An autonomous multi-agent automatic analysis system and method for multi-industry liquid chromatography-tandem mass spectrometry quantitative detection scene

Abstract

The application relates to an autonomous multi-agent automatic analysis system and method for multi-industry liquid chromatography-tandem mass spectrometry quantitative detection scenes, belonging to the technical field of instrumental analysis and artificial intelligence. The system comprises a data access module, an autonomous multi-agent module, a quality determination module, a compliance report generation module and a manual review module. The autonomous multi-agent module comprises a central agent, a detection scene recognition agent, a process configuration agent and a quantitative analysis agent, and is locally deployed and does not transmit data. The application solves the problems of low data flow efficiency, poor scene adaptability and insufficient traceability in the prior art by cooperating with multiple agents to analyze natural language requirements, automatically recognizing detection scenes, adapting corresponding quality control rules and report templates, completing full-automatic quantitative analysis, quality determination, compliance report generation and full-process audit tracking, and is suitable for drug, clinical, customs and food multi-industry detection scenes, and greatly improves detection automation, standardization and traceability.

Description

Technical Field

[0001] This invention relates to the fields of instrumental analysis and artificial intelligence, and in particular to an autonomous multi-agent automated analysis system and method for quantitative detection scenarios of liquid chromatography-tandem mass spectrometry in multiple industries. Background Technology

[0002] Liquid chromatography-tandem mass spectrometry (LC-MS / MS) is widely used for the detection of small molecule targets in pharmaceutical production, clinical testing, customs inspection, and food inspection due to its high sensitivity and strong quantitative capabilities. Since the test results are the core basis for intermediate control, finished product release, clinical diagnosis, and law enforcement supervision, the testing process must be highly accurate, standardized, and traceable while pursuing high efficiency to ensure the authority and reliability of the conclusions.

[0003] Currently, the above-mentioned detection process typically relies on a combination of tools, including liquid chromatography-tandem mass spectrometry workstations, laboratory information management systems, online automated detection equipment, spreadsheet software, and report templates (Henderson, CM, Shulman, NJ, MacLean, B., MacCoss, MJ, & Hoofnagle, AN (2018). SkylinePerforms as Well as Vendor Software in the Quantitative Analysis of Serum 25-Hydroxy Vitamin D and Vitamin D Binding Globulin). Clinical Chemistry , 64(2), 408–410). These tools have improved detection efficiency and data management in their respective stages and have promoted the standardization of detection processes to some extent. However, the above models still have the following problems:

[0004] (1) Insufficient processing timeliness in rapid release scenarios. Instrument workstations and laboratory information management systems are usually independent of each other, with raw data, sample information, quality control information, and report fields stored in different systems, requiring manual data transfer, verification, and summarization. The above scenarios usually require quantitative calculation, quality judgment, and report generation to be completed in a short period of time, but the existing process still relies on manual switching and data verification between multiple systems. When the sample batch is large or the testing tasks are concentrated, data processing and report generation can easily become bottlenecks restricting rapid release.

[0005] (2) Insufficient standardization adaptability for specific testing tasks. In specific testing tasks, different sample types, testing items, and testing purposes correspond to different quality control rules, calculation methods, judgment thresholds, report fields, and review processes. Existing tools usually require manual selection of applicable standards, application of report templates, and completion of conclusions, making it difficult to automatically match the corresponding rules according to specific testing tasks, resulting in insufficient consistency in result judgment and report format.

[0006] (3) Insufficient report generation and audit trail capabilities. Existing tools can usually only save the processing results of a single step, making it difficult to unify and link the original input files, instrument acquisition information, sample metadata, quantitative calculation process, quality judgment basis, manual confirmation records and final report, resulting in high costs for result verification, regulatory inspection and cross-institutional traceability.

[0007] In summary, existing tools for quantitative detection using liquid chromatography-tandem mass spectrometry still fall short of meeting actual detection needs in terms of overall automation, standardization, and traceability. There is an urgent need for an automated analysis system and method specifically developed for quantitative detection using liquid chromatography-tandem mass spectrometry. Summary of the Invention

[0008] The purpose of this invention is to address the shortcomings of existing technologies in terms of automation, standardization, and traceability by providing an autonomous multi-agent automated analysis system and method for quantitative detection in multiple industries using liquid chromatography-tandem mass spectrometry. This will improve data processing efficiency, consistency of quality judgment, standardization of report generation, and traceability of the detection process in scenarios such as pharmaceutical production, clinical testing, customs inspection, and food inspection.

[0009] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0010] In a first aspect, the present invention provides an autonomous multi-agent automatic analysis system for liquid chromatography-tandem mass spectrometry detection scenarios in multiple industries, including a data access module, an autonomous multi-agent module, a quality judgment module, a compliance report generation module, and a manual review module;

[0011] The data access module is used to obtain the data to be analyzed from the liquid chromatography-tandem mass spectrometry instrument workstation, laboratory information management system, online or near-line automated detection equipment, database or local file directory. The data to be analyzed includes one or more of the following: sample batch information, injection sequence, instrument acquisition data, calibration sample data, quality control sample data and unknown sample data.

[0012] The autonomous multi-agent module is deployed on a local server, enterprise intranet, or laboratory intranet environment. It is used to parse natural language analysis requirements, identify detection scenarios, specific detection tasks, and target outputs without uploading detection data to an external server, and autonomously generate or adjust structured quantitative analysis processes. The autonomous multi-agent module includes a central agent, a detection scenario identification agent, a process configuration agent, and a quantitative analysis agent. The central agent generates structured task plans and coordinates task execution. The detection scenario identification agent identifies detection scenarios based on user needs, sample type, and detection items, including one or more of pharmaceutical production, clinical testing, customs inspection, and food inspection. The process configuration agent determines specific detection tasks based on the detection scenario, sample type, detection items, and target output, autonomously configures the quantitative analysis process, quality judgment process, and report fields, and selects or adjusts corresponding quality control rules, judgment thresholds, and report templates. The quantitative analysis agent performs peak area reading, internal standard correction, calibration curve fitting, and sample concentration back-calculation.

[0013] The quality judgment module is used to judge the quality of calibration samples, quality control samples and unknown samples according to the quality control rules and judgment thresholds determined by the process configuration agent, and generate judgment results of qualified, unqualified, pending review or out of range.

[0014] The compliance report generation module is used to generate standardized test reports for corresponding test scenarios according to the report template determined by the process configuration agent, and to generate audit trail records that include data sources, calculation processes, judgment criteria, manual review records and report version information;

[0015] The manual review module is used to pause the automatic process and receive user confirmation or correction when data is missing, quality control rules or report templates are unclear, quality control fails, sample concentration back-calculation results exceed the calibration range, limit judgment is abnormal, or report fields are missing.

[0016] Secondly, this invention provides an autonomous multi-agent automated analysis method for multi-industry liquid chromatography-tandem mass spectrometry detection scenarios. Based on the aforementioned system, this method takes user-input natural language analysis requirements and data to be analyzed as input. The autonomous multi-agent module identifies the detection scenario, specific detection task, and target output in the local environment. It generates a structured task plan according to different detection scenarios and specific detection tasks, and calls the corresponding modules to complete data access, autonomously configure quantitative analysis processes and applicable rules, perform quantitative calculations, quality judgment, and generate compliance reports. Finally, it generates standardized detection reports and audit trail records for the corresponding detection scenarios, adapting to the automated compliance detection needs of multiple industries.

[0017] The technical problem solved by this invention:

[0018] This invention aims to address the technical challenges of insufficient automation, standardization adaptability, and traceability in existing liquid chromatography-tandem mass spectrometry (LC-MS / MS) quantitative detection processes across multiple industries, including pharmaceutical manufacturing, clinical testing, customs inspection, and food inspection. Specifically, in existing multi-system collaborative processes, detection data, sample information, and report information are typically scattered across different tools, resulting in low data flow efficiency and a heavy burden of manual verification and summarization. Furthermore, different sample types, detection items, and detection purposes have varying standard rules, quality judgment methods, and reporting requirements, making it difficult for existing processes to automatically adapt to specific detection tasks. In addition, the lack of continuous correlation between detection data, calculation processes, judgment criteria, manual review records, and final reports leads to high costs for result verification and regulatory traceability. By addressing these issues, this invention makes the LC-MS / MS quantitative detection process more suitable for the rapid release and standardized testing needs of multiple industries.

[0019] This invention innovatively introduces an autonomous multi-agent scheduling mechanism into the entire process of quantitative detection using liquid chromatography-tandem mass spectrometry. Compared with existing technologies, this invention has the following outstanding advantages:

[0020] 1. This invention uses multi-agent collaborative scheduling to unify and coordinate all business processes, including data access, quantitative calculation, quality judgment, and report generation. This replaces the cumbersome operations of manual cross-system switching, manual verification, and data aggregation in traditional solutions, reducing manual workload and effectively solving data processing bottlenecks in batch testing and rapid release scenarios, thereby improving overall testing efficiency.

[0021] 2. To address the diverse testing needs across various industries, including pharmaceutical manufacturing, clinical testing, customs inspection, and food inspection, this invention can automatically match corresponding quality control rules, judgment thresholds, and standardized report templates based on specific sample types, testing items, and testing objectives. This eliminates the need for manual screening of suitable standards and templates, preventing issues such as inconsistent judgment standards and chaotic report formats caused by human intervention, and improving the consistency of result judgment and report output for different testing tasks.

[0022] 3. This invention can link and bind the original test data, instrument acquisition information, quantitative calculation process, quality judgment basis, manual review records and final test report in the whole chain and archive them in a unified manner to form a complete audit trail record, effectively reducing the time and manpower costs of result review, regulatory inspection and cross-institutional traceability, and fully adapting to the compliance and regulatory requirements of various industries.

[0023] 4. This invention adopts an intranet localized deployment approach. All test data can be parsed, calculated, and stored on local servers, enterprise intranets, or laboratory intranets without uploading to external servers. This avoids the risk of data leakage from the source and meets the data localization and security processing requirements of highly compliant industries such as pharmaceuticals, clinical practice, customs, and food. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the structural composition of the autonomous multi-agent automatic analysis system according to an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the workflow of the process configuration intelligent agent in an embodiment of the present invention.

[0026] Figure 3 This is a diagram showing the batch release test results of atorvastatin in pharmaceutical production during Example 1 of the present invention.

[0027] Figure 4 This is a graph showing the accuracy results of the development of a quantitative method for carbamazepine in clinical therapeutic drug monitoring in Example 2 of the present invention.

[0028] Figure 5 This is a diagram showing the detection results of malachite green and leuco malachite green in customs inspection during Embodiment 3 of the present invention.

[0029] Figure 6 This is a graph showing the results of chloramphenicol residue detection in food inspection during Example 4 of the present invention. Detailed Implementation

[0030] 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.

[0031] In this embodiment, the autonomous multi-agent automatic analysis system for multi-industry liquid chromatography-tandem mass spectrometry detection scenarios includes a data access module, an autonomous multi-agent module, a quality assessment module, a compliance report generation module, and a manual review module. The data access module, autonomous multi-agent module, quality assessment module, compliance report generation module, and manual review module are interconnected to achieve data access, detection task identification, quantitative process configuration, quantitative analysis, quality assessment, report generation, and manual review. The structural composition of this embodiment is as follows. Figure 1 As shown.

[0032] 1. Data Access Module

[0033] The data access module supports multi-source data adaptation and can simultaneously acquire data to be analyzed from liquid chromatography-tandem mass spectrometry instrument workstations, laboratory information management systems, online or near-line automated detection equipment, databases, or local file directories. The data to be analyzed includes one or more of the following: sample batch information, injection sequence, instrument acquisition data, calibration sample data, quality control sample data, and unknown sample data. After acquiring the data to be analyzed, the data access module identifies the detection items, data sources, file formats, sample types, sample numbers, injection sequences, analyte information, internal standard information, and acquisition time, and maps data fields from different sources to a unified data structure. This unified data structure includes at least a sample identification field, a batch field, an injection field, an analyte field, an internal standard field, and a quality control status field. The data access module also performs integrity checks on the data to be analyzed. When it detects missing sample numbers, discontinuous injection sequences, missing calibration samples, missing quality control samples, inconsistent units, or unidentifiable fields, it generates a data anomaly marker and transmits the anomaly information to the manual review module.

[0034] 2. Autonomous Multi-Agent Module

[0035] The autonomous multi-agent module is deployed on a local server, enterprise intranet, or laboratory intranet environment. It is used to parse natural language analysis requirements, identify detection scenarios, specific detection tasks, and target outputs without uploading detection data to an external server, and autonomously generate or adjust structured quantitative analysis processes. The autonomous multi-agent module includes a central agent, a detection scenario identification agent, a process configuration agent, and a quantitative analysis agent.

[0036] 2.1 After receiving the user's input of natural language analysis requirements, the central agent, in conjunction with the data list generated by the data access module, determines the tasks to be executed and generates a structured task plan. The structured task plan includes the detection scenario, specific detection tasks, input data paths, field mapping relationships, target objects to be tested, calling modules, execution order, quantitative analysis process, quality judgment process, task-level rule configuration, task-level report templates, output directory, and exception handling strategies. The central agent is also used to receive the returned results after each module's execution and, based on the task plan, determine whether to proceed to the next step, whether to trigger manual review, or whether to re-execute a certain calculation step.

[0037] 2.2 The detection scenario recognition agent is used to identify detection scenarios and specific detection tasks based on user needs, sample types, detection items, and reporting objectives. The detection scenarios include one or more of the following: pharmaceutical production, clinical testing, customs inspection, and food inspection. The specific detection tasks include one or more of the following: raw material testing, intermediate control, finished product release, stability studies, quantification of clinical targets, import / export sample testing, contaminant testing, additive testing, and residue testing. When the user's requirements clearly specify the detection scenario, the detection scenario recognition agent directly determines the detection scenario based on the user input; when the user's requirements do not specify the detection scenario, the detection scenario recognition agent generates candidate detection scenarios based on sample type, detection items, and user-input keywords, and submits them to the manual review module for confirmation through the central agent.

[0038] 2.3 The process configuration agent is used to autonomously configure quantitative analysis processes, quality judgment processes, and report fields based on the detection scenario, specific detection task, sample type, analyte, and target output. It also selects or adjusts corresponding quality control rules, judgment thresholds, and report templates. Specifically, the process configuration agent calls rules and templates applicable to the current task from the local rule base and template library. Without changing the original standard rule content, it generates task-level rule configurations and task-level report templates based on the data type, sample grouping, calibration range, quality control level, report purpose, and audit requirements of the current task. The task-level rule configuration includes one or more of the following: calibration curve fitting method, weighting method, calibration sample acceptance standard, quality control sample acceptance standard, unknown sample judgment threshold, out-of-range handling method, review triggering condition, and audit node. The task-level report template includes one or more of the following: report title, chapter structure, table fields, graph fields, conclusion fields, signature fields, and audit trail fields. The local rule base can include the standard number, standard name, applicable analyte, applicable sample type, quality control acceptance standard, calibration curve requirements, judgment threshold, conclusion expression method, and audit node corresponding to different detection tasks. The template library can include report titles, chapter structures, table fields, graph fields, conclusion fields, and audit trail fields corresponding to different detection scenarios or specific detection tasks. When multiple candidate rules exist, applicable rules are missing, task-level rule configurations conflict, template fields are incomplete, or rule versions are inconsistent, the process configuration agent generates process configuration exception information and passes the candidate rules, task-level rule configurations, template fields, discrepancies, and recommended configurations to the manual review module. The workflow of the process configuration agent is as follows: Figure 2 As shown.

[0039] 2.4 The quantitative analysis agent is used to configure and execute quantitative calculations based on the task-level rules generated by the structured task plan and process configuration agent. The quantitative analysis agent reads the peak area of ​​the analyte, the peak area of ​​the internal standard, the theoretical concentration of the calibration sample, the theoretical concentration of the quality control sample, and the response value of the unknown sample. It calculates the response ratio of the analyte to the internal standard and fits a calibration curve based on the theoretical concentration and response ratio of the calibration sample. The calibration curve can be linearly fitted or weighted linearly fitted, and the weighting method can include one or more of unweighted, 1 / x, 1 / x², 1 / y, or 1 / y². The quantitative analysis agent performs concentration back-calculation on the calibration sample, quality control sample, and unknown sample based on the fitted calibration curve and outputs the calibration curve parameters, back-calculated concentration, deviation, response ratio, sample concentration, and out-of-range flag. When there are missing peak areas, abnormal internal standard responses, calibration curve fitting failures, sample concentration back-calculation results exceeding the calibration range, or calculation results not meeting preset rules, the quantitative analysis agent generates a calculation anomaly flag and transmits the anomaly information to the central agent and the manual review module.

[0040] 3. Quality Judgment Module

[0041] The quality assessment module is used to assess the quality of calibration samples, quality control samples, and unknown samples based on task-level rule configurations generated by the process configuration agent. Specifically, the quality assessment module determines the number of calibration samples, calibration range, back-calculation deviation, and curve validity based on calibration curve requirements; it determines the accuracy, precision, and batch acceptance of quality control samples based on quality control rules; and it determines whether unknown samples are qualified, unqualified, out of range, or require re-verification based on limit standards or judgment thresholds. The judgment results generated by the quality assessment module include qualified, unqualified, requiring re-verification, or out of range, and record the rule basis, calculated value, threshold, and sample number corresponding to each judgment result.

[0042] 4. Compliance Report Generation Module

[0043] The compliance report generation module generates standardized test reports for corresponding testing scenarios based on task-level report templates generated by the intelligent agent according to the process configuration. The module automatically fills in sample information, test items, instrument information, calibration curve information, quality control results, results for unknown samples, standard basis, quality judgment conclusions, and manual review records based on the task-level report template. The compliance report generation module also generates audit trail records, which include data source, original file path, data access time, field mapping results, structured task plan, task-level rule configuration, rule version, task-level report template, template version, calculation process, quality judgment basis, manual review records, report generation time, and report version information. After the report is generated, the system associates and saves the standardized test report and audit trail records.

[0044] 5. Manual review module

[0045] The manual review module is used to pause the automated process when uncertainties or abnormal results occur during automatic analysis, and to receive user confirmation or correction. Uncertainties or abnormal results include one or more of the following: missing data, unidentifiable fields, unclear rule configuration or report template, quality control failure, calibration curve not meeting requirements, sample concentration back-calculation results exceeding the calibration range, abnormal limit judgment, missing report fields, or incomplete report conclusions. The manual review module displays the anomaly type, associated samples, raw data, candidate rule configurations, candidate report templates, calculation results, and system suggestions to the user. Users can choose to confirm, correct, exclude the anomaly and record the reasons, recalculate, or mark for retesting. The manual review module writes the user's actions, review time, review content, and review conclusion into the audit trail record and returns the review results to the central agent so that the system can continue executing subsequent processes.

[0046] The autonomous multi-agent automatic analysis method of the present invention, based on the above system, includes the following steps:

[0047] S1. Multi-source data access and standardized processing: The data access module collects data to be analyzed from various sources, completes field identification, unified structured mapping and integrity verification, marks abnormal data, and generates a standardized data list;

[0048] S2. Requirements Analysis and Scene Task Identification: User inputs natural language to analyze requirements, the central intelligent agent breaks down tasks by combining standardized data lists, and the scene identification intelligent agent accurately identifies the detection scene and specific detection task to generate a preliminary task plan.

[0049] S3. Autonomous configuration of processes and rules: The process configuration agent matches local rule base and template base resources to generate exclusive quality control rules, judgment thresholds and report templates adapted to the current task, forming a complete task-level configuration;

[0050] S4. Fully Automated Quantitative Calculation: The quantitative analysis agent, based on its proprietary configuration, completes peak area reading, internal standard correction, calibration curve fitting, full sample concentration recalculation, and simultaneously self-checks for calculation anomalies.

[0051] S5. Layered Quality Assessment: The quality assessment module, based on proprietary rules, completes multi-level quality assessments for calibration, quality control, and unknown samples, and outputs compliance assessment results and supporting evidence.

[0052] S6. Manual Review of Abnormalities: If any abnormal or uncertain point occurs during the testing process, manual review will be automatically triggered. After the user completes the confirmation and correction, the review information will be recorded and the process will be restarted.

[0053] S7. Compliance Report and Traceability Record Generation: The compliance report generation module outputs standardized industry testing reports and simultaneously generates full-dimensional audit trace records, which are then linked and archived to complete the testing process.

[0054] The following examples fully demonstrate the practical application effects of this invention in four core scenarios: pharmaceutical production, clinical testing, customs inspection, and food inspection. The entire process involves no manual rule configuration or template application, and the analysis is completed autonomously.

[0055] Based on the above system structure and operation mode, the following embodiments demonstrate the practical application effects in four core scenarios: drug production, clinical testing, customs inspection, and food inspection. These are only used to illustrate the present invention and are not intended to limit the scope of protection of the present invention.

[0056] Example 1: Batch release detection of atorvastatin by liquid chromatography-tandem mass spectrometry in pharmaceutical production.

[0057] The user inputs a natural language request: "Perform quantitative detection of atorvastatin in a certain batch of atorvastatin calcium tablets using liquid chromatography-tandem mass spectrometry and generate a batch release report." The system's data access module reads the batch information, injection sequence, instrument acquisition data, calibration sample data, quality control sample data, and unknown sample data of the atorvastatin calcium tablet sample. It then performs field identification, unified structured mapping, and data integrity verification on the multi-source heterogeneous data to form a standardized structured data list. The detection scenario identification agent in the autonomous multi-agent module automatically identifies the current detection scenario as a pharmaceutical production scenario and the detection task as batch release testing of finished pharmaceutical products based on the user's requirements, sample dosage form, and detection purpose. The central agent generates a complete structured task plan based on the identification results. The process configuration agent selects applicable calibration curve requirements, quality control acceptance standards, release judgment thresholds, and batch release report templates from the local rule base and template library based on the detection purpose, sample type, analyte atorvastatin, and target output, and generates task-level rule configurations and task-level report templates.

[0058] The quantitative analysis agent reads the peak area of ​​atorvastatin and its corresponding internal standard, calculates the response ratio, fits a calibration curve that meets the drug release standards, and performs concentration back-calculation for quality control samples and unknown samples. The quality judgment module, based on task-level rule configuration, judges whether the calibration curve, quality control samples, and unknown samples meet the release requirements, and generates a judgment result of "qualified," "unqualified," "pending review," or "out of range." The system-generated results are as follows: Figure 3 As shown. After the review is completed, the compliance report generation module generates an atorvastatin batch release test report and audit trail record, realizing full traceability, reproducibility, and supervision of the drug release test process.

[0059] Example 2: Development of a quantitative method for carbamazepine by liquid chromatography-tandem mass spectrometry in clinical therapeutic drug monitoring.

[0060] The user inputs a natural language request: "Establish a liquid chromatography-tandem mass spectrometry quantitative analysis method for carbamazepine in plasma samples for clinical therapeutic drug monitoring, and generate a methodology development report."

[0061] The system data access module reads carbamazepine standard information, internal standard information, instrument acquisition data, calibration sample data, quality control sample data, and other methodological validation data. The detection scenario identification agent recognizes the task as a small molecule drug quantitative methodological development task. The workflow configuration agent selects applicable calibration curve requirements, quality control acceptance criteria, methodological validation items, and methodological development report templates from the local rule base and template base based on the detection purpose, analyte, sample type, and target output, and generates task-level rule configurations and task-level report templates. The quantitative analysis agent reads the peak areas of carbamazepine and its internal standard, calculates the response ratio, and fits calibration curves based on calibration samples at different concentration levels, performing concentration back-calculations for calibration samples, quality control samples, and validation samples. The quality judgment module judges items such as linearity, accuracy, precision, selectivity, residue, matrix effect, dilution accuracy, and stability according to the task-level rule configuration, and records non-compliant items, original calculated values, and judgment criteria. The accuracy results generated by the system are as follows: Figure 4 As shown.

[0062] When the calibration curve fails to meet requirements, quality control samples exceed the acceptable range, sample concentration back-calculation results exceed the calibration range, or there are anomalies in the methodology validation items, the system prompts the user through the manual review module to confirm whether to correct the rule configuration, eliminate anomalies and record the reasons, recalculate, or mark for retesting. After the review is completed, the compliance report generation module generates a carbamazepine quantitative methodology development report based on the task-level report template, and forms an audit trail record that includes data sources, task-level rule configurations, task-level report templates, calculation processes, judgment criteria, manual review records, and report version information.

[0063] Example 3: Quantitative detection of malachite green and leucomalle green by liquid chromatography-tandem mass spectrometry in customs inspection.

[0064] The user inputs a natural language request: "Quantitative detection of malachite green and leucomalle green in imported aquatic products using liquid chromatography-tandem mass spectrometry, and generation of a customs inspection report." The system data access module reads imported aquatic product sample information, instrument acquisition data, injection sequence, calibration sample data, quality control sample data, and unknown sample data. The detection scenario identification agent recognizes this task as a prohibited substance regulatory inspection task under a customs inspection scenario. The process configuration agent selects applicable limit standards, quality control rules, judgment thresholds, review processes, and customs inspection report templates from the local rule base and template library based on the sample type, detection items, target analytes malachite green and leucomalle green, and target output, and generates task-level rule configurations and task-level report templates. The quantitative analysis agent completes peak area reading, internal standard correction, calibration curve fitting, and sample concentration back-calculation. The quality judgment module determines whether the unknown sample contains the target analyte and whether it exceeds the judgment threshold based on the task-level rule configuration, and generates a judgment result of qualified, unqualified, pending review, or out of range. The system-generated results are as follows: Figure 5 As shown. After the review is completed, the compliance report generation module generates a customs inspection report and an audit trail record.

[0065] Example 4: Detection of chloramphenicol residues by liquid chromatography-tandem mass spectrometry in food inspection.

[0066] The user inputs a natural language request: "Quantitatively detect chloramphenicol residues in milk samples using liquid chromatography-tandem mass spectrometry and generate a food inspection report." The system data access module reads milk sample information, instrument acquisition data, calibration sample data, quality control sample data, and unknown sample data. The detection scenario identification agent recognizes this task as a veterinary drug residue quantitative detection task within a food inspection scenario. The process configuration agent selects applicable limit standards, quality control acceptance standards, judgment thresholds, and food inspection report templates from the local rule base and template base based on the food category, detection item, target analyte chloramphenicol, and target output, and generates task-level rule configurations and task-level report templates. The quantitative analysis agent reads the peak areas of chloramphenicol and its internal standard, calculates the response ratio, fits the calibration curve, and performs back-calculation of unknown sample concentrations. The quality judgment module determines whether the chloramphenicol residues in the milk sample meet the limit requirements based on the task-level rule configuration and generates a judgment result of qualified, unqualified, pending review, or out of range. The system-generated results are as follows: Figure 6 As shown. After the review is completed, the compliance report generation module generates a food inspection report and audit trail record.

[0067] Validated through multiple batches of samples, compared to traditional manual analysis processes, this invention, through multi-agent autonomous scheduling throughout the entire process, can reduce the time spent on data analysis, quality control judgment, and report generation for liquid chromatography-mass spectrometry (LC-MS) by more than 60%, and reduce manual operations by more than 90%. The automatic adaptation of multi-scenario testing rules achieves 100% accuracy, completely eliminating judgment biases and inconsistent report formats caused by manual standard adaptation. Simultaneously, this invention enables data association and archiving throughout the entire testing process, improving audit traceability efficiency by more than 80%. The localized offline deployment mode completely eliminates the risk of data leakage. Testing has shown that it can stably adapt to batch testing and compliance regulatory scenarios in multiple industries, including pharmaceuticals, clinical settings, customs, and food, fully validating its technical stability, versatility, and engineering practicality.

[0068] This invention introduces autonomous multi-agent collaborative scheduling technology into the field of liquid chromatography-tandem mass spectrometry (LC-MS / MS) quantitative detection, overcoming industry pain points such as fragmented multi-system processes, high reliance on manual intervention, poor scenario adaptability, and missing traceability chains in traditional detection processes. Through a layered multi-agent architecture, it achieves fully automated, unmanned operation throughout the entire process, including natural language requirement parsing, autonomous detection scenario identification, adaptive task rule configuration, fully automated quantitative calculation, quality standardization judgment, and intelligent output of compliance reports. This eliminates the traditional model of manual cross-system operation, manual standard adaptation, and manual report compilation. Simultaneously, this invention constructs a full-chain traceability system covering data access, calculation process, judgment criteria, review records, and final report. Combined with a fully localized intranet deployment mode, it simultaneously solves four major technical challenges in LC-MS / MS detection within a single system: automation efficiency, multi-industry standardization adaptability, compliance traceability, and data security. It is widely applicable to quantitative detection scenarios in pharmaceuticals, clinical settings, customs, and food, demonstrating strong versatility and engineering practicality.

[0069] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, partial improvements, and scenario expansions made by those skilled in the art based on the structure, method, and application logic of the present invention without departing from the core principles, technical ideas, and scope of disclosure of the present invention should fall within the scope of protection of the present invention. Technical features, implementation methods, and industry-adapted scenarios not described in detail in the present invention can all be implemented by combining existing technologies with the core technical solutions of the present invention, without affecting the completeness and practicality of the technical solutions of the present invention.

Claims

1. An autonomous multi-agent automated analysis system for quantitative detection scenarios using liquid chromatography-tandem mass spectrometry across multiple industries, characterized in that: It includes a data access module, an autonomous multi-agent module, a quality judgment module, a compliance report generation module, and a manual review module, wherein the data access module, autonomous multi-agent module, quality judgment module, compliance report generation module, and manual review module are connected in pairs for communication. The data access module is used to obtain the data to be analyzed from a liquid chromatography-tandem mass spectrometry instrument workstation, a laboratory information management system, online or near-line automated detection equipment, a database or a local file directory. The data to be analyzed includes one or more of the following: sample batch information, injection sequence, instrument acquisition data, calibration sample data, quality control sample data and unknown sample data. The autonomous multi-agent module is deployed locally on a local server, enterprise intranet, or laboratory intranet environment. It is used to parse natural language analysis requirements, identify detection scenarios, specific detection tasks, and target outputs, and autonomously generate or adjust structured quantitative analysis processes without uploading detection data to an external server. The autonomous multi-agent module includes a central agent, a detection scenario identification agent, a process configuration agent, and a quantitative analysis agent. The central intelligent agent is used to generate structured task plans and coordinate the execution of tasks in each module, receive the execution results returned by each module, and determine whether the process is progressing, manual review is triggered, or steps are re-executed. The detection scenario identification intelligent agent is used to identify the detection scenario based on user needs, sample type and detection items. The detection scenario includes one or more of the following: pharmaceutical production, clinical testing, customs inspection and food inspection. The process configuration agent is used to determine specific testing tasks based on the testing scenario, sample type, testing items and target output, and autonomously configure quantitative analysis process, quality judgment process and report fields, and select and adjust matching quality control rules, judgment thresholds and report templates. The quantitative analysis agent is used to perform peak area reading, internal standard correction, calibration curve fitting, and sample concentration back calculation operations. The quality judgment module is used to judge the quality of calibration samples, quality control samples and unknown samples according to the quality control rules and judgment thresholds determined by the process configuration agent, and generate judgment results of qualified, unqualified, pending review or out of range. The compliance report generation module is used to generate standardized test reports for corresponding test scenarios according to the matched report template, and to generate audit trail records that include data sources, calculation processes, judgment criteria, manual review records, and report version information; The manual review module is used to pause the automatic process and receive confirmation or correction from the user when there are data anomalies, unclear rule templates, quality control failures, abnormal concentration results, or missing report fields.

2. The system according to claim 1, characterized in that, After acquiring the data to be analyzed, the data access module identifies the detection items, data sources, file formats, sample types, sample numbers, injection sequences, analyte information, internal standard information, and collection time. It then maps data fields from different sources into a unified data structure. The unified data structure includes at least a sample identification field, a batch field, an injection field, an analyte field, an internal standard field, and a quality control status field. The data access module is also used to verify the integrity of the data to be analyzed. When it detects missing sample numbers, discontinuous injection sequences, missing calibration samples, missing quality control samples, inconsistent units, or unidentifiable fields, it generates a data anomaly marker and transmits it to the manual review module.

3. The system according to claim 1, characterized in that, The structured task plan generated by the central intelligent agent includes the detection scenario, specific detection task, input data path, field mapping relationship, target object to be tested, calling module, execution order, quantitative analysis process, quality judgment process, task-level rule configuration, task-level report template, output directory and exception handling strategy.

4. The system according to claim 1, characterized in that, The specific detection tasks identified by the intelligent agent in the detection scenario include one or more of the following: raw material detection, intermediate control, finished product release, stability study, quantification of clinical targets, import and export sample detection, contaminant detection, additive detection, and residue detection. The recognition logic of the detection scene recognition intelligent agent is as follows: when the user clearly defines the detection scene, the corresponding scene is directly determined. When the user's requirements do not specify the testing scenario, candidate testing scenarios are generated based on sample type, testing items, and keywords, and submitted to the manual review module for confirmation.

5. The system according to claim 1, characterized in that, The process configuration agent has a built-in local rule library and template library. After calling the appropriate rules and templates, it generates task-level rule configurations and task-level report templates based on the current task data type, sample grouping, calibration range, quality control level, report purpose and review requirements without modifying the original standard rules. The task-level rule configuration includes one or more of the following: calibration curve fitting method, weighting method, calibration sample acceptance standard, quality control sample acceptance standard, unknown sample judgment threshold, out-of-range handling method, review triggering condition, and audit node. The task-level report template includes one or more of the following: report title, chapter structure, table fields, graph fields, conclusion fields, approval fields, and audit trail fields. When the process configuration agent detects multiple candidate rules, missing rules, configuration conflicts, incomplete template fields, or inconsistent versions, it generates process configuration exception information and pushes it to the manual review module.

6. The system according to claim 1, characterized in that, The calibration curve fitting method of the quantitative analysis agent includes linear fitting and weighted linear fitting, and the weighting method includes one or more of the following: unweighted, 1 / x, 1 / x², 1 / y, and 1 / y². The quantitative analysis agent fits the calibration curve based on the response ratio of the analyte to the internal standard and the theoretical concentration of the calibration sample, completes the back-calculation of the concentration of various samples, and outputs the curve parameters, back-calculated concentration, deviation, response ratio, and out-of-range marker. When peak area is missing, internal standard response is abnormal, curve fitting fails, concentration exceeds calibration range, or calculation is abnormal, a calculation abnormality marker is generated and pushed to the central intelligent agent and manual review module.

7. The system according to claim 1, characterized in that, The quality assessment module is specifically used for: determining the number of calibration samples, calibration range, back-calculation deviation, and curve validity according to calibration curve requirements; determining the accuracy, precision, and batch acceptance of quality control samples according to quality control rules; and determining the status of unknown samples according to limit standards and assessment thresholds. Each assessment result is associated with the corresponding rule basis, calculated value, threshold, and sample number.

8. The system according to claim 1, characterized in that, The compliance report generation module automatically fills in sample information, testing items, instrument information, calibration curve information, quality control results, results for unknown samples, standard basis, judgment conclusions, and manual review records; The audit trail record includes data source, original file path, data access time, field mapping result, structured task plan, rule configuration and version, template configuration and version, calculation process, judgment basis, review record, report generation time and report version information, and the report is saved in association with the audit trail record.

9. The system according to claim 1, characterized in that, The abnormal triggering scenarios of the manual review module include missing data, unidentifiable fields, unclear rules or templates, quality control failure, substandard calibration curves, concentration back-calculation results exceeding the range, abnormal limit judgment, missing report fields or incomplete conclusions; The manual review module displays the anomaly type, associated samples, raw data, candidate configurations, and system suggestions. It supports users in confirming, correcting, excluding anomalies, recalculating, or marking retests, and writes all review operation information into the audit trail record.

10. An autonomous multi-agent automated analysis method for quantitative detection scenarios using liquid chromatography-tandem mass spectrometry in multiple industries, characterized in that, The system applied to any one of claims 1-9 includes the following steps: S1. The data access module collects data from multiple sources to be analyzed, completes data integrity verification and unified field mapping, and generates a standardized data list. S2. The autonomous multi-agent module receives the user's natural language analysis requirements, the scene recognition agent identifies the current detection scene and specific detection task, and the central agent generates a structured task plan. S3. The process configuration agent matches the appropriate quality control rules, judgment thresholds and report templates according to the task information to generate task-specific configurations. S4. The quantitative analysis agent completes peak area reading, internal standard correction, calibration curve fitting, and concentration calculation of each sample according to the task configuration. S5, the quality assessment module completes the quality assessment of calibration, quality control and unknown samples based on exclusive rule configuration, and outputs the assessment results and basis; S6. When an abnormal or uncertain point occurs during the detection process, the manual review module is triggered to complete manual confirmation and correction, and the review information is saved synchronously. S7, the compliance report generation module generates standardized testing reports and full-process audit trail records, and completes data association and archiving.

Images