A Proteomics Data Analysis and Data Mining System Based on LLM

Abstract

This invention discloses a proteomics data analysis and data mining system based on LLM (Limited Language Understanding). The system includes a user input module for acquiring natural language descriptions input by the user; an LLM module responsible for parsing the user's input, extracting key entities, relationships, and operational intentions, and converting them into computer-readable analysis statements and SQL statements; a data storage module responsible for structured storage of proteomics analysis data; a proteomics analysis module that acquires analysis statements generated by the LLM module and performs proteomics analysis; and a report display module that presents the results generated by the proteomics analysis module to the user in an easily understandable format. This system breaks down the barriers between humans and complex proteomics data analysis, reducing the complexity of the analysis. Ordinary users can complete complex proteomics data analysis and data mining tasks using natural language descriptions to obtain the desired results.

Description

Technical Field

[0001] This invention belongs to the fields of artificial intelligence technology and proteomics technology, specifically relating to an LLM-based proteomics data analysis and data mining system. Background Technology

[0002] Proteomics is the study of all the proteins that make up an organism's cells or tissues; these proteins are collectively called the proteome. Proteomics focuses on understanding the structure, function, interactions, expression levels, and post-translational modifications of proteins, and how these properties change under different conditions, including health and disease states. The importance of proteomics research lies in the fact that, although the genome provides the basis of genetic information, proteins are the entities that perform most biological functions. Protein expression levels, activity, and interactions can vary depending on cell type, developmental stage, physiological state, and environmental factors; therefore, proteomics helps reveal how organisms respond to internal and external stimuli, and the molecular mechanisms of disease. With advancements in technology, especially mass spectrometry and bioinformatics, proteomics research has enabled large-scale, high-throughput analysis, greatly promoting the understanding and application of life sciences, particularly in precision medicine, drug discovery, and biomarker identification. Analyzing proteomics data is a multidisciplinary task requiring a range of specialized skills and knowledge. People who work on proteomics analysis are mostly from a biology background and have a solid biological foundation, but they lack programming skills, especially the ability to use languages ​​such as R, Python or Perl for data processing and scripting. Summary of the Invention

[0003] Purpose of the invention: To address the problems existing in the prior art, this invention provides a proteomics data analysis and data mining system based on LLM.

[0004] Technical solution: To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0005] A proteomics data analysis and data mining system based on LLM includes a user input module, a natural language understanding module (LLM), a data storage module, a proteomics analysis module, and a report display module.

[0006] (1) The user input module is responsible for receiving and obtaining the natural language description input by the user;

[0007] (2) The Natural Language Understanding (LLM) module is responsible for receiving the natural language instructions from the user input module, extracting the key entities, relationships, and operational intentions, rewriting them into a standard analysis mode through the LLM model, and feeding them back to the user. The user confirms that the content is correct, and the user intention is parsed. The user intention is divided into three types: new sample proteomics analysis, historical sample data query, and historical multi-sample joint analysis. The intention is rewritten into the corresponding analysis statement and SQL statement and sent to the downstream module.

[0008] (3) Data storage module, receives the SQL statement after the historical sample data query and historical multi-sample joint analysis intention in (2) is converted, performs historical sample data query, and further forwards the analysis statement in the historical multi-sample joint analysis intention in (2) to the proteomics analysis module.

[0009] (4) The proteome analysis module is responsible for receiving the analysis statement in (2) that interprets the proteome analysis intent of the new sample, and performing proteome analysis of the new sample; and receiving the historical sample data and analysis statement in the historical multi-sample joint analysis intent queried by the data storage module in (3) to perform historical multi-sample joint analysis.

[0010] (5) The report display module is responsible for receiving the results generated by the proteomics analysis module in (4) and presenting them to the user in an easy-to-understand form; and receiving the results of historical sample data query from the data storage module in (3) and presenting them to the user in an easy-to-understand form.

[0011] As a specific implementation scheme, in the user input module, natural language description refers to text combined with target analysis, including natural language descriptions of new sample analysis, historical sample queries, and historical multi-item joint analysis;

[0012] Specifically, during the analysis of new samples, the natural language description includes the new sample matrix and the content to be analyzed; during the query of historical samples, the natural language description includes the description of previous analysis items; during the joint analysis of historical multiple items, the natural language description includes the description of previous analysis items and the content to be jointly analyzed.

[0013] More specifically, when analyzing new samples, the natural language description includes the protein matrix to be analyzed by the user, a natural language description of the protein matrix, a grouping description of the matrix, and a description of the analysis mode, wherein the description of the analysis mode includes PCA analysis, heatmap analysis, data difference comparison analysis, annotation analysis of data difference comparison results, and a description of the visualization requirements for the analysis results.

[0014] When querying historical samples, the natural language description includes a description of the historical analysis items, including a historical item analysis table and the analysis results for a specific historical item;

[0015] When performing joint analysis of multiple historical projects, the natural language description includes the project names of the multiple projects, the method of merging the multiple projects, the grouping method, the analysis mode, including PCA analysis, heatmap analysis, data difference comparison analysis, annotation analysis of data difference comparison results, and a description of the visualization requirements for the analysis results.

[0016] As a specific implementation plan, in the Natural Language Understanding (LLM) module, the standard analysis pattern rewritten by the LLM model is fed back to the user. After receiving the feedback, if the intent of the LLM model rewritten matches the user's intent... Figure 1 To confirm, users simply need to describe their intent using affirmative statements and then submit. If the intent of the rewritten LLM model does not match the user's intent, the user rewrites the analysis schema and submits it. The LLM model then obtains the rewritten analysis schema, extracts the key entities, relationships, and operational intents, rewrites it into a standard analysis schema, and feeds it back to the user. This process repeats until the user confirms their intent. Figure 1 After that, the LLM model parses the user intent again.

[0017] As a specific implementation scheme, the Natural Language Understanding (LLM) module includes:

[0018] After receiving natural language instructions from the user input module, the key entities, relationships, and operational intentions are extracted. If these entities contain descriptions of the analysis matrix and sample grouping information, the mapping relationship between row and column labels and sample features, and the operation involves performing data dimension analysis, performing difference comparison analysis, annotating the data difference comparison results, and visualizing the analysis results, then the user intention is parsed as a new sample proteomics analysis. The order and dependencies of the intentions are further deconstructed and organized into a standard analysis mode with dependencies, which is then fed back to the user.

[0019] After receiving the natural language instructions from the user input module, the key entities, relationships and operation intentions are extracted. If these entities include historical project names, project query fields and how the query results should be displayed, the user intention is parsed into a historical sample data query. These intentions are then further deconstructed and fed back to the user.

[0020] After receiving natural language instructions from the user input module, the key entities, relationships, and operational intentions are extracted. If these entities contain multiple historical project names, the fields corresponding to the project names are regrouped, and the operation is data dimension analysis or data difference analysis, the results of the difference analysis are annotated, and all analysis results are visualized, then the user intention is parsed into a historical multi-sample joint analysis. These intentions are further deconstructed and fed back to the user.

[0021] As a specific implementation scheme, the Natural Language Understanding (LLM) module includes:

[0022] When the analysis intent is to perform proteomics analysis on a new sample, the intent is rewritten into a corresponding analysis statement, and then the analysis statement is transmitted to the proteomics analysis module.

[0023] When the analysis intent is to query historical sample data, the analysis intent is converted into an SQL statement and sent to the data storage module;

[0024] When the analysis intent is to perform a joint analysis of historical multiple samples, the data used in the multiple projects involved are first converted into SQL statements and sent to the data storage module along with the analysis statements in the analysis intent.

[0025] As a specific implementation plan, the proteomics analysis module includes:

[0026] The analysis of the new sample proteome includes PCA analysis, differential comparison analysis, and annotation of differential comparison results; the analysis results are then structured and sent to the data storage module for structured storage; further, the analysis results are sent to the report display module for data display.

[0027] The historical multi-sample joint analysis utilizes historical sample data and the analytical statements in the historical multi-sample joint analysis intent to perform proteomic analysis, including PCA analysis, differential comparison analysis, and differential comparison result annotation. The analysis results are then structured and sent to the data storage module for structured storage, and further sent to the report display module for data display.

[0028] As a specific implementation plan, the report display module includes easily understandable formats such as graphs and tables.

[0029] As a specific implementation plan, the report display module also includes a download function.

[0030] Large Language Models (LLMs) refer to a class of artificial intelligence language models trained using deep learning techniques, characterized by an extremely large number of parameters. The development of LLMs represents a significant milestone in the field of natural language processing. LLMs can not only understand human language but also generate computer code based on it, assisting those without programming skills in writing computational programs and completing tasks. The capabilities of large LLM models can effectively assist biologists engaged in proteomics data analysis, helping them handle complex programming tasks. Therefore, this invention proposes an LLM-based proteomics data analysis and data mining system that addresses the aforementioned problems by enabling complex proteomics analysis tasks to be completed through natural language descriptions.

[0031] Beneficial effects: Compared with the prior art, the present invention provides a proteomics data analysis and data mining system based on LLM. This system can break down the barriers between humans and complex proteomics data analysis, reduce the complexity of analysis, that is, ordinary users can complete complex proteomics data analysis and data mining work through natural language description and obtain the desired results. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the process of the present invention;

[0033] Figure 2 This refers to the analytical functions of the proteomics analysis module of the present invention.

[0034] Figure 3 This is the protein expression matrix.

[0035] Figure 4 This is a plot of the analysis results (PCA) in the report display module.

[0036] Figure 5 This is a graph (heatmap analysis) of the analysis results in the report display module. Detailed Implementation

[0037] The present invention will now be described clearly and in detail with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0038] Example 1

[0039] An LLM-based proteomics data analysis and data mining system is proposed, primarily designed to assist researchers in biological studies in conducting proteomics data analysis and mining. The technical solution is as follows:

[0040] The system includes a user input module, a Natural Language Understanding (LLM) module, a data storage module, a proteomics analysis module, and a report display module; its main functions and usage methods are as follows:

[0041] 1. In the user input module, enter the content you want to analyze, including analysis of new samples, query of historical samples, and joint analysis of historical multiple items; for analysis of new samples, enter the new sample matrix and the content you want to analyze; for query of historical samples, enter a description of the content of historical sample analysis; for joint analysis of historical multiple items, enter a description of the previous analysis items and the content you want to analyze together.

[0042] When analyzing new samples, the natural language description includes the protein matrix to be analyzed by the user, a natural language description of the protein matrix, a grouping description of the matrix, and a description of the analysis mode. The description of the analysis mode includes PCA analysis, heatmap analysis, data difference comparison analysis, annotation analysis of data difference comparison results, and a description of the visualization requirements for the analysis results.

[0043] When querying historical samples, the natural language description includes a description of the historical analysis items, including a historical item analysis table and the analysis results for a specific historical item;

[0044] When performing joint analysis of multiple historical projects, the natural language description includes the project names of the multiple projects, the method of merging the multiple projects, the grouping method, the analysis mode, including PCA analysis, heatmap analysis, data difference comparison analysis, annotation analysis of data difference comparison results, and a description of the visualization requirements for the analysis results.

[0045] 2. The Natural Language Understanding (LLM) module vectorizes the obtained user input into text. The vectorization algorithm is based on the BGE (BAAI General Embedding) algorithm built by BERT. The constructed vectorized text is then compared with the vector database for context retrieval based on vector similarity. The retrieved text contains three types of intent: new sample analysis intent, historical sample query intent, and historical multi-sample joint analysis intent.

[0046] After receiving natural language instructions from the user input module, the system extracts key entities, relationships, and operational intentions. If these entities contain descriptions of analysis matrices and sample grouping information, mapping relationships between row and column labels and sample features, and the operation involves performing data dimension analysis, differential comparison analysis, annotating the data differential comparison results, and visualizing the analysis results, then the user intention is parsed as a new sample proteomics analysis. Furthermore, the order and dependencies of the intentions are restructured and organized into a standard analysis pattern with dependencies, which is then fed back to the user.

[0047] After receiving natural language instructions from the user input module, the key entities, relationships, and operational intentions are extracted. If these entities include historical project names, project query fields, and how the query results should be displayed, the user intention is parsed into a historical sample data query. These intentions are then further deconstructed and fed back to the user.

[0048] After receiving natural language instructions from the user input module, the key entities, relationships, and operational intentions are extracted. If these entities contain multiple historical project names, the fields corresponding to the project names are regrouped, and the operation is data dimension analysis or data difference analysis, the results of the difference analysis are annotated, and all analysis results are visualized. Then, the user intention is parsed into a historical multi-sample joint analysis, and these intentions are further deconstructed and fed back to the user.

[0049] 3. After receiving the modified analysis intent of the LLM large model, if the intent aligns with the user's... Figure 1 To achieve this, the form of user input affirmative intent, along with the modified intent of the LLM large model. Figure 1 The same text is sent to the LLM large model; if the intent is inconsistent with the user's intent, the user needs to re-describe the inconsistent parts of the intent in the modified LLM large model, and then send it along with the modified intent in the LLM large model. Figure 1 Send it to the large LLM model.

[0050] 4. After receiving confirmation of the user's intent, the LLM model proceeds to the next step based on the different analytical intents:

[0051] Furthermore, if the analysis intent is to perform proteomics analysis on a new sample, the intent is converted into an analysis statement and sent to the proteomics analysis module.

[0052] Furthermore, if the analysis intent is a historical sample query, the intent is converted into an SQL statement and sent to the data storage module.

[0053] Furthermore, if the analysis intent is a joint analysis of multiple historical projects, the intent is converted into SQL statements and analysis statements and then sent to the data storage module.

[0054] 5. The data storage module receives queries including those for historical projects, which are then sent directly to the report display module; and queries for combined analysis of multiple historical projects are sent to the proteomics analysis module for combined analysis.

[0055] 6. The proteomics analysis module receives analyses including proteomics analyses of new samples and combined analyses of historical multiple samples.

[0056] The analysis of the new sample proteome included PCA analysis, differential comparison analysis, and annotation of the differential comparison results; the analysis results were then structured and sent to the data storage module for structured storage; further, the analysis results were sent to the report display module for data presentation.

[0057] Historical multi-sample joint analysis utilizes historical sample data and analytical statements in the historical multi-sample joint analysis intent to perform proteomic analysis, including PCA analysis, differential comparison analysis, and annotation of differential comparison results. The analysis results are then structured and sent to the data storage module for structured storage, and further sent to the report display module for data presentation.

[0058] The more specific methods are as follows:

[0059] The proteomic analysis of the new samples included:

[0060] Normalize the mean and variance of the protein data: x i For the expression level of one of the proteins, μ i σ is the mean of one of the proteins across all samples. i Let x′ be the standard deviation of one of the proteins across all samples. i This represents the normalized protein expression level of one of the proteins.

[0061] PCA analysis based on the scikit-learn module in Python;

[0062] Heatmap analysis based on the ComplexHeatmap module in R;

[0063] Difference analysis was performed using a t-test. When the homogeneity of variance of the two proteins was unequal in the two groups, a t-test was used. When two proteins are equal in homogeneity of variance test in the two groups, the following is adopted: in, This represents the average value of a certain protein in group A. This represents the average value of a certain protein in group B. Let V be the variance of a certain protein in group A. Let n be the variance of a certain protein in group B. A Let n be the number of samples A. B This represents the number of samples B.

[0064] GO / KEGG enrichment analysis, enrichment analysis N represents the total number of proteins with GO / KEGG annotation information, n represents the number of differentially expressed proteins in N, M represents the total number of proteins annotated with a specific GO / KEGG entry, and x represents the number of differentially expressed proteins annotated with a specific GO / KEGG entry.

[0065] Differential protein drug database annotation: The obtained differential proteins are annotated with our own drug database.

[0066] Furthermore, all analysis results are stored in a structured data storage module.

[0067] 7. The report display module presents the data obtained from the proteomics analysis module in doc, html, or pdf format.

[0068] Example 2

[0069] like Figure 1-2 As shown, an LLM-based proteomics data analysis and data mining system includes a user input module, a natural language understanding module (LLM), a proteomics analysis module, a data storage module, and a report display module.

[0070] Step 1: In the user input module, the user describes in detail and clearly the analysis they want to perform using natural language. For example, if the user wants to conduct proteomics analysis based on a new project, such as... Figure 3 The protein expression matrix example shown shows that users first transfer their analysis data to the system as a file, and then describe the data as follows: "Columns represent features, rows represent samples, column labels represent protein names, group labels are in the second column, features start from the third column, samples start from the second row, group A is the control group, group B is the experimental group, PCA analysis is required, heatmap analysis is required, differential analysis is required, GO / KEGG enrichment analysis is required after differential analysis, and drug annotation analysis is required."

[0071] Step 2: After receiving the natural language input from the user input module, the Natural Language Understanding (LLM) module parses the user description and finds that it includes a description of matrix information and analysis content. It determines that the user's intention is to perform proteomic analysis on new samples and rewrites the intention as "Protein matrix (rows: samples; starting from the second row, columns: features, starting from the third column); grouped as [A,B], located in the second column; analysis content is [PCA,Heatmap,Diff,GO,KEGG,Drug]".

[0072] Step 3: After receiving the modified intent from the Natural Language Understanding (LLM) module in Step 2, the user input module checks if it matches its own description and adds a positive message to the modified intent. The intent then becomes: "Protein matrix (rows: samples; starting in the second row, columns: features, starting in the third column); grouped as [A,B], located in the second column; analysis content is [PCA,Heatmap,Diff,GO,KEGG,Drug]; no problem." After user confirmation, the modified intent is sent to the LLM module. If the modified intent does not match the user's analysis intent, the user modifies the analysis intent, such as changing the description after grouping: "Protein matrix (rows: samples; starting in the second row, columns: features, starting in the third column); grouped as [A,B], located in the second column; analysis content is [PCA,Heatmap,Diff,GO,KEGG,Drug]; grouping error, grouped as [A,C], everything else correct," and sends the modified intent to the LLM module.

[0073] Step 4: After receiving the user's confirmed intent in Step 3, the Natural Language Understanding (LLM) module converts the analysis intent into an analysis statement: "feature = col[3:]; sample = row[2:]; group = col[2]; groups = [A, B]; analysis = [PCA, Heatmap, Diff, GO, KEGG, Drug]" and sends it to the proteomics analysis module. If the received statement is a modification of the intent in Step 3, the intent is further rewritten as "protein matrix (row: sample; starting from the second row, column: feature, starting from the third column); grouped as [A, C], located in the second column; analysis content is [PCA, Heatmap, Diff, GO, KEGG, Drug]", and returns to Step 3. This process is repeated until the user confirms that the intent is correct. In this example, the user only modified the intent once and confirmed that there were no problems. Finally, the intent after grouping was modified to "Protein matrix (row: sample; starting from the second row, column: feature, starting from the third column); grouped as [A,C], located in the second column; analysis content is [PCA,Heatmap,Diff,GO,KEGG,Drug]; no problem"; After receiving the intent, LLM rewrote it into the analysis statement: "feature=col[3:]; sample=row[2:]; group=col[2]; groups=[A,C]; analysis=[PCA,Heatmap,Diff,GO,KEGG,Drug]" and sent it to the proteomics analysis module.

[0074] Step 5: The proteomics analysis module receives the new sample proteomics analysis statement "feature=col[3:];sample=row[2:];group=col[2];groups=[A,C];analysis=[PCA,Heatmap,Diff,GO,KEGG,Drug]" sent by the natural language understanding module LLM. It verifies the protein data against the analysis statement, splits the analysis statement by ";", reads the keywords "feature, sample, group, groups, and analysis", verifies the assignment information corresponding to the keywords "feature, sample, group, groups" against the protein matrix, and then determines whether the analysis corresponding to "analysis" is included in the following... Figure 2 In the analysis module shown, if there are no errors, the corresponding analysis in "analysis" will be performed. After the analysis is completed, the analysis results will be sent to the report display module, and the structured analysis results will be sent to the data storage module for data formatting and storage. If a discrepancy is found between the sample and the sample description, such as a protein matrix grouped as A and C but described as A and B, or the analysis content in "analysis" is inconsistent with the description, the analysis will be performed accordingly. Figure 2 If there is a conflict in the analysis module, the analysis will end and the reason for the error will be sent to the report display module.

[0075] Step 6: The data storage module receives the formatted protein analysis results from Step 5, including protein item information, group analysis, and the path of the analysis results, and stores them in the database.

[0076] Step 7: The report display module, upon receiving the analysis results from the proteomics analysis module in Step 5, displays the results to the user in doc, html, and pdf formats, providing download options. Some analysis results are available in [link to relevant documentation]. Figure 4 (PCA) Figure 5 (Heatmap analysis)

[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A proteomics data analysis and data mining system based on LLM, characterized in that, It includes a user input module, a Natural Language Understanding (LLM) module, a data storage module, a proteomics analysis module, and a report display module; (1) The user input module is responsible for receiving and obtaining the natural language description input by the user; (2) The Natural Language Understanding (LLM) module is responsible for receiving the natural language instructions from the user input module, extracting the key entities, relationships, and operational intentions, rewriting them into a standard analysis mode through the LLM model, and feeding them back to the user. The user confirms that the content is correct, and the user intention is parsed. The user intention is divided into three types: new sample proteomics analysis, historical sample data query, and historical multi-sample joint analysis. The intention is rewritten into the corresponding analysis statement and SQL statement and sent to the downstream module. When the user's intent is to perform proteomics analysis on a new sample, the intent is rewritten into a corresponding analysis statement, and then the analysis statement is transmitted to the proteomics analysis module. When the user's intent is to query historical sample data, the intent is converted into an SQL statement and sent to the data storage module. When the user intent is to perform joint analysis of historical multiple samples, the data used in the multiple projects involved are first converted into SQL statements and sent to the data storage module along with the analysis statements in the intent. (3) Data storage module, receives the SQL statement after the historical sample data query and historical multi-sample joint analysis intention in (2) is converted, performs historical sample data query, and further forwards the analysis statement in the historical multi-sample joint analysis intention in (2) to the proteomics analysis module. (4) The proteomics analysis module is responsible for receiving the analysis statement in (2) that interprets the proteomics analysis intent of the new sample, and performing proteomics analysis of the new sample; and receiving the historical sample data and analysis statement in the historical multi-sample joint analysis intent queried by the data storage module in (3) to perform historical multi-sample joint analysis. (5) The report display module is responsible for receiving the results generated by the proteomics analysis module in (4) and presenting them to the user in an easy-to-understand form; and receiving the results of historical sample data query from the data storage module in (3) and presenting them to the user in an easy-to-understand form.

2. The LLM-based proteomics data analysis and data mining system according to claim 1, characterized in that, In the user input module, natural language description refers to text combined with target analysis, including natural language descriptions of new sample analysis, historical sample queries, and historical multi-item joint analysis. When analyzing new samples, the natural language description includes the new sample matrix and the content to be analyzed; when querying historical samples, the natural language description includes the description of previous analysis items; when performing joint analysis of historical multiple items, the natural language description includes the description of previous analysis items and the content to be jointly analyzed. When querying historical samples, the natural language description includes a description of the historical analysis project, including a historical project analysis table and the analysis results for a specific historical project. When performing the historical multi-project joint analysis, the natural language description includes the project names of the multiple projects, the method of merging the multiple projects, the grouping method, the analysis mode, including PCA analysis, heatmap analysis, data difference comparison analysis, annotation analysis of data difference comparison results, and a description of the visualization requirements for the analysis results.

3. The LLM-based proteomics data analysis and data mining system according to claim 2, characterized in that, When analyzing the new sample, the natural language description includes the protein matrix to be analyzed by the user, a natural language description of the protein matrix, a grouping description of the matrix, and a description of the analysis mode. The description of the analysis mode includes PCA analysis, heatmap analysis, data difference comparison analysis, annotation analysis of data difference comparison results, and a description of the visualization requirements for the analysis results.

4. The LLM-based proteomics data analysis and data mining system according to claim 1, characterized in that, In the Natural Language Understanding (LLM) module, the LLM model rewrites the data into a standard analysis pattern and then feeds it back to the user. If the user's intent is consistent with the intent of the LLM model after receiving the feedback, the user only needs to describe it with affirmative statements and then submit it. If the intent rewritten by the LLM model is inconsistent with the user's intent, the user rewrites the analysis schema and submits it. The LLM model then obtains the rewritten analysis schema, extracts the key entities, relationships, and operational intents, rewrites it into a standard analysis schema, and feeds it back to the user. This process is repeated until the user confirms that the intents are consistent, at which point the LLM model parses the user's intent again.

5. The LLM-based proteomics data analysis and data mining system according to claim 1, characterized in that, In the Natural Language Understanding Module (LLM): After receiving natural language instructions from the user input module, the key entities, relationships, and operational intentions are extracted. If these entities contain descriptions of the analysis matrix and sample grouping information, the mapping relationship between row and column labels and sample features, and the operation involves performing data dimension analysis, performing difference comparison analysis, annotating the data difference comparison results, and visualizing the analysis results, then the user intention is parsed as a new sample proteomics analysis. The order and dependencies of the intentions are further deconstructed and organized into a standard analysis mode with dependencies, which is then fed back to the user. After receiving the natural language instructions from the user input module, the key entities, relationships and operation intentions are extracted. If these entities include historical project names, project query fields and how the query results should be displayed, the user intention is parsed into a historical sample data query. These intentions are then further deconstructed and fed back to the user. After receiving natural language instructions from the user input module, the key entities, relationships, and operational intentions are extracted. If these entities contain multiple historical project names, the fields corresponding to the project names are regrouped, and the operation is data dimension analysis or data difference analysis, the results of the difference analysis are annotated, and all analysis results are visualized, then the user intention is parsed into a historical multi-sample joint analysis. These intentions are further deconstructed and fed back to the user.

6. The LLM-based proteomics data analysis and data mining system according to claim 1, characterized in that, In the proteomics analysis module: The analysis of the new sample proteome includes PCA analysis, differential comparison analysis, and annotation of differential comparison results; the analysis results are then structured and sent to the data storage module for structured storage; further, the analysis results are sent to the report display module for data display. The historical multi-sample joint analysis utilizes historical sample data and the analytical statements in the historical multi-sample joint analysis intent to perform proteomic analysis, including PCA analysis, differential comparison analysis, and differential comparison result annotation. The analysis results are then structured and sent to the data storage module for structured storage, and further sent to the report display module for data display.

7. The LLM-based proteomics data analysis and data mining system according to claim 1, characterized in that, In the report display module, the easy-to-understand formats include graphs and tables.

8. The LLM-based proteomics data analysis and data mining system according to claim 1, characterized in that, The report display module also includes a download function.

Images