A multi-agent automated analysis system, method and medium for liquid chromatography-tandem mass spectrometry qualitative and quantitative analysis

Abstract

A multi-agent automatic analysis system and method for liquid chromatography-tandem mass spectrometry qualitative and quantitative analysis and a medium, relate to instrument analysis and artificial intelligence field. The system comprises an input module, a central agent, a sub-task agent, a manual confirmation module and a report output module. The input module pre-scans the detection data and generates a data list. The central agent with a large language model analyzes the natural language requirements, formulates a structured task plan in combination with the list, and schedules various sub-task agents to complete the spectrum qualitative analysis, concentration quantitative analysis, instrument method configuration and methodology verification. The system suspends the process and enters manual review under abnormal conditions such as insufficient confidence, missing path, and out-of-specification quality control. The report output module generates a report and an audit tracking file by summarizing the data. The mass spectrometry detection process is fully automated, and the problems of dispersion in traditional links, complicated manual operation, scattered qualitative basis and poor process traceability are improved, and the detection efficiency, result stability and compliance traceability are improved.

Description

Technical Field

[0001] This invention relates to the fields of instrumental analysis and artificial intelligence, and in particular to a multi-agent automated analysis system, method and medium for qualitative and quantitative analysis by liquid chromatography-tandem mass spectrometry. Background Technology

[0002] Liquid chromatography-tandem mass spectrometry (LC-MS / MS) is a crucial technological platform for drug analysis, clinical diagnostics, food testing, and environmental monitoring. A complete LC-MS / MS analysis workflow involves multiple stages, including sample pretreatment, instrument parameter configuration, sample sequence generation, raw data acquisition, spectral retrieval and qualitative compound annotation, spectral integration and quantitative compound determination, method validation, results summarization, and report writing. The entire analytical process is time-consuming, labor-intensive, and the results are easily affected by differences in operator experience.

[0003] Existing analytical workflows have achieved a certain degree of automation at the level of individual software modules, such as peak identification (Louail, P., Brunius, C., Garcia-Aloy, M., Kumler, W., Storz, N., Stanstrup, J., Treutler, H., Vangeenderhuysen, P., Witting, M., Neumann, S., & Rainer, J. (2025). xcms in Peak Form: Now Anchoring a Complete Metabolomics DataPreprocessing and Analysis Software Ecosystem. Analytical Chemistry, 97(50), 27639–27645) and spectral library retrieval (Huber, F., van der Burg, S., van der Hooft, JJJ, & Ridder, L. (2021). MS2DeepScore: a novel deep learning similarity measure to compare tandem mass spectra). Journal of Cheminformatics , 13(1), 84), Compound Notes (Stravs, MA, Dührkop, K., Böcker, S., & Zamboni, N. (2022). MSNovelist: denovo structure generation from mass spectra. Nature Methods, 19(7), 819–825) and quantitative integration (Henderson, CM, Shulman, NJ, MacLean, B., MacCoss, MJ, & Hoofnagle, AN (2018). Skyline Performs as Well as Vendor Software in theQuantitative Analysis of Serum 25-Hydroxy Vitamin D and Vitamin D BindingGlobulin. Clinical Chemistry Tools such as 64(2), 408–410 can improve data processing efficiency at each stage, reduce repetitive manual operations, and to some extent improve the standardization level of liquid chromatography-tandem mass spectrometry analysis. However, the above-mentioned models still have the following problems:

[0004] (1) Automation usually stops at a single step rather than covering the entire analysis task. Analysts usually need to frequently transfer data and select parameters between instrument workstations, spectral library search tools, quantitative software, statistical scripts and report templates, and manually summarize to form the final report conclusions.

[0005] (2) Insufficient evidence integration capability for qualitative annotation. Existing methods usually output candidate compound lists, matching scores or ranking results, which are difficult to automatically integrate the results of different algorithms and different evidence sources. High-quality annotation results still need to rely on expert experience.

[0006] (3) The validation of quantitative methods lacks a standardized automated judgment process. Liquid chromatography-tandem mass spectrometry quantitative analysis includes the regular judgment of calibration samples, quality control samples and unknown samples. Core indicators such as linearity, selectivity and residue need to be manually evaluated by the analyst based on the back-calculated concentration obtained by the software. Exclusion points, threshold basis and original calculated values ​​are not easily preserved completely.

[0007] (4) The analysis process lacks a complete audit chain. Existing tools can usually only save the processing results of a single software, making it difficult to link the original input files, actual operation procedures, intermediate calculation processes and version information in a unified manner to obtain a standardized report that is traceable but cannot be modified, resulting in high costs for result verification, cross-laboratory reproduction and management.

[0008] These challenges mean that existing tools for liquid chromatography-tandem mass spectrometry analysis still fall short of meeting practical analytical needs in terms of practicality, standardized analysis, and traceability. Summary of the Invention

[0009] The purpose of this invention is to address the shortcomings of existing technologies in terms of automation, standardization, and traceability by providing a multi-agent automated analysis system, method, and medium for qualitative and quantitative analysis using liquid chromatography-tandem mass spectrometry, thereby improving analytical efficiency, result reliability, and process traceability.

[0010] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0011] In a first aspect, the present invention provides a multi-agent automated analysis system for qualitative and quantitative analysis by liquid chromatography-tandem mass spectrometry, including an input module, a central agent, sub-task agents, a report output module, and a manual confirmation module. The central agent establishes data connections with the input module, all sub-task agents, the manual confirmation module, and the report output module, respectively.

[0012] The input module is used to acquire natural language analysis requirements and structured data to be analyzed, including one or more of sample information, instrument parameters, mass spectrometry data, and report templates; the input module can pre-scan the acquired structured data to be analyzed and generate a data list to constrain the task scheduling scope;

[0013] The central intelligent agent is equipped with a large language model and generates a structured task plan based on natural language analysis needs, including analysis type, data path, execution order and called sub-task intelligent agents;

[0014] The sub-task intelligent agent includes a method configuration sub-agent, a qualitative analysis sub-agent, a quantitative analysis sub-agent, and a method validation sub-agent. Each sub-task intelligent agent contains at least a qualitative analysis sub-agent and a quantitative analysis sub-agent as core execution units. The method configuration sub-agent is used to generate instrument method parameters, sample sequences, and analysis procedures. The qualitative analysis sub-agent is used to perform spectral retrieval, candidate compound screening, evidence fusion, and consensus ranking. The quantitative analysis sub-agent is used to perform spectral integration, response ratio calculation, calibration curve fitting, and compound concentration determination. The method validation sub-agent is used to perform methodological evaluation on calibration samples, quality control samples, and unknown samples according to preset rules.

[0015] The report output module is used to generate standardized reports and audit trail records based on qualitative annotation results, quantitative measurement results, methodological validation results, and manual verification records;

[0016] The manual confirmation module is used to pause the automatic process and receive user confirmation or correction when the analysis task is unclear, the file path is unclear, there are low-confidence spectral annotations, or the calibration or quality control samples are outside the verification range.

[0017] Secondly, this invention provides a multi-agent automated analysis method for qualitative and quantitative analysis using liquid chromatography-tandem mass spectrometry. This method starts with user-input natural language analysis requirements and the file to be analyzed. A large language model automatically identifies the analysis type and target output, generates a structured task plan, and calls corresponding sub-task agents to complete method configuration, qualitative annotation, quantitative determination, and methodological validation, ultimately generating a standardized report and audit trail.

[0018] Thirdly, the present invention provides a computer-readable storage medium. The computer-readable storage medium stores a computational program that, when executed, implements the aforementioned multi-agent automated analysis method.

[0019] The technical problem solved by this invention:

[0020] This invention addresses the problems of insufficient automation and difficulty in unified scheduling of various stages in existing liquid chromatography-tandem mass spectrometry analysis workflows. It reduces the workload of analysts who frequently switch between various software and analysis stages, repeatedly manually set parameters, and summarize results. At the same time, it solves the problems of difficulty in automatically integrating multiple spectral evidences in qualitative annotation, reliance on manual judgment for quantitative method validation, difficulty in fully preserving exclusion points and threshold criteria, and difficulty in unifying and associating original input files, intermediate calculation processes, manual confirmation records, and final reports. Thus, it achieves complete automation and traceability management of the entire workflow from input of analytical requirements to generation of standardized reports.

[0021] Compared with the prior art, the advantages of the present invention are:

[0022] This invention utilizes a central intelligent agent to parse natural language analysis requirements and structured data, and then invokes corresponding sub-task intelligent agents to execute each step, extending the automation of liquid chromatography-tandem mass spectrometry (LC-MS / MS) analysis from single-step automation to complete workflow automation. This invention integrates sample information, instrument parameters, mass spectrometry data, and report templates into a unified scheduling process, improving analytical efficiency and reducing manual workload. Specifically, the qualitative analysis sub-agent is used for candidate compound screening, evidence fusion, and consensus ranking, enhancing the reliability and verifiability of qualitative annotation results. The quantitative analysis sub-agent and the methodology validation sub-agent complete spectral integration, calibration curve fitting, concentration determination, and evaluation of calibration samples, quality control samples, and unknown samples, improving the consistency and standardization of quantitative and validation results. The report output module generates standardized reports and audit trail records, unifying qualitative annotation, quantitative determination, methodology validation, and manual verification records for easy result verification, cross-laboratory reproducibility, and quality management. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the structural composition of an embodiment of the present invention.

[0024] Figure 2 This is a flowchart of qualitative analysis evidence fusion.

[0025] Figure 3 This is a flowchart for the validation of quantitative analysis methodologies.

[0026] Figure 4 This is a graph showing the results of a quantitative analysis of carbamazepine. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the following embodiments will be used in conjunction with the accompanying drawings to further illustrate the invention. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0028] In this embodiment, the multi-agent automated analysis system for qualitative and quantitative analysis by liquid chromatography-tandem mass spectrometry includes an input module, a central agent, sub-task agents, a report output module, and a manual verification module. The input module, central agent, sub-task agents, report output module, and manual verification module are interconnected to realize the input of analytical requirements, task planning, task execution, result output, and manual verification. The overall structure is as follows: Figure 1 As shown.

[0029] The input module receives user-input natural language analysis requests and files to be analyzed, and pre-scans the files to be analyzed to generate a candidate data list. The files to be analyzed include one or more of the following: sample information, instrument parameter table, injection sequence table, qualitative analysis data, quantitative analysis data, method validation data, and report template. The sample information includes sample number, sample type, batch number, sampling time, theoretical concentration, and internal standard information. The instrument parameter table includes information such as ion mode, scan mode, multiple reaction monitoring (MRM) ion pairs, mobile phase composition, gradient elution program, and injection volume. The injection sequence table includes sample injection order, sample location, sample name, and sample type information. The qualitative analysis data, quantitative analysis data, and method validation data can be converted mzML files or readable data files exported from the instrument workstation. The report template is a standardized report template pre-configured in the system of this invention, including a report title, chapter structure, table fields, graphic fields, and audit trail fields, used to constrain the report output module to generate a uniformly formatted analysis report.

[0030] The central agent employs a large language model as the core of task parsing, used to generate structured task plans based on natural language analysis requirements and a candidate data list. The structured task plan includes the analysis type, input file path, output file path, execution order, called sub-task agents, and necessary analysis parameters. The central agent determines the tasks to be executed based on the candidate data list and preset task types, rather than directly generating analysis results arbitrarily. The preset task types include one or more of the following: method parameter generation, injection sequence generation, qualitative annotation, quantitative determination, pharmacokinetic analysis, method validation, and report generation.

[0031] The sub-task agents include a method configuration sub-agent, a qualitative analysis sub-agent, a quantitative analysis sub-agent, and a method validation sub-agent. The method configuration sub-agent is used to generate instrument method parameters and injection sequences; the qualitative analysis sub-agent is used to perform spectral retrieval, candidate compound screening, evidence fusion, and consensus ranking; the quantitative analysis sub-agent is used to perform MRM peak extraction, response ratio calculation, calibration curve fitting, and sample concentration determination; and the method validation sub-agent is used to perform methodological evaluation on calibration samples, quality control samples, and unknown samples according to preset rules.

[0032] The report output module generates standardized reports and audit trails based on qualitative annotation results, quantitative measurement results, methodological validation results, and manual verification records. The manual verification module pauses the automatic process and receives user confirmation or corrections when there are issues such as low-confidence spectral annotations, unclear file paths, inconsistent sample metadata, or abnormal calibration or quality control samples.

[0033] Based on the above system, the multi-agent automated analysis method provided by the present invention includes the following steps.

[0034] S100 obtains the user's input natural language analysis requirements and the file to be analyzed, and performs a pre-scan on the file to be analyzed.

[0035] Specifically, the input module reads the project data directory and identifies the qualitative data directory, quantitative data directory, method validation data directory, instrument parameter directory, and injection sequence directory. The system records the relative path, file type, executable task type, and corresponding analysis module for each file, generating a candidate data list. The candidate data list is used to limit the task planning scope of the subsequent central agent, preventing the central agent from selecting non-existent file paths or generating unsupported analysis tasks.

[0036] S200: The central intelligent agent generates a structured task plan based on the needs of natural language analysis.

[0037] Specifically, the central agent reads the user's input analysis requirements, such as "perform qualitative analysis and generate a report," "perform quantitative determination based on MRM data," "generate a pharmacokinetic analysis report," or "generate a method validation injection protocol." The central agent determines the analysis type based on the requirements and selects the corresponding input file, data path, and output path from the candidate data list to form a structured task plan. The structured task plan includes the analysis type, execution order, input file path, output file path, invoked sub-task agents, and necessary parameters.

[0038] S300, the method configuration sub-agent generates instrument method parameters and injection sequences.

[0039] Specifically, when the structured task plan includes a method parameter generation task, the method configuration sub-agent reads the instrument parameter table or user-provided methodological text, extracts parameters such as ion mode, scan mode, MRM ion pairs, retention time, collision energy, declustering voltage, injection volume, and chromatographic gradient, and combines this with a preset instrument method template to generate a method file that can be read or imported by the instrument workstation. When the structured task plan includes an injection sequence generation task, the method configuration sub-agent reads the sample information table and generates an injection sequence file based on the sample number, sample type, batch number, data file name, injection volume, blank sample, and quality control sample location. This reduces issues such as path errors, sample name errors, and missing quality control samples that occur when manually editing instrument methods and injection sequences.

[0040] S400, the qualitative analysis sub-agent performs spectral retrieval and candidate compound annotation.

[0041] Specifically, the qualitative analysis agent reads the mass spectrometry data to be analyzed and the background data, performs peak extraction, background subtraction, and MS / MS spectrum processing to obtain the spectrum to be annotated. Subsequently, the qualitative analysis agent matches the spectrum to be annotated with standard spectra in the spectrum library to obtain a list of candidate compounds.

[0042] In a preferred embodiment, the qualitative analysis sub-agent uses cosine similarity, adjusted cosine similarity, Spec2Vec, and MS2DeepScore algorithms to match the same MS / MS spectrum, with each method outputting candidate compounds and their ranking results. The system merges the candidate compounds output by each method into a restricted candidate set and records the source method, ranking position, and matching score of each candidate compound. Subsequently, the central agent or the qualitative analysis sub-agent performs evidence fusion and consensus ranking of the candidate compounds based on their frequency of occurrence in different methods, ranking consistency, matching score, precursor ion mass, fragment ions, high-intensity product ions, and neutral loss information. The consensus ranking is performed only within the restricted candidate set and does not add candidate compounds not supported by the spectrum search. The confidence level of candidate compounds is measured by calculating the average similarity of each method. When the overall confidence level of a candidate compound meets a preset threshold (set to 0.9 in this embodiment), the system outputs qualitative annotation results. When the candidate results have low confidence (such as ambiguity in adduct ions, difficulty in distinguishing structural isomers, or poor spectral quality), the system submits the spectrum to the manual confirmation module for user confirmation or correction before continuing execution. The overall execution flow is as follows: Figure 2 As shown.

[0043] The S500 quantitative analysis sub-agent performs MRM peak extraction, calibration curve fitting, and concentration determination.

[0044] Specifically, the quantitative analysis sub-agent locates the quantitative data catalog according to the structured task plan and identifies the batch data within it. For each batch, the quantitative analysis sub-agent reads the MRM data, extracts and integrates the chromatographic peaks of the analyte and internal standard, and obtains the peak area of ​​the analyte, the peak area of ​​the internal standard, and the analyte / internal standard response ratio.

[0045] The quantitative analysis sub-agent fits a calibration curve based on the theoretical concentration and response ratio of the calibration sample. The calibration curve can be unweighted, or weighted using 1 / x, 1 / x², 1 / y, or 1 / y² methods. The system calculates the back-calculated concentration of the calibration sample and determines whether the calibration curve is acceptable based on preset acceptance criteria. For unknown samples, the system calculates the sample concentration based on the calibration curve; for samples outside the calibration range, the system marks them as outside the validation range and submits them to the manual confirmation module for processing. In the pharmacokinetic analysis embodiment, the quantitative analysis sub-agent also reads the subject number, sampling time, and sample concentration, generating a concentration-time data table. The system can further generate a pharmacokinetic analysis report based on the concentration-time data.

[0046] S600, the methodology verification sub-agent performs standardized verification and evaluation.

[0047] Specifically, the methodology validation agent reads the methodology validation parameters and injection results, and performs regularized evaluations on calibration samples, quality control samples, and blank samples. The regularized evaluations include one or more of the following: linearity, selectivity, residual properties, dilution accuracy, matrix effect, accuracy, precision, and stability.

[0048] In one implementation, the method validation sub-agent first reads the method validation parameters and injection results to determine the validation objects, such as the calibration curve concentration level, the quality control sample concentration level, the diluted quality control sample, and samples from different matrix sources. Subsequently, the method validation sub-agent evaluates each method validation item according to preset acceptance criteria. For linearity items, the system determines whether the calibration curve is acceptable based on the theoretical concentration, response ratio, regression equation, and back-calculated concentration of the calibration sample; for selectivity items, the system determines whether there are interfering peaks in the blank matrix; for residual items, the system determines whether there is a residual response exceeding a threshold in the blank sample after injection of a high-concentration sample; for dilution accuracy items, the system determines whether the back-calculated concentration of the diluted quality control sample meets the preset deviation range; for matrix effect items, the system determines whether the matrix effect meets the requirements based on the response differences of samples from different matrix sources; for accuracy and precision items, the system calculates the accuracy and variability based on the measured concentration, theoretical concentration, and repeated measurement results of the quality control sample, and determines whether it meets the preset criteria; for stability items, the system determines whether the analyte is stable based on the measurement results of samples under different storage or processing conditions. When a verification item fails to meet the preset criteria, the system records the failed item, the original calculated value, the threshold basis, the exclusion point information, and the corresponding batch information. This information, along with the final verification conclusion, is saved for subsequent review. The overall execution flow is as follows: Figure 3 As shown.

[0049] S700, the report output module generates standardized reports and audit trail records.

[0050] Specifically, the report output module reads qualitative annotation results, quantitative determination results, method validation results, and manual verification records to generate a standardized report. The standardized report includes the project name, generation time, input file path, analysis type, execution parameters, candidate compound ranking, quantitative result table, calibration curve information, quality control judgment, method validation conclusion, anomaly alerts, and manual verification records.

[0051] The audit trail includes the original input file, candidate data list, structured task plan, invoked subtask agents, intermediate calculation results, model or rule version, manual confirmation content, and final report path. This enables continuous tracking from the input of natural language analysis requirements to the generation of the final report.

[0052] The technical solutions provided by the present invention will be described in detail below with reference to the embodiments. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0053] Application Example 1: Qualitative analysis of mass spectrometry data of licorice extract.

[0054] The user inputs a natural language request: "Qualitatively annotate and generate a report on the mass spectrometry data of licorice extract." The system input module first reads the mass spectrometry data, sample information, and report template of the licorice extract. The central agent generates a structured task plan based on the user's request, including peak extraction, background subtraction, spectral retrieval, candidate compound screening, evidence fusion, consensus ranking, and report generation. The qualitative analysis sub-agent reads the MS / MS spectra of the licorice extract in positive and / or negative ion modes, matches the spectrum to be annotated with standard spectra in the spectral library, and forms a restricted candidate set based on the candidate results of various spectral matching algorithms. Subsequently, the system performs consensus ranking based on the frequency of occurrence of candidate compounds in different algorithms, ranking consistency, matching score, precursor ion mass, fragment ions, and neutral loss information. For candidate compounds that meet the preset confidence threshold, the system outputs the qualitative annotation results, the first three rows of which are shown in Table 1; for low-confidence spectra or candidate results with isomer ambiguity, the system submits them to the manual confirmation module.

[0055] Table 1

[0056]

[0057] As shown in Table 1, the multi-agent system of this invention can accurately separate and precisely annotate active ingredients of different polarities and types in licorice extract. Arginine, a polar small-molecule amino acid, has the shortest retention time, abundant secondary fragments, and a 100% response intensity for the characteristic fragment 70.0659, indicating extremely high matching degree. Glycyrrhizin B, a characteristic glycoside active ingredient of licorice, stably eluted at 8.029 min, with a base peak at 127.0393, exhibiting strong spectral specificity. Isoliquiritigenin, a core flavonoid active substance of licorice, has a moderate retention time, regular fragment ion peak distribution, and prominent characteristic fragment response. After multi-algorithm evidence fusion and sorting, the overall confidence level of the three types of compounds is higher than the preset threshold of 0.9, with no isomer ambiguity or low matching degree issues, requiring no manual verification. This fully verifies the technical advantages of the qualitative analysis sub-agent system of this invention in multi-algorithm fusion, precise screening, and high-reliability annotation, effectively avoiding the matching error of a single algorithm and achieving precise qualitative identification of target components in complex plant extract matrices.

[0058] The report output module ultimately generates a qualitative analysis report of licorice extract and an audit trail, completing the fully automated analysis.

[0059] Application Example 2: Quantitative analysis of carbamazepine blood concentration data by mass spectrometry.

[0060] The user inputs a natural language request: "Generate a pharmacokinetic curve and output a report based on carbamazepine blood concentration data." The system reads the analyte carbamazepine and internal standard MRM data, injection sequence list, subject number, plasma sampling time, and report template corresponding to the carbamazepine sample. The central agent matches the candidate data list to generate a structured task plan for quantitative analysis, including peak integral, response ratio calculation, calibration curve fitting, concentration conversion, pharmacokinetic curve plotting, methodological quality control verification, and report archiving. The system then schedules the quantitative analysis sub-agent to execute the entire calculation process.

[0061] The quantitative analysis agent extracts the chromatographic peaks of carbamazepine and bupropion (internal standard) from the MRM chromatogram based on batch data, automatically performs baseline correction and peak integration, obtains the peak areas of the two substances, and calculates the response ratio of carbamazepine peak area / bupropion peak area. In this embodiment, a 1 / x² weighted fitting is used to fit the calibration curve, and the final calibration curve equation is: Y = 0.143 × X + 0.022, with a coefficient of determination R² = 0.9979 (see...). Figure 4 (The calibration curve is shown in the left-middle figure). All calibration scatter points uniformly fit the fitted straight line, demonstrating excellent linear correlation, proving that the calibration curve meets the quantitative linearity requirements. The system automatically recalculates the concentrations at all calibration points, confirming the curve's validity. Based on the valid calibration curve, the quantitative analysis sub-agent substitutes the response ratios of each subject at different sampling times to batch calculate the measured carbamazepine concentrations of all plasma samples.

[0062] The system binds and summarizes subject IDs, sampling durations, and plasma concentration measurements, generating a concentration-time data table and automatically plotting pooled drug-time curves (mean ± standard deviation) for each subject. Figure 4 As shown in the right-middle figure, the pharmacokinetic curve clearly illustrates the in vivo metabolic pattern of carbamazepine concentration after administration, showing a rapid initial increase, peak, and then a gradual and stable decline. The error bars in the figure represent the standard deviation of test results from different subjects, visually demonstrating the repeatability of the test within the group. In this study, the concentrations of all quality control samples and unknown samples were within the calibration curve range, with no data exceeding the concentration limit. All quality control indicators met the preset standards, requiring no manual confirmation. If any abnormalities are detected, such as sample concentrations exceeding the calibration range or quality control deviations exceeding the threshold, the system will mark the abnormal entry and pause the process, pushing it to the manual confirmation module for user review and correction.

[0063] The report output module integrates the calibration curve equation, R² parameter, all sample concentration data, pharmacokinetic curve spectroscopy, and quality control judgment conclusions to generate a standardized pharmacokinetic analysis report for carbamazepine. The audit trail record fully includes the original MRM data path, calibration curve fitting parameters, all intermediate calculation values, quality control judgment basis, spectroscopy generation record, manual intervention record (no manual intervention in this case), and the final report storage path, realizing full-process verification and traceability of quantitative analysis.

[0064] Experiments show that this invention establishes a multi-agent collaborative quantitative analysis architecture for the entire process of LC-MS / MS quantitative detection of drugs in biological body fluids. The core innovations include:

[0065] 1. Innovative Intelligent Fitting of Weighted Calibration Curve: The quantitative agent can autonomously select weighting methods such as 1 / x² to complete the standard curve fitting based on the response characteristics of the matrix and compound to be tested, no longer limited to the ordinary least squares method. Taking carbamazepine as an example, the weighted fitting yields the calibration equation Y=0.143×X+0.022, with an R² as high as 0.9979, which significantly reduces the quantitative deviation at high and low concentrations, adapts to the need for accurate quantification over a wide range of blood drug concentrations, and solves the defect of large errors at low concentration points in conventional equal-weighted fitting.

[0066] The system features a fully automated closed-loop system for peak integration, concentration conversion, and pharmacokinetic plotting. It completes MRM peak extraction, peak area integration, analyte / internal standard response ratio calculation, concentration deduction from calibration curve, time-concentration data matching, mean and standard deviation statistics, and automatic plotting of pharmacokinetic curves in one stop. This replaces segmented manual software operation and avoids manual input errors and data matching problems caused by multiple software transfers.

[0067] Innovation in automatic anomaly classification and control: It sets up a dual-layer anomaly judgment logic for calibration exceeding limits and quality control exceeding standards. Normal samples automatically complete report archiving, while samples with concentrations exceeding the linear range or quality control results failing to meet standards are actively transferred to the manual review module. This achieves fully automated routine testing and intervention for difficult anomalies, balancing high throughput efficiency with rigorous testing compliance.

[0068] Innovation in end-to-end traceability auditing: The entire process automatically retains raw mass spectrometry data, fitting parameters, intermediate values, spectral files, anomaly handling records, and the final report path, meeting the GLP traceability requirements of the pharmaceutical testing industry. Compared with traditional manual record keeping, the traceability information is more complete and easier to retrieve.

[0069] This invention is applicable to the quantitative detection of various blood drugs, tissue fluids, and extract matrices by liquid chromatography-mass spectrometry. It is highly versatile and can be widely used in various quantitative detection scenarios such as clinical blood drug concentration monitoring, pharmacokinetic experiments, and natural product content determination.

[0070] The above embodiments are merely preferred implementations of the present invention, used to illustrate the complete working logic, operational procedures, and actual detection effects of the multi-agent mass spectrometry quantitative system of the present invention, and do not constitute a limitation on the scope of protection of the present invention. Those skilled in the art can make conventional adjustments, substitutions, and optimizations to the type of drug to be tested, mass spectrometry acquisition mode, curve weighting method, quality control threshold, and agent division of labor logic without departing from the overall technical concept of the present invention; all such improved solutions derived from the core architecture of the present invention fall within the protection scope of the claims of the present invention.

Claims

1. A multi-agent automated analysis system for qualitative and quantitative analysis by liquid chromatography-tandem mass spectrometry, comprising an input module, a central agent, sub-task agents, a report output module, and a manual verification module; wherein the input module is used to collect mass spectrometry analysis data, the central agent is used for task parsing and scheduling, the sub-task agents are used to complete mass spectrometry data analysis, the manual verification module is used for process anomaly control, and the report output module is used to generate and output analysis files; characterized in that: The input module is used to collect natural language analysis requirements and structured data to be analyzed, pre-scan the structured data to be analyzed, and generate a data list to constrain the task scheduling scope; the structured data includes at least one of sample information, instrument parameters, raw mass spectrometry data, and report templates; The central intelligent agent is equipped with a large language model, and generates a structured task plan based on the natural language analysis requirements and the data list. The structured task plan records the analysis type, data storage path, task execution order, and sub-task intelligent agents to be called. The subtask agent includes at least a qualitative analysis sub-agent and a quantitative analysis sub-agent; the qualitative analysis sub-agent is used to perform mass spectrometry library retrieval, candidate compound screening, multi-dimensional matching evidence fusion, and candidate credibility ranking; the quantitative analysis sub-agent is used to complete chromatographic peak integration, peak area response parameter calculation, calibration model fitting, and concentration conversion of the sample to be tested. The manual confirmation module is configured to terminate the automated process and receive manual review and modification operations in the following situations: the natural language requirement is semantically ambiguous, the path of the data to be tested is missing, the confidence of qualitative matching is lower than the preset confidence threshold, the compliance of the calibration model is not up to standard, the quality control test result exceeds the limit range, or the concentration to be tested exceeds the quantitative range. The report output module is used to collect qualitative results, quantitative results, and manual operation records to generate standardized test reports and audit trail record files.

2. The multi-agent automated analysis system according to claim 1, characterized in that, The subtask agent also includes a method configuration sub-agent and a method validation sub-agent; the method configuration sub-agent is configured to automatically generate liquid chromatography-tandem mass spectrometry instrument method parameters and sample injection sequence based on preset parameter templates and sample information; The methodology verification sub-agent is configured to automatically evaluate the methodological indicators of mass spectrometry detection based on preset judgment rules for calibration samples, quality control samples, and blank samples, and retain the verification process data and judgment basis.

3. The multi-agent automated analysis system according to claim 1, characterized in that, The qualitative analysis sub-agent uses multiple mass spectrometry matching algorithms to perform parallel matching of secondary mass spectrometry spectra, and summarizes the candidate compound results output by each algorithm to form a restricted candidate set; based on the multi-algorithm matching features of candidate compounds, it completes multi-dimensional evidence fusion and consensus ranking, and outputs only the qualitative results within the restricted candidate set.

4. The multi-agent automated analysis system according to claim 3, characterized in that, The qualitative analysis sub-agent comprehensively determines the confidence level of candidate compounds through the matching results of multiple algorithms, and only outputs qualitative results whose confidence level meets the preset confidence threshold; when there are low confidence scenarios such as compound identification ambiguity or substandard spectrum quality, it is pushed to the manual confirmation module for processing.

5. The multi-agent automated analysis system according to claim 1, characterized in that, The quantitative analysis sub-agent supports fitting calibration models with multiple weighting coefficients, verifies the compliance of the calibration model by back-calculating the concentration of the calibration sample, and marks the test sample with a concentration exceeding the quantitative range as abnormal and pushes it to the manual confirmation module.

6. The multi-agent automated analysis system according to claim 2, characterized in that, The methodology verification sub-agent can perform automated evaluation of multiple mass spectrometry detection methodology indicators, automatically record unqualified verification items, raw calculation data, threshold judgment basis and sample batch information, and realize full-chain traceability of the verification process.

7. A multi-agent automated analytical method for qualitative and quantitative analysis using liquid chromatography-tandem mass spectrometry, characterized in that, The multi-agent automated analysis system according to any one of claims 1 to 6 is used to perform the following steps: S1. The input module collects the user's input natural language analysis requirements and the data to be analyzed and detected, pre-scans the detection data directory, and generates a data list within the scope of the task scheduling. S2. The central intelligent agent analyzes natural language analysis requirements through a built-in large language model and generates standardized structured task plans by combining data lists. S3. The central agent schedules the corresponding sub-task agents according to the structured task plan, and selectively executes at least one analytical task among instrument method configuration, sample injection sequence generation, mass spectrometry qualitative analysis, mass spectrometry quantitative analysis, and methodology verification. S4. The system monitors and analyzes the process operation status in real time to determine whether preset abnormal conditions are triggered. If an exception is triggered, the automated process will be paused and transferred to manual review and correction. Once the correction is completed, the task will resume execution. If no abnormalities are found, continue to complete the data analysis and calculations; S5, the report output module summarizes the qualitative results, quantitative results, methodology verification data and manual operation records of the entire process, generates a standardized analysis report, and simultaneously retains the traceability records of the entire chain.

8. The multi-agent automated analysis method according to claim 7, characterized in that, In step S3, the qualitative analysis process includes parallel matching of spectra using multiple algorithms, fusion of evidence for candidate compounds and consensus ranking, and confidence screening; The quantitative analysis process includes fitting a multi-weighted calibration model, converting sample concentrations, and marking out-of-range anomalies; the method validation process includes automated evaluation of multi-dimensional methodological indicators and archiving and retaining validation data.

9. The multi-agent automated analysis method according to claim 7, characterized in that, In step S4, all manually reviewed and corrected operational data and corrections are automatically included in the full-chain traceability record, ensuring that the analysis process is traceable and reproducible.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-agent automated analysis method according to any one of claims 7 to 9.

Images