching decisions are performed on each candidate adjacent node: Retention time interval check: confirming that the retention time difference between the candidate adjacent node and the current node does not exceed a preset threshold; Mass-to-charge ratio change rate check: verifying whether the change rate of the mass-to-charge ratio between the two nodes meets the preset standard; Resolution adaptability test: assessing whether the mass-to-charge ratio difference between the two nodes is greater than the minimum resolvable difference based on the mass spectrometer's resolution. Experimental batch and temperature adjustment: if two nodes belong to the same experimental batch, the weight of the edge between them is appropriately increased; if they are not in the same experimental batch, the edge weight is adjusted according to the temperature difference at the time of acquisition, and the weight is reduced if the temperature difference exceeds a certain threshold. Comprehensive judgment of similarity constraints: for candidate adjacent nodes that meet all the above conditions, further checks are performed to see if their mass-to-charge ratio difference, retention time difference, peak intensity, and dynamically adjusted graph edge weights all meet the preset similarity constraints. Specifically, this includes: mass-to-charge ratio difference less than the set threshold; retention time difference less than the specified allowable range; peak intensity within the effective range; and dynamically adjusted graph edge weights higher than the predefined minimum requirements. Once a candidate neighbor node satisfies all these conditions, it is added to the current matching cluster, and the search continues deeper, centered on this new node. The identification and iterative optimization of valid matching groups: When no more nodes can be added to the current matching cluster, if the cluster contains at least three peaks, it is considered a valid matching group and recorded in the matching results table. If a round of traversal fails to produce a new valid matching group, the similarity constraints are appropriately relaxed, and the matching process is re-executed until the predetermined iteration termination condition is reached, resulting in... Figure 4 The image shows the metabolite matching results after processing metabolomics data using a graph theory-based strategy. (Compared to...) Figure 4 In contrast, the metabolite matching results after processing with the traditional XCMS peak matching method are shown in the figure below. Figure 3As shown, the weight adjustment mechanism for specific situations increases the connection strength of nodes within the same experimental batch, while reducing the connection weight for nodes across batches with significant temperature differences. The first and second offsets can be flexibly set according to the actual situation, and can be equal or different to better reflect the influence of the actual experimental environment.

[0098] In the process of metabolite peak matching based on peak matching graph, this embodiment not only relies on the edge weights in the graph structure, but also comprehensively introduces multi-dimensional constraints such as sample relationship information, experimental condition parameters, chromatographic-mass spectrometry continuity requirements and instrument physical limitations, so as to improve the biological rationality and technical reliability of the matching results. Specifically, it includes the following aspects: (1) Dynamic similarity adjustment of sample relationship and experimental conditions. In order to reflect the impact of real experimental scenarios on data consistency, the following dynamic adjustment strategy is introduced when using inter-peak similarity (i.e., graph edge weight): Priority enhancement for the same batch: If two peaks to be matched come from samples of the same experimental batch, it is considered that their systematic deviation is small and the matching credibility is higher. Therefore, a positive offset (preferred, +0.1) is added on the basis of the original similarity to improve its matching priority; Temperature difference penalty: If two peaks come from different batches and the difference in ambient temperature when their samples are collected exceeds 5°C, it is considered that there may be retention time drift or ionization efficiency change caused by temperature. Therefore, a negative offset (preferred, +0.1) is subtracted on the basis of the original similarity. ), to reduce the probability of matching. The specific value of the above offset can be set according to the experimental system calibration. Its core purpose is to enable the matching process to adaptively respond to the differences in experimental conditions. (2) Chromatography-mass spectrometry continuity constraints. In order to ensure that the matched peaks have continuity in chemical and physical behavior, the following hard constraints are set: retention time continuity: the retention time difference between any two matched peaks shall not exceed 1 minute, so as to exclude peaks that may belong to different compounds due to significant differences in elution behavior; mass-to-charge ratio change rate limit: for adjacent peak pairs in the figure (such as candidate neighbors considered in the DFS expansion process), the relative change rate of their mass-to-charge ratio, i.e.

[0099] The ratio must not exceed 5% to prevent incorrect association of peaks with excessively large mass differences. (3) Instrument resolution constraints: the matching process must strictly adhere to the physical resolution limit of the mass spectrometer used. For example, if a mass spectrometer with a resolution R of 5000 is used, the minimum mass-to-charge ratio difference that it can resolve within the target mass range is approximately 0.1 Da. The specific value can be determined according to the resolution formula. For peak pairs with a mass-to-charge ratio difference less than the minimum resolvable difference, even if other characteristics are similar, they are not considered independent metabolites to avoid splitting the same signal for matching; conversely, if the mass-to-charge ratio difference is small but still greater than the instrument's resolvable limit, it is allowed to participate in the matching judgment. This constraint ensures that the matching results conform to the instrument's physical performance and avoids over-resolution misjudgment. (4) Comprehensive judgment conditions for successful matching: For a peak pair to be finally included in the same matching group, it must simultaneously meet the following basic similarity thresholds: dynamically adjusted graph edge weight ≥ 0.6; mass-to-charge ratio difference < 0.1 Da; retention time difference < 0.3 minutes; peak intensity is within the effective detection range (e.g., 1000–10,000 counts) to exclude interference from noise peaks or saturation peaks. In actual operation, due to the complexity and diversity of biological samples, some special situations or limiting factors may be encountered, resulting in the inability to find enough effective matching groups by strictly following the initially set strict thresholds. For example, the performance of some metabolites in different samples may be affected by changes in the experimental environment, causing slight changes in their characteristic parameters (such as mass-to-charge ratio, retention time, etc.). Therefore, when a new effective matching group is not generated in one round of traversal, appropriately relaxing these constraints can help identify peaks that, although slightly varied, still belong to the same metabolite, thereby increasing the success rate and coverage of matching. The iterative process will stop when any of the following conditions are met: the preset maximum number of iterations is reached; the number of newly added effective matching groups no longer increases, that is, no new matching relationship is found in several consecutive iterations; all potential matching combinations have been tried and no new matching that meets the conditions appears. In this way, the quality of matching can be guaranteed, and all possible related peaks can be covered as comprehensively as possible, avoiding the omission of important metabolite information. (5) Matching group size control: In order to balance the sensitivity and specificity of matching, a size constraint is set for each effective matching group: the minimum number of matches is 3: that is, a matching group must contain peaks from at least 3 different samples to ensure its statistical stability and biological reproducibility; the maximum number of matches is dynamically set: it can be flexibly adjusted according to the size of the dataset and the research objectives. For example, in a small-scale validation experiment, the upper limit can be set to 10 to prevent multiple different metabolites from being excessively aggregated into a single cluster due to accidental similarity. Through this series of meticulously designed matching rules, the present invention can efficiently and accurately identify and associate the same metabolite peak in different samples, greatly improving the quality and efficiency of metabolomics data analysis.

[0100] Furthermore, determining the matching relationship between peaks representing the same metabolite in different samples also includes:

[0101] When processing large-scale metabolomics data, a greedy strategy is used to select a subset of unmatched neighboring nodes as candidate neighboring nodes to reduce the number of matching decisions.

[0102] When studying rare metabolites and requiring extremely high matching accuracy, a local matching subgraph is constructed based on the current node and its unmatched neighboring nodes. The Hungarian algorithm is then used to solve for the globally optimal one-to-one match in this subgraph to improve matching accuracy.

[0103] Specifically, in the peak matching process, to balance computational efficiency and matching accuracy, this embodiment dynamically adopts different neighbor node processing strategies based on the data scale and research objectives: For large-scale metabolomics data (such as more than 100 samples or hundreds of thousands of peaks), to reduce computational complexity, a greedy selection strategy is adopted for the unmatched neighbor nodes of the current node: that is, only a few neighbor nodes with the highest overall similarity ranking (e.g., the top 5 with the highest similarity) are retained as candidate neighbor nodes, giving priority to high-confidence matching paths, thereby significantly reducing the number of subsequent matching decisions and improving overall processing efficiency. For rare metabolite research or scenarios with extremely high matching accuracy requirements (such as biomarker validation), a local matching subgraph is constructed based on the current node and all its unmatched neighbor nodes, and the matching problem in this subgraph is modeled as a bipartite graph maximum weight matching problem. The classic Hungarian algorithm is used to solve its globally optimal one-to-one matching scheme. Although this method has a large computational cost, it can avoid the local suboptimal solutions that may be introduced by the greedy strategy, thereby significantly improving the accuracy and reliability of matching.

[0104] Furthermore, based on the peak matching diagram, the matching relationship between peaks representing the same metabolite in different samples is determined to obtain multiple valid matching groups, which also includes:

[0105] The matching results are evaluated based on a validation dataset containing known metabolite annotations, where true matches are determined by standards or manual correction. The evaluation metrics include, but are not limited to, the following:

[0106] Matching accuracy: The ratio of the number of peak pairs judged as correct matches in the matching results to the total number of matching peak pairs output by the algorithm, in order to evaluate the accuracy of the matching algorithm;

[0107] Recall: The ratio of the number of correctly matched peak pairs in the matching results to the total number of all real-world homologous peak pairs in the validation dataset, used to evaluate the completeness of the matching algorithm;

[0108] F1 score: The harmonic mean of the matching accuracy and recall is calculated to more comprehensively evaluate the overall performance of the matching algorithm.

[0109] Specifically, this embodiment introduces a quantitative evaluation mechanism based on a known labeled validation dataset. The validation dataset contains a set of authoritatively confirmed metabolite peaks and their true cross-sample matching relationships. These true matching relationships are determined through one of the following methods: using standards to collect data under the same experimental conditions to clarify the chemical identity and correspondence of each peak; or having domain experts manually calibrate and compare the matching results with mass spectrometry libraries (such as NIST and HMDB) to form a high-confidence reference matching set. Based on this, the following standardized metrics are used to quantitatively evaluate the matching results: Matching accuracy: the proportion of all matching peak pairs output by the algorithm that are judged as correct matches. This metric reflects the reliability of the matching results, i.e., how many of the matched results are true. Recall: the proportion of all truly existing homologous peak pairs in the validation dataset that are successfully matched by the algorithm. This metric reflects the coverage of the matching results, i.e., how many truly existing matches are found. F1 score: as the harmonic mean of accuracy and recall, it comprehensively reflects the algorithm's performance in balancing precision and completeness, and is particularly suitable for scenarios where both false positives and false negatives need to be controlled in matching tasks. The above evaluation process provides an objective basis for the optimization and parameter tuning of the matching algorithm, ensuring that it has verifiable performance under different experimental systems.

[0110] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A graph theory-based method for metabolomics data processing, characterized in that, The method includes: Obtain raw peak data of metabolomics from multiple biological samples. The raw peak data includes the mass-to-charge ratio and retention time of each peak, as well as the peak intensity, peak area and peak width of each peak in each sample. The original peak data is cleaned to obtain a cleaned peak set. Each peak in the peak set consists of multidimensional feature data composed of its mass-to-charge ratio, retention time and peak intensity. A peak matching graph is constructed based on the multidimensional feature data, where graph nodes represent each peak and are associated with its multidimensional feature data, and graph edge weights reflect the comprehensive similarity between peaks of any two different samples. The comprehensive similarity is obtained by normalizing the similarity based on mass-to-charge ratio, retention time, and peak intensity and then weighting and fusing them. Based on the peak matching diagram, the matching relationship between peaks representing the same metabolite in different samples is determined, resulting in multiple effective matching groups. Each effective matching group contains a set of peaks from different samples that are associated with the same metabolite. The matching relationship satisfies basic similarity constraints as well as advanced matching constraints based on chromatographic-mass spectrometry continuity and instrument resolution. The construction of the peak matching graph includes: Each peak in the peak set is used as a graph node, and each peak is associated with its corresponding multidimensional feature data, including mass-to-charge ratio, retention time and peak intensity. The space of the mass-to-charge ratio is divided into multiple intervals of preset width, and only peak pairs falling into the same or adjacent intervals are retained as candidate correlation pairs. The peak intensity is preprocessed by performing a logarithmic transformation to obtain the preprocessed peak intensity value. The logarithmic transformation formula is as follows: ； Where I' is the preprocessed peak intensity value, and I is the peak intensity in the multidimensional feature data; Calculate the overall similarity of each peak pair in the candidate association pairs, where each peak pair comes from peaks in different samples, using the following formula: (1) Calculate the mass-to-charge ratio similarity: ； in, and Let be the mass-to-charge ratios of any two peaks. and These are the maximum and minimum mass-to-charge ratios for all peaks in the entire set of peaks; (2) Calculate retention time similarity: ； in, and , where represents the retention time of any two peaks. and These are the maximum and minimum values ​​of all retention times in the entire set of peaks, respectively; (3) Calculate the intensity similarity: ； in, and These are the peak intensity values ​​after logarithmic transformation of any two peaks. and These are the maximum and minimum values ​​of the peak intensity after logarithmic transformation in the entire peak set, respectively; The three similarities mentioned above are weighted and fused to obtain the comprehensive similarity score, as shown in the following formula: Sim=w1·Sim(m / z)+w2·Sim(tR)+w3·Sim(I'); Where Sim represents the overall similarity, w1, w2, and w3 are preset weight coefficients, and w1+w2+w3=1, w1=0.4, w2=0.35, and w3=0.25; Using the comprehensive similarity as the graph edge weights, an initial peak matching graph is constructed; The initial peak matching graph is subjected to graph optimization processing, including: removing self-looping edges and sparsifying edges with a degree less than 3 to obtain an optimized peak matching graph; Determining the matching relationship between peaks representing the same metabolite in different samples includes: Obtain the optimized peak matching diagram and initialize the matching result table, which is used to record the formed valid matching groups; The nodes in the optimized peak matching graph are traversed sequentially as seed nodes. For each unmatched seed node, a current matching cluster is constructed starting from it, and the current node is used as the current node. The cluster is expanded using a depth-first search strategy. During the expansion process, each unmatched neighbor of the current node in the peak matching graph is treated as a candidate neighbor node, and the following matching decision is performed: (a) Determine whether the retention time interval between the candidate neighbor node and the current node does not exceed a preset threshold; (b) Determine whether the rate of change of the mass-to-charge ratio between the candidate adjacent node and the current node does not exceed a preset rate of change threshold; (c) Obtain the mass spectrometer resolution, determine the minimum resolvable mass-to-charge ratio difference at that resolution, and determine whether the difference between the candidate adjacent node and the current node is not less than the minimum resolvable mass-to-charge ratio difference. If all of the above (a) to (c) are satisfied, the experimental batch identifiers of the candidate adjacent nodes and the current node are obtained. If they belong to the same experimental batch, the graph edge weight between the two nodes is increased. If they do not belong to the same experimental batch, the temperature when each sample is collected is obtained. If the temperature difference exceeds the preset temperature difference threshold, the graph edge weight between the two nodes is decreased. Determine whether the mass-to-charge ratio difference, retention time difference, peak intensity, and dynamically adjusted graph edge weights between the candidate adjacent nodes and the current node meet the preset similarity constraints. If the conditions are met, the candidate neighbor node is included in the current matching cluster, and the candidate neighbor node is used as the new current node to continue the depth-first search. If the current matching cluster cannot be further expanded, it is recorded as a valid matching group if the number of peaks it contains is not less than 3. If a round of traversal fails to generate a new valid matching group, the similarity constraint is relaxed and the matching process is re-executed until the preset iteration termination condition is met. After matching is complete, output the matching result table.

2. The metabolomics data processing method based on graph theory strategy according to claim 1, characterized in that, Before acquiring raw metabolomics peak data from multiple biological samples, the process includes: Obtain raw chromatographic-mass spectrometry data from multiple biological samples; The original chromatographic-mass spectrometry data are preprocessed, including noise-reduced chromatograms and mass spectra, to obtain noise-reduced chromatographic-mass spectrometry data. Based on the noise-reduced chromatographic-mass spectrometry data, peak detection and feature extraction are performed to generate an initial peak data table. The initial peak data table includes the mass-to-charge ratio, retention time, peak intensity, peak area, and peak width of each peak in each sample, forming raw metabolomics peak data for multiple biological samples.

3. The metabolomics data processing method based on graph theory strategy according to claim 1, characterized in that, The original peak data is cleaned to obtain a cleaned peak set, including: The quality cleaning includes removing abnormal peaks, which is achieved based on at least one characteristic parameter of retention time, peak area, and peak width, and is accomplished in the following manner: Set the allowable range for each characteristic parameter, and take the peaks that exceed the allowable range as candidate abnormal peaks. Further confirm whether they are abnormal peaks through box plot analysis. If they are, remove them to obtain the cleaned peak set.

4. The metabolomics data processing method based on graph theory strategy according to claim 1, characterized in that, Constructing the initial peak matching plot also includes: Before constructing the initial peak matching map, potential isotope peak pairs are identified based on the cleaned peak set: if the mass-to-charge ratio difference between two peaks falls within the range of [0.99, 1] Da, the retention time difference is less than the preset tolerance, and the intensity ratio conforms to the distribution law of natural isotopes, then the two are marked as the parent peak and the isotope peak. When constructing the initial peak matching diagram, an edge connection is established between the corresponding nodes of the parent peak and the isotope peak, and the weight of the edge is a preset high confidence constant.

5. The metabolomics data processing method based on graph theory strategy according to claim 1, characterized in that, Constructing the initial peak matching plot also includes: Before constructing the initial peak matching map, based on the mass-to-charge ratio of each peak in the cleaned peak set, peaks that satisfy the multi-charge distribution characteristics are identified. Based on their isotopic spacing or charge sequence characteristics, the number of charges z they carry is inferred. If z ≥ 2, a multi-charge peak is obtained. When constructing the initial peak matching map, for the identified multi-charge peaks, their mass-to-charge ratio is converted into a neutral mass M based on their charge number z. The conversion formula is as follows: ； Where m / z is the mass-to-charge ratio of the multi-charge peak, and z is the charge number. The mass of the proton is 1.0078 Da. For peaks not identified as having multiple charges, their charge number z = 1 is set, and their equivalent mass M is calculated using the same formula; when calculating the mass-to-charge ratio similarity between peaks, the equivalent mass M is used instead of the original mass-to-charge ratio.

6. The method for processing metabolomics data based on graph theory strategy according to claim 1, characterized in that, Determining whether the mass-to-charge ratio difference, retention time difference, peak intensity, and dynamically adjusted graph edge weights between the candidate adjacent nodes and the current node satisfy preset similarity constraints includes: Obtain the mass-to-charge ratio difference, retention time difference, peak intensity, and dynamically adjusted graph edge weights between the candidate adjacent nodes and the current node; Determine whether the mass-to-charge ratio difference is less than a preset mass-to-charge ratio difference threshold; Determine whether the retention time difference is less than a preset retention time difference threshold; Determine whether the peak intensity is within a preset effective range; Determine whether the dynamically adjusted edge weights of the graph are greater than the preset weight threshold; If all the above-mentioned preset similarity constraints are met, the candidate adjacent nodes will be included in the current matching cluster. Among them, increasing the graph edge weight between two nodes means adding a first offset to the graph edge weight between the candidate adjacent node and the current node. Reduce the graph edge weight between two nodes, that is, reduce the second offset based on the graph edge weight between the candidate adjacent node and the current node; The first offset and the second offset may be the same or different.

7. The metabolomics data processing method based on graph theory strategy according to claim 1, characterized in that, Also includes: When processing large-scale metabolomics data, a greedy strategy is used to select a subset of unmatched neighboring nodes as candidate neighboring nodes to reduce the number of matching decisions. When studying rare metabolites and requiring extremely high matching accuracy, a local matching subgraph is constructed based on the current node and its unmatched neighboring nodes. The Hungarian algorithm is then used to solve for the globally optimal one-to-one match in this subgraph to improve matching accuracy.

8. The metabolomics data processing method based on graph theory strategy according to claim 1, characterized in that, Also includes: The matching results are evaluated based on a validation dataset containing known metabolite annotations, where true matches are determined by standards or manual calibration. The evaluation metrics include the following aspects: Matching accuracy: The ratio of the number of peak pairs judged as correct matches in the matching results to the total number of matching peak pairs output by the algorithm, in order to evaluate the accuracy of the matching algorithm; Recall: The ratio of the number of correctly matched peak pairs in the matching results to the total number of all real-world homologous peak pairs in the validation dataset, used to evaluate the completeness of the matching algorithm; F1 score: The harmonic mean of the matching accuracy and recall is calculated to more comprehensively evaluate the overall performance of the matching algorithm.

Images