Intelligent system, method and equipment for assisting multi-step genome data analysis

Abstract

The invention relates to the technical field of genome data analysis, and discloses an intelligent system, method and equipment for assisting multi-step genome data analysis, and the system comprises a dialogue agent which is used for generating a corresponding answer according to a question of a user, or reading an analysis plan file generated by a workflow agent, generating an analysis interpretation text for the analysis plan file; the workflow agent is used for generating a structured task execution plan according to the to-be-executed analysis task and executing the to-be-executed analysis task; and the modeling analysis agent is used for generating a configuration file and a script based on the user request, constructing a model and generating an analysis result corresponding to the user request in combination with the workflow agent. Through multi-agent cooperation, task division and cooperative scheduling are realized, each agent independently completes task planning, execution control, model analysis and other functions, the bottleneck problem of processing of a traditional single model in a complex process is avoided, error accumulation is reduced, and the execution efficiency and stability of the whole process are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of genomic data analysis, and particularly relates to an intelligent system, a method and equipment for assisting multi-step genomic data analysis. BACKGROUND

[0002] In multi-step and complex biological data analysis, workflow based on automated intelligent agent system has made significant progress in fields such as single-cell genomics, ribonucleic acid (RNA) editing and multi-omics data processing. Commonly used artificial intelligence-based automated bioinformatics analysis tools such as AutoBA (Automated Bioinformatics Analysis) and CellAgent process task planning, execution and verification through role-specific intelligent agents, coordinate each step through memory mechanism, and dynamically extract domain-specific knowledge to support workflow execution by combining retrieval-augmented generation (RAG).

[0003] With the increasing complexity of bioinformatics workflow, existing systems often cannot efficiently handle error propagation problems in long-time processes, and are difficult to adapt to the introduction of emerging tools, resulting in low efficiency of workflow execution. Although existing systems combine RAG to improve tool adaptability, the integration problem between knowledge acquisition and actual execution is still prominent in cross-domain complex analysis, resulting in a lack of deep support for multi-step genomic data analysis. Most existing systems lack effective memory management strategies or unreasonable role allocation when managing multiple steps and large intermediate results, resulting in performance degradation when processing large-scale biological data. SUMMARY

[0004] Therefore, the present application provides an intelligent system, a method and equipment for assisting multi-step genomic data analysis to solve the problems of low efficiency and high error rate of workflow in the prior art.

[0005] In a first aspect, the present application provides an intelligent system for assisting multi-step genomic data analysis, the system comprising: a multi-agent collaboration module, the multi-agent collaboration module comprising: a dialogue agent, a workflow agent, and a modeling analysis agent.

[0006] The dialogue agent is configured to generate a corresponding answer according to a user question, or read an analysis plan file generated by the workflow agent and generate an analysis interpretation text for the analysis plan file, the analysis plan file comprising a structured task execution plan generated by the workflow agent.

[0007] A workflow agent is configured to generate a structured task execution plan according to an analysis task to be executed, and execute the analysis task to be executed according to the structured task execution plan.

[0008] A modeling analysis agent is configured to generate a configuration file and a script based on a user request, build a model, and generate an analysis result corresponding to the user request in combination with the workflow agent.

[0009] The intelligent system for assisting multi-step genomic data analysis provided by the application achieves task division and collaborative scheduling through multi-agent cooperation, each agent independently completes functions such as task planning, execution control, and model analysis, avoids the bottleneck problem of traditional single models in complex processes, and significantly reduces error accumulation and improves the execution efficiency and stability of the overall process.

[0010] In an optional embodiment, the system further comprises a front-end interaction module, a back-end processing module, a knowledge enhancement module, and a memory module.

[0011] The front-end interaction module is configured to create a project and a session, configure parameters, obtain a user question or a user request, and view an answer or an analysis result.

[0012] The back-end processing module is in communication connection with the front-end interaction module, and is configured to schedule the multi-agent cooperation module to execute a corresponding task according to the user question or the user request, and feed back a task execution state and an analysis result to the front-end interaction module.

[0013] The knowledge enhancement module is configured to generate a vector knowledge base in the field of genomics, perform semantic retrieval on the vector knowledge base to obtain a knowledge fragment according to the user question or the user request, and add the knowledge fragment to the multi-agent cooperation module to execute a corresponding task.

[0014] The memory module is configured to store historical user interaction information for a long time, and assist the multi-agent cooperation module in understanding a user intention according to the historical user interaction information, and generate a result meeting a user demand.

[0015] The intelligent system for assisting multi-step genomic data analysis provided by the application, the front-end interaction module supports project management, task flow, configuration editing and the like, the back-end processing module automatically executes scheduling and log tracking, ensures that user requests can accurately trigger multi-agent collaborative execution of corresponding tasks, and can timely feedback task status and results to the user, ensures real-time and smoothness of interaction, realizes organic unification of user friendliness and engineering stability, the knowledge enhancement module assists the generated genomic field vector knowledge base to perform semantic retrieval on user requests and provide knowledge fragments, provides professional knowledge support for multi-agent task execution, improves accuracy and professionalism of task execution, the memory module long-term stores historical interaction information, assists multi-agent to better understand user intent, makes the generated results more in line with user past needs and use habits, and enhances personalized service capability and intelligence of the system.

[0016] In an optional implementation, generating a corresponding answer according to the user question, comprising:

[0017] Configuring a service interface by using a locally deployed large model interface service address, and selecting a target large language model;

[0018] Building a dialogue template including system task instructions, typical questions and question and answer examples, and real-time user input;

[0019] Obtaining the user question based on the dialogue template, and submitting the user question to the target large language model for inference calculation based on the service interface, and automatically converting the output parser into text content as the corresponding answer to the user question.

[0020] The intelligent system for assisting multi-step genomic data analysis provided by the application, by configuring the service interface of the locally deployed large model, can ensure data security and reduce dependence on external networks; the dialogue template containing system task instructions, typical question and answer and the like can standardize the interaction logic, so that the large model can more accurately understand the task; the output parser automatically converts the inference result into text, which not only improves the efficiency of generating answers, but also ensures that the answer format is standard and easy to read, thereby providing users with more efficient, accurate and demand-oriented question and answer experience.

[0021] In an optional implementation, reading an analysis plan file generated in the background, and generating an analysis interpretation text for the analysis plan file, comprising:

[0022] Reading the analysis plan file generated by the workflow intelligent agent, and extracting file parameters therefrom, the file parameters including: task step description, input file, output file, and used tool;

[0023] Retrieving a script file for task execution in the back end, and parsing commands and parameters in the script file;

[0024] Based on the file parameters and the commands and parameters in the script file, the user is asked, the execution plan is executed, the current task step is analyzed, and an analysis interpretation text is generated.

[0025] The intelligent system for assisting multi-step genomic data analysis provided by the application extracts key file parameters from the analysis plan file, retrieves script commands and parameters executed by the analysis backend, and comprehensively analyzes the user questions, execution plans and current task steps based on the above, so that the analysis interpretation text generated by the user can clearly and intuitively understand the analysis plan and task execution, effectively reduce the difficulty of understanding professional analysis content for the user, and improve the user's cognition of the system analysis process.

[0026] In an optional embodiment, the workflow intelligent agent comprises: a plan intelligent agent, a task intelligent agent, a correction intelligent agent, and a check intelligent agent.

[0027] According to the structured task execution plan, the analysis task to be executed is executed, comprising:

[0028] The plan intelligent agent obtains the analysis task to be executed, and finds related knowledge from the vector knowledge base in the field of genomics, splits the analysis task to be executed into specific steps and executes them in order.

[0029] During the sequential execution of the specific steps, the task intelligent agent finds related knowledge from the tool library and the tool knowledge base to select a target tool, and generates a script corresponding to each specific step using the target tool.

[0030] The scripts corresponding to each specific step are executed in turn, and the correction and checking process is executed in a loop until the checking results of all scripts meet the preset standard, and the correction and checking process comprises: if the execution is unsuccessful, the code is modified according to the execution result using the correction intelligent agent and executed again until the execution is successful or the number of repeated executions exceeds the repetition threshold, the execution result is checked using the check intelligent agent, and if the checking result does not meet the preset standard, the process of modifying the code according to the execution result using the correction intelligent agent is returned.

[0031] In an optional embodiment, if the execution is unsuccessful, the code is modified according to the execution result using the correction intelligent agent and executed again, comprising:

[0032] If the execution is unsuccessful, the correction intelligent agent analyzes the unsuccessful reason according to the execution result.

[0033] Based on the unsuccessful reason, related knowledge is found from the tool library and the tool knowledge base, and the code is modified based on the related knowledge and executed again.

[0034] The intelligent system for assisting multi-step genomic data analysis provided by the application can obtain knowledge from a genomic field vector knowledge base through a planning intelligent agent, can accurately split a to-be-executed analysis task into specific steps and execute them in order, and guarantees the professionalism and orderliness of task execution; a task intelligent agent selects a target tool and generates a corresponding script by means of a tool library and a tool knowledge base, and realizes efficient adaptation of the tool and automatic generation of the script; an error correction intelligent agent can modify code and repeatedly execute in the case of unsuccessful execution, and a checking intelligent agent strictly checks the execution result, and the error correction and checking mechanism formed by the two can discover and solve problems in task execution in time, and ensure that all script execution results meet preset standards, thereby improving the accuracy and reliability of the entire analysis task execution.

[0035] In an optional implementation, the configuration file and the script are generated based on the user request, the model is built, and the analysis result corresponding to the user request is generated in combination with the workflow intelligent agent, including:

[0036] The task target and the data list are parsed based on the user request, and a parameter agent is called to generate a standardized parameter set of the population genetic analysis tool;

[0037] Based on the standardized parameter set, a configuration file generation intelligent agent is called to generate a standard configuration file;

[0038] A script generation intelligent agent is called to generate a script file for executing the population genetic modeling tool, and an execution model corresponding to the user request is established by executing the script file;

[0039] The analysis result corresponding to the user request is generated by using the execution model and the standard configuration file.

[0040] The intelligent system for assisting multi-step genomic data analysis provided by the application can accurately parse the task target and the data list based on the user request, generate a standardized parameter set by means of a parameter agent, and then automatically generate a standard configuration file and an execution script by means of a configuration file generation intelligent agent and a script generation intelligent agent, efficiently build an execution model, quickly generate an analysis result meeting the user demand in combination with a workflow intelligent agent, and improve the automation and efficiency of analysis task execution.

[0041] In an optional implementation, the modeling analysis intelligent agent is further used for interpreting and analyzing the result, including:

[0042] An analysis plan file generated by the workflow intelligent agent is read, and the latest execution step is extracted therefrom;

[0043] Relevant knowledge is obtained by using a knowledge enhancement module based on the user request;

[0044] According to the latest execution step and the relevant knowledge, a result interpretation text and a model analysis description text corresponding to the user request are generated.

[0045] The intelligent system for assisting multi-step genomic data analysis provided by the present application can automatically read a result file after task execution is completed, call a proper modeling tool to perform inference analysis, and output in the form of a visual chart or a statistical summary, thereby greatly improving the data interpretation and model fitting efficiency of researchers. In combination with relevant knowledge obtained by a knowledge enhancement module, the intelligent system generates a result interpretation and model analysis explanation text, so that users can more clearly and deeply understand the analysis result, and the analysis result is more interpretable and easy to use, thereby assisting users to better grasp the core information of genomic data analysis.

[0046] In a second aspect, the present application provides an intelligent method for assisting multi-step genomic data analysis, which is applied to the intelligent system for assisting multi-step genomic data analysis in any one of the first aspect, and the method comprises the following steps.

[0047] generating a corresponding answer according to a user question, or reading an analysis plan file generated by the workflow agent, and generating an analysis interpretation text for the analysis plan file, wherein the analysis plan file comprises a structured task execution plan generated by the workflow agent;

[0048] generating a structured task execution plan according to a to-be-executed analysis task, and executing the to-be-executed analysis task according to the structured task execution plan;

[0049] generating a configuration file and a script based on a user request, building a model, and generating an analysis result corresponding to the user request in combination with the workflow agent.

[0050] In a third aspect, the present application provides a computer device, which comprises a memory and a processor, the memory and the processor are communicatively connected with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method of the second aspect or any one of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or the prior art description. Obviously, the drawings described below are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0052] Figure 1 is a structural block diagram of the intelligent system for assisting multi-step genomic data analysis according to an embodiment of the present application;

[0053] Figure 2 is a flowchart of the data processing of the dialogue agent in the intelligent system for assisting multi-step genomic data analysis according to an embodiment of the present application;

[0054] Figure 3 is a flowchart of a workflow agent processing data in an intelligent system for assisting multi-step genomic data analysis according to an embodiment of the present application;

[0055] Figure 4 is a flowchart of a model analysis agent processing data in an intelligent system for assisting multi-step genomic data analysis according to an embodiment of the present application;

[0056] Figure 5 is a schematic diagram of a modeling analysis agent processing data results in one specific embodiment of an intelligent system for assisting multi-step genomic data analysis according to an embodiment of the present application;

[0057] Figure 6 is a flowchart of an intelligent method for assisting multi-step genomic data analysis according to an embodiment of the present application;

[0058] Figure 7 is a hardware structure schematic diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0059] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0060] The embodiments of the present application provide an intelligent system for assisting multi-step genomic data analysis, which realizes task division and collaborative scheduling through multi-agent cooperation, so as to reduce error accumulation and improve the execution efficiency and stability of the overall process.

[0061] Genomic data analysis includes multiple branches such as variant detection, functional annotation, phenotype association and population genetics. The population genetics analysis of biological information takes population-scale genotype / variant data as the object, infers population structure, kinship, genetic diversity, historical scale and migration, selection signal, etc. The typical process often involves multi-step serial and parallel (quality control → dimension reduction / inference → statistics and visualization → modeling and interpretation), and there are many tools and parameters, coupling and error propagation risks between steps, which are extremely in need of automatic and recoverable execution system.

[0062] The multi-agent collaboration framework (planning / execution / debugging / inspection / interpretation / conversation) provided by the embodiment is designed for such multi-step and long process scenarios, and stably maps the population genetic analysis from the "natural language requirement" to the "executable, traceable and recoverable" pipeline, and dynamically adapts the tools and parameters through the RAG and memory mechanism during the execution process.

[0063] According to the embodiment of the application, an intelligent system for assisting multi-step genomic data analysis is provided, which comprises a multi-agent collaboration module, the multi-agent collaboration module comprising a plurality of agents, each agent playing a different role, such as Figure 1 As shown, the multi-agent collaboration module comprises a conversation agent 1, a workflow agent 2 and a modeling analysis agent 3.

[0064] The conversation agent 1 is configured to generate a corresponding answer according to a user question, or read an analysis plan file generated by the workflow agent 2 and generate an analysis interpretation text for the analysis plan file, the analysis plan file comprising a structured task execution plan generated by the workflow agent 2.

[0065] Specifically, the conversation agent 1 (ChatAgent) integrates a large amount of bioinformatics literature and research data to provide comprehensive knowledge support for the task. The user inputs a user question (for example, a task request or a question information) through a front-end interaction module, the front-end page provides a network service using a Django framework, and listens to and receives the input request of the user in real time; the Django server of the system back-end processing module obtains the user request information, and directly transmits the input data to the ChatAgent of the back-end multi-agent collaboration module for processing through a pre-defined routing mechanism.

[0066] The workflow agent 2 is configured to generate a structured task execution plan according to a to-be-executed analysis task, and execute the to-be-executed analysis task according to the structured task execution plan.

[0067] Specifically, the workflow agent 2 (Execute Agent) is responsible for the planning, execution and error checking of the task, and ensures the smooth execution of the entire workflow. Compared with the existing agent system, the workflow agent 2 can support longer and more complex bioinformatics population genetic analysis, and has better scalability.

[0068] The user inputs a requirement or a target data file of a to-be-executed analysis task through a front-end interaction module, the front-end page provides a network service using a Django framework, and receives and listens to the execution request of the user in real time; the Django server of the system back-end processing module transmits the related task information, data path and configuration information to the Execute Agent of the back-end multi-agent collaboration module for automatic processing through a pre-defined interface mechanism after receiving the analysis task request submitted by the user.

[0069] The Execute Agent automatically completes task planning (PLAN) based on the LangChain framework and a predefined Prompt template, and enters the sub-task execution (TASK) phase according to the automatically generated task planning, to automatically execute each task.

[0070] The modeling and analysis agent 3 is configured to generate configuration files and scripts based on user requests, build models, and generate analysis results corresponding to user requests in combination with the workflow agent 2.

[0071] Specifically, the modeling and analysis agent 3 (Analysis Agent) can analyze pictures and data in the workflow, use tools such as FastSimCoal to perform modeling and generate corresponding charts. The user inputs a user request (such as an analysis requirement or target data) through the front-end interaction module, and the system front-end interface receives the user request in real time through the Django framework and transmits the related task requirements and data paths to the Analysis Agent through the back-end interface. After receiving the user request, the Analysis Agent of the system back-end initializes the environment and performs model configuration, including API key, model service address, and model type configuration. It should be noted that the Analysis Agent supports multiple model configuration methods, including but not limited to external interface services (such as OpenAI's GPT-4o model) and locally deployed models (such as the open-source models Deepseek-R1, Qwen3-235B deployed by Ollama), thereby enabling completely offline analysis task execution; the Analysis Agent has a built-in Prompt template to achieve standardized task reasoning and parameter generation, ensuring the efficiency and repeatability of analysis.

[0072] The intelligent system for assisting multi-step genomic data analysis provided in the embodiment achieves task division and collaborative scheduling through multi-agent cooperation, and each agent independently completes task planning, execution control, model analysis, and other functions, thereby avoiding the bottleneck problem of traditional single models in complex processes and significantly reducing error accumulation and improving the execution efficiency and stability of the overall process.

[0073] In some optional embodiments, the system further includes a front-end interaction module, a back-end processing module, a knowledge enhancement module, and a memory module.

[0074] The front-end interaction module is configured to create projects and sessions, configure parameters, obtain user questions or user requests, and view answers or analysis results.

[0075] Specifically, the system has a complete front-end interactive module and a back-end processing module supporting multi-threading. After the system is opened, the default main interface is entered, which includes "Project", "New session" and other entries.

[0076] After clicking the "Project" button, the system will pop up or switch to the project management page, which is usually used to create a new session, manage existing projects, or view the analysis status under the current project. The "Project" page includes the "New session" or "create session" option. Clicking it will automatically generate a new session or workflow entry for subsequent bioinformatics analysis tasks in the session. After successfully creating a session, the system will automatically generate a default name (such as "NewSession"). At this time, the user can customize the session name (such as "population genetics analysis").

[0077] After completing the session creation, you can enter the system configuration center by setting the button. In the settings page, complete the relevant settings, including: ① API Setting (the user needs to fill in the API key of the model call (such as OpenAI's Key), which is used for subsequent communication with large models), ② Model Setting (set the model service address. For example, fill in https: / / one-api.xxx.xxx / v1 as the default large model API interface path), ③ Execute or not (used to start or stop the actual execution flow control function. Setting to True indicates that the actual task execution is supported; setting to False, the system only performs session reasoning, and the task will not be sent to the back-end execution system), ④ Manage Docs (click the top "NEW DOCUMENT" button to add a task document (such as detailed steps of population genetics analysis process). Fill in the analysis content and planned steps in the pop-up window to complete the addition of the task document. The document information is stored for subsequent task planning or explanation), ⑤ ManageTools (list the current analysis tool description documents connected by the system. You can extend the tool library by uploading.txt format files, or delete existing documents. Tool information will be called by the agent in the execution phase for reasoning and execution matching), ⑥ Upload Data (the user can upload data links for analysis. The system will pull the data to the back-end analysis environment through the cloud interface).

[0078] In the "Project" page, there are "Chat Agent", "Execute Agent", "Analysis Agent" buttons. Users can click the corresponding button according to their needs to enter different agents.

[0079] (1) In the page of "Chat Agent", input questions, the system will call Chat Agent for professional knowledge based on the current session and configuration state, such as asking professional questions in the field of population genetics "What is the difference between PCA and Admixture in population analysis?", requesting task plan suggestions "Please generate a stepwise plan to perform population structure analysis", asking questions about the results of workflow agent 2 or modeling analysis agent 3 "Please help explain the results from the ADMIXTURE step" "Can you interpret the PCA plot from AnalysisAgent?", just as examples, but not limited to. Chat Agent is the most versatile and flexible agent, which can schedule information across modules, interpret results, assist teaching and guide execution, and is suitable for users to understand and control the system.

[0080] (2) In the page of "Execute Agent", the task instruction can be inputted and the required data can be selected, with the functions of inputting task description in the dialogue box, uploading data files, etc. After confirming that the text and data input are correct, the "Send" button is clicked to automatically call the agent, plan and execute the task to be analyzed. The workflow agent 2 analyzes the planning of the task according to the content input by the user and the data uploaded, displays the complete execution plan (PLAN), and parses the task into multiple sub-steps, each sub-step containing four parameters: Description (explaining the function and goal of the step), Input Files (the input file path required for each step), Output Files (the output file path generated by the step), and Tools (the tool name used in the task, such as PLINK or custom R / Python script). In the PLAN display interface, there are two operation buttons: "Execute PLAN" and "Edit PLAN". "Execute PLAN" means confirming the entire task step planned by the system and executing it immediately. The system will sequentially schedule the required agents (Task Agent, Debug Agent, etc.) and tools to execute each step of operation, and the progress and result output of each step can be viewed. "Edit PLAN" can modify the task planned by the system on the current basis, including: adding / deleting steps, replacing tools or modifying parameters, inserting new analysis processes (such as adding principal component analysis or population structure analysis), and after modification, the updated task process can be executed again. The task is mapped from natural language input by the user to machine executable steps, and the uploaded data path can be automatically read, and the standardized process operation is completed in the backend.

[0081] After completing the execution planning of the task plan, the entire task plan is automatically executed, and the task execution progress is displayed in percentage form. During the execution process, "Stop Plan" can be clicked at any time to stop the task execution, and after stopping, the execution script can be viewed, the command can be modified, and the task can be executed again. After execution is completed, a report can be generated according to the execution result and the corresponding interpretation.

[0082] (3) In the page of "Analysis Agent", the analysis instruction can be inputted, the execution result or original data file can be uploaded, and the output of the workflow agent 2 can be combined to automatically call the modeling method to execute modeling, parameter fitting or result evaluation. The output result can also be discussed through the dialogue agent 1 and the modeling analysis agent 3, for example: through FastSimCoal to perform population history modeling and parameter estimation, which is only an example and is not limited thereto.

[0083] The back-end processing module is in communication connection with the front-end interaction module, and is configured to dispatch the multi-agent collaboration module to perform corresponding tasks according to the user questions or user requests, and feed back the task execution status and analysis results to the front-end interaction module.

[0084] Specifically, the back-end processing module provides technical support for data processing for the front-end interaction module, selects different agents to perform corresponding tasks according to user questions or user requests, and feeds back the task execution status and analysis results to the front-end interaction module. The back-end processing module can be a Django server, which, after receiving the analysis task request submitted by the user, transmits the related task information, data path and configuration information to the corresponding agent in the multi-agent collaboration module through a predefined interface mechanism for automatic processing.

[0085] The knowledge enhancement module is configured to generate a vector knowledge base in the field of genomics, perform semantic retrieval on the vector knowledge base to obtain a knowledge fragment according to the user questions or user requests, and add the knowledge fragment to the multi-agent collaboration module to perform corresponding tasks.

[0086] Specifically, the knowledge enhancement module runs through all the agents, and is configured to improve the vector knowledge base in the field of genomics according to the historical operation data of each agent. The knowledge contained in the vector knowledge base of each agent belongs to the field of genomics, and each knowledge base can be optimized using the execution analysis results of different agents, but the specific knowledge content contained in each knowledge base is different. The dialogue agent focuses on answering questions, and the dialogue knowledge base contains basic bioinformatics common sense. The workflow agent focuses on process planning, and the plan knowledge base of the planning agent contains common bioinformatics processes. The tool knowledge base of the task agent and the error correction agent is a manual for using tools. The modeling and analysis agent focuses on analyzing user requests through modeling, and the analysis knowledge base contains modeling parameters and other knowledge. The knowledge enhancement module of different agents specifically includes:

[0087] (1) The dialogue agent 1 introduces a retrieval-augmented generation (RAG) function, preloads knowledge documents related to the field of population genetics, including but not limited to PDF and TXT format files, and uses a local vector embedding model (granite-embedding) deployed by ollma to generate vector representations of the documents, which are stored in a Chroma vector database. During the processing of user requests, the current user input is automatically searched semantically, and the highly relevant knowledge document content is retrieved and obtained. The relevant document fragments retrieved are automatically fused into the Prompt input, which significantly improves the completeness of the background knowledge of the ChatAgent and the accuracy of the answers, and realizes intelligent interactive knowledge query and reasoning.

[0088] (2) Workflow Agent 2 integrates dynamic tool matching and RAG functions, specifically including: before the execution of each sub-task step, Execute Agent automatically loads the usage instructions and technical details of the relevant tools from the tool description document or network link according to the task requirements; Execute Agent uses Chroma vector database to store bioinformatics analysis tool related knowledge documents, including but not limited to PDF and TXT format files, and uses the local vector embedding model (granite-embedding) deployed by ollma to generate vector representations of the documents, which are stored in the Chroma vector database; During the execution of the sub-task, the content of the current task step and the tool are automatically searched semantically, and highly relevant knowledge fragments are retrieved and fused into the Prompt input in real time, significantly enhancing the completeness of the knowledge background and the accuracy of command generation during task execution.

[0089] (3) Modeling analysis agent 3 integrates RAG mechanism, specifically including: preloading knowledge documents containing population genetic analysis tool usage instructions and configuration methods, including but not limited to PDF and TXT format files, and using the local vector embedding model (granite-embedding) deployed by ollma to generate vector representations of the documents, which are stored in the Chroma vector database; During the execution of the modeling, the system automatically performs real-time semantic retrieval based on the current analysis task target and data characteristics to accurately obtain highly relevant document fragments for the current modeling task; automatically fuse the retrieved knowledge fragments into the Prompt input content, effectively enhancing the completeness of the knowledge background of Analysis Agent and improving the accuracy of task execution.

[0090] Memory module for long-term storage of historical user interaction information, and assisting the multi-agent collaboration module in understanding user intent and generating results that meet user needs based on historical user interaction information.

[0091] Specifically, the dialogue agent 1 and the modeling analysis agent 3 have long-term memory mechanisms built-in, which store and analyze user interaction history information and preferences for a long time, dynamically adjust Prompt generation strategies and knowledge retrieval directions, and continuously optimize personalized interaction experience and analysis accuracy.

[0092] The intelligent system assisting multi-step genomic data analysis provided by the embodiment, the front-end interaction module supports project management, task transfer, configuration editing and other operations, the back-end processing module automatically executes scheduling and log tracking, ensures that user requests can accurately trigger multi-agent collaborative execution of corresponding tasks, and can timely feedback task status and results to users, ensures the real-time and smoothness of interaction, realizes the organic unity of user-friendly and engineering stability, the knowledge enhancement module assists the generated genomic field vector knowledge base to perform semantic retrieval on user requests and provides knowledge fragments, provides professional knowledge support for multi-agent task execution, improves the accuracy and professionalism of task execution, the memory module long-term stores historical interaction information, assists multi-agent to better understand user intent, makes the generated results more in line with user past needs and use habits, enhances the personalized service ability and intelligence of the system.

[0093] In some optional embodiments, generating a corresponding answer according to the user question comprises:

[0094] Step a1, configuring a service interface by using a locally deployed large model interface service address, and selecting a target large language model.

[0095] Specifically, as shown in Figure 2 Fig. 1 is a process diagram for the dialogue agent 1 to process data, the dialogue agent 1 initialization process connects the interface by configuring the interface through the locally deployed large model interface service address, and selects the required large language model for interaction, which supports Deepseek-R1 model and the like.

[0096] Step a2, constructing a dialogue template including system task instructions, typical questions and answer examples, and real-time user input.

[0097] Specifically, a special Prompt template is constructed, which includes system task instructions, typical questions and answer examples (Few-Shot examples), and real-time user input, and generates a complete Prompt.

[0098] Step a3, obtaining a user question based on the dialogue template, and submitting the user question to the target large language model for inference calculation based on the service interface, and automatically converting the output parser into text content as the corresponding answer to the user question.

[0099] Specifically, the complete Prompt is submitted to the large language model for inference calculation through the selected model service interface, and is automatically converted into structured and easy-to-read text content by the output parser (StrOutputParser) for front-end interface display.

[0100] The intelligent system for assisting multi-step genomic data analysis provided by the embodiment can guarantee data security and reduce dependence on external networks by configuring a service interface for locally deploying a large model; can standardize interaction logic by constructing a dialogue template containing system task instructions and typical questions and answers, so that the large model can more accurately understand the task; and can automatically convert inference results into text by means of an output parser, thereby improving the efficiency of generating answers, ensuring that the answer format is standard and easy to read, and providing users with a more efficient, accurate and demand-oriented question and answer experience.

[0101] In some optional embodiments, the analysis plan file generated in the background is read, and an analysis interpretation text is generated for the analysis plan file, including:

[0102] Step a4, reading the analysis plan file generated by the workflow intelligent agent 2 and extracting file parameters therefrom, the file parameters including: task step description, input file, output file, and used tool.

[0103] Specifically, the dialogue intelligent agent 1 can read the analysis plan file generated by the workflow intelligent agent 2, which is usually a JSON format file, and extract information such as task step description, input file, output file, and used tool.

[0104] Step a5, retrieving a script file for backend task execution and parsing commands and parameters in the script file.

[0105] Specifically, a shell script file (such as step_01.sh) for backend task execution is retrieved, and specific commands and parameters in the script are parsed.

[0106] Step a6, analyzing the user's question, execution plan, and current task step based on the file parameters and the commands and parameters in the script file to generate an analysis interpretation text.

[0107] Specifically, the current question of the user, the execution plan, and the latest step are comprehensively analyzed to automatically generate an analysis interpretation text that is easy to understand, and real-time execution status feedback and professional problem solving are provided.

[0108] In addition to the basic question and answer and analysis functions, the dialogue intelligent agent 1 also provides automatic generation of analysis reports and management mechanism of historical interaction records, and the specific implementation is as follows:

[0109] (1) After the user completes the task, ChatAgent automatically enters the report generation mode, the specific process includes: the system automatically loads all execution result files related to the current analysis task from the predefined output directory, including task plan file (PLAN.json), intermediate debugging output file (DEBUG_Output.json) and final result file (such as result.json); automatically parse the step description and expected output file information in the task plan, and compare the actual generated file with the planned expectation; combine the actual generated file content and the task execution process, ChatAgent constructs a standardized Markdown format report Prompt, automatically calls large language model to generate a task execution analysis report with clear structure and complete content; the report content includes task plan overview, key execution step summary, comparison analysis of expected generated file and actual generated file, specific presentation of each generated file content and intuitive display of result image; the automatically generated report is saved in Markdown format in the specified output directory (for example. / output / {task number}_report.md), for user to download, view or further edit.

[0110] (2) ChatAgent has a special historical interaction record management mechanism, the specific process includes: automatically create or call existing interaction history record folder (such as. / history / {task number} / ), automatically store user asking (asking) and agent response (response) in each interaction process; the interaction history record is saved in chat.json file in JSON format, and has automatic incremental update function; users can call the latest interaction record (default 10) at any time to facilitate analysis and decision-making and background understanding of subsequent questions; when generating analysis report, ChatAgent automatically integrates the final generated Markdown report content into the execution record file (execute.json) of the interaction history, ensuring the unity and integrity of the whole task information.

[0111] ChatAgent not only can real-time and efficiently carry out user interaction and knowledge enhancement professional answer, but also has automatic report generation ability and perfect history record management mechanism, greatly improves the execution convenience, result traceability and information integrity of group genetic analysis task, significantly improves the overall intelligence level and user interaction experience of the system, has good practical value and application prospect.

[0112] The intelligent system for assisting multi-step genomic data analysis provided by the embodiment extracts key file parameters from the analysis plan file, retrieves script commands and parameters executed by the analysis backend, comprehensively analyzes the user questions, execution plan and current task steps, and generates analysis interpretation text to enable the user to more clearly and intuitively understand the analysis plan and task execution, effectively reducing the difficulty for the user to understand professional analysis content and improving the user's understanding of the system analysis process.

[0113] In some optional embodiments, the workflow intelligent agent 2 includes a plan intelligent agent, a task intelligent agent, a correction intelligent agent, and a check intelligent agent.

[0114] Generating a structured task execution plan according to the to-be-executed analysis task, and executing the to-be-executed analysis task according to the structured task execution plan, includes:

[0115] Step b1, using the plan intelligent agent to obtain the to-be-executed analysis task, and searching for related knowledge from the vector knowledge base in the genomic field, splitting the to-be-executed analysis task into specific steps and executing them in order.

[0116] Specifically, as shown in Figure 3 the process diagram for processing data by the workflow intelligent agent 2, the Execute Agent automatically completes the task planning and sub-task execution process based on the LangChain framework and the pre-defined Prompt template. The task planning process includes: in the initialization stage, the interface configuration is connected to the large language model, supporting external model interfaces (such as the GPT-4 model of OpenAI) or locally deployed open source models (such as the Llama4, Phi-3, Mistral, Deepseek-R1, etc. models deployed by Ollama), realizing completely localized analysis task reasoning and execution, supporting completely localized large model deployment (such as using Ollama to deploy open source models), realizing the efficiency, stability and data security of task analysis; the Execute Agent automatically reads the pre-defined tool description file (.config), establishes a tool set suitable for the current task, and generates a special task planning Prompt template using the tool set, including task objectives, data descriptions and a list of available tools; the user input task objectives, uploaded data file information and tool set are integrated to generate a complete Prompt text, which is submitted to the large language model service for reasoning, and a structured task execution plan containing step descriptions, input data paths, output data paths and analysis tools is automatically generated; the generated task plan is automatically stored in the specified output directory in JSON format for subsequent task execution module calling and user front-end viewing.

[0117] Step b2, during the sequential execution of the specific steps, the task agent searches the tool library and tool knowledge base for relevant knowledge to select the target tool, and uses the target tool to generate scripts corresponding to each specific step.

[0118] Specifically, the sub-task execution process specifically includes: automatically parsing each specific step in the task plan, automatically calling the TASK agent to generate the execution command of the corresponding step according to the step description, input data and specified tool, and saving it in the form of a shell script (such as Step_01.sh); when the system is configured to allow execution (executor = True), the ExecuteAgent automatically calls the system shell environment to execute the generated shell script to complete the specific analysis step; if it is configured not to execute (executor = False), the system only generates the shell script for the user to manually confirm before manually or automatically executing later.

[0119] Step b3, sequentially execute the scripts corresponding to each specific step, and loop the error correction checking process until the checking results of all scripts meet the preset standard, the error correction checking process includes: if the execution is not successful, the error correction agent is used to modify the code according to the execution result and execute again until the execution is successful or the number of repeated executions exceeds the repetition threshold, the checking agent is used to check the execution result, and if the checking result does not meet the preset standard, the error correction agent is used to modify the code according to the execution result.

[0120] Specifically, after the execution of each step is completed, the execution result (including standard output and error output) is automatically checked, and a special debugging agent (DEBUG Agent) is called to intelligently analyze and diagnose the execution status; the system automatically supports the error retry mechanism of task execution, and according to the preset repetition threshold (such as 5 times by default), the task step is automatically reattempted when the task step fails, until the task is successfully executed or the maximum number of retries is reached; the intermediate input, output and execution status data of each execution step are stored in the output directory of the system in real time, realizing accurate tracking and recovery of the execution status of the task.

[0121] In some optional embodiments, if the execution is not successful, the error correction agent is used to modify the code according to the execution result and execute again, which includes:

[0122] If the execution is not successful, the error correction agent analyzes the reason for the unsuccessful execution based on the execution result; based on the unsuccessful reason, the error correction agent searches the tool library and tool knowledge base for relevant knowledge, and modifies the code based on the relevant knowledge and executes again.

[0123] Specifically, after the execution of each sub-task is completed, if an execution exception or error is found, the system will automatically give specific error reason analysis and error correction suggestions.

[0124] During the task execution, the workflow agent 2 supports real-time interruption and recovery mechanism, including: real-time monitoring of the task execution suspension request sent by the user through the front-end interaction module, after detecting the task suspension request, automatically and safely terminating the current task execution, keeping the execution progress and intermediate state data, and recording the current execution log and intermediate result file; the user can modify the task step content, execution instruction or parameter setting based on the stored interruption state file and log, and then reactivate the task execution process, which can automatically recover the task execution from the interruption point, ensuring the continuity and controllability of the task execution.

[0125] The workflow agent 2 has the functions of task automatic planning, automatic execution control, error automatic diagnosis and correction, knowledge dynamic enhancement, and task interruption and recovery, which can significantly improve the automation level, reliability and use flexibility of the group genetic analysis workflow.

[0126] The intelligent system for assisting multi-step genomic data analysis provided in the embodiment can accurately split the to-be-executed analysis task into specific steps and execute them in order by acquiring knowledge from the genomic field vector knowledge base through the planning agent, ensuring the professionalism and orderliness of the task execution; the task agent selects the target tool and generates the corresponding script by means of the tool library and the tool knowledge base, realizing efficient adaptation of the tool and automatic generation of the script; the error correction agent can modify the code and repeat the execution in the case of unsuccessful execution, and the checking agent strictly checks the execution result, and the error correction and checking mechanism formed by the two can timely find and solve the problems in the task execution, ensuring that all script execution results meet the preset standards, thereby improving the accuracy and reliability of the entire analysis task execution.

[0127] In some optional embodiments, the configuration file and the script are generated based on the user request, the model is built, and the analysis result corresponding to the user request is generated in combination with the workflow agent 2, including:

[0128] Step c1, based on the user request, the task target and the data list are analyzed, and the parameter agent is called to generate a standardized parameter set of the group genetic analysis tool.

[0129] Specifically, as Figure 4As shown, it is a schematic diagram of the process of modeling analysis agent 3 processing data. After receiving a user request, modeling analysis agent 3 immediately initializes the environment and performs model configuration, including: application programming interface (API) key, model service address and model type configuration. Analysis Agent supports multiple model configuration methods, including but not limited to external interface services (such as OpenAI's GPT-4o model) and locally deployed models (such as open source models Deepseek-R1, Qwen3-235B deployed by Ollama), thereby realizing completely offline analysis task execution; Analysis Agent has built-in special Prompt templates to realize standardized task reasoning and parameter generation, ensuring the efficiency and repeatability of analysis.

[0130] Based on the user request, first analyze the current task target and data list, and automatically call the parameter agent to generate a standardized parameter set for the population genetic analysis tool.

[0131] Step c2, based on the standardized parameter set, call the configuration file generation agent to generate the standard configuration file.

[0132] Specifically, Analysis Agent automatically calls a special configuration file generation agent (fastsimcoal agent) to generate the standard configuration file required for analysis based on the standardized parameter set, including TPL file (simulation.tpl), EST file (simulation.est) and PAR file (simulation.par). All files are automatically stored in the specified output directory.

[0133] Step c3, call the script generation agent to generate the script file for executing the population genetic modeling tool, and establish the execution model corresponding to the user request by executing the script file.

[0134] Specifically, Analysis Agent calls a special script generation agent (script_agent) to automatically generate a standard Linux shell script file (run_ana.sh) for executing the population genetic modeling tool. The script explicitly calls the generated TPL, EST and PAR configuration files to ensure fast and error-free execution of the analysis task.

[0135] Step c4, use the execution model and standard configuration file to generate the analysis result corresponding to the user request.

[0136] Specifically, in the process of processing user requests, the execution model and standard configuration file are used to generate the corresponding analysis results.

[0137] The intelligent system for assisting multi-step genomic data analysis provided by the embodiment is based on accurate analysis of task targets and data lists based on user requests, generates a standardized parameter set with the help of a parameter agent, generates an agent and a script agent through a configuration file, automatically generates a standard configuration file and an execution script, efficiently constructs an execution model, and quickly generates an analysis result meeting user needs in combination with a workflow agent 2, thereby improving the automation and efficiency of analysis task execution.

[0138] In some optional embodiments, the modeling analysis agent 3 is further used for interpreting and analyzing the results, including:

[0139] Step c5, reading the analysis plan file generated by the workflow agent 2 and extracting the latest execution step therefrom.

[0140] Specifically, the task analysis plan (PLAN) file and the latest execution step (shell script) generated by the background workflow agent 2 are automatically read to comprehensively grasp the latest progress status of the current analysis.

[0141] Step c6, obtaining relevant knowledge based on user requests with a knowledge enhancement module.

[0142] Step c7, generating a result interpretation text and a model analysis explanation text corresponding to the user request according to the latest execution step and the relevant knowledge.

[0143] Specifically, intuitive and popular result interpretation and model analysis explanation texts are automatically generated according to the questions raised by the user and the current execution status in combination with the real-time retrieval results of the knowledge enhancement mechanism, and are fed back to the user in real time. The AnalysisAgent realizes automatic population genetic data modeling, parameter optimization and configuration file generation, as well as real-time interpretation and interactive analysis of results.

[0144] The intelligent system for assisting multi-step genomic data analysis provided by the embodiment can automatically read the result file after the completion of the task execution, call appropriate modeling tools for reasoning analysis, and output in the form of a visual chart or a statistical summary, thereby greatly improving the data interpretation and model fitting efficiency of researchers. In combination with the relevant knowledge obtained by the knowledge enhancement module, a result interpretation and model analysis explanation text is generated, so that the user can more clearly and deeply understand the analysis result, thereby enhancing the interpretability and ease of use of the analysis result and assisting the user in better grasping the core information of the genomic data analysis.

[0145] In a specific embodiment, the intelligent system for assisting multi-step genomic data analysis provided by the embodiment is used for RNA-seq differential expression (DE), and the content corresponding to each agent can be:

[0146] (1) The user in the dialogue agent asks: "What are the steps to do a good RNA-seq differential expression, and what is the significance of each step?"

[0147] (2) The analysis task to be performed in the workflow agent is: compare the gene expression differences between "treatment group vs control group", get the list of differential genes, visualization graphs (PCA, MA / volcano plot, enrichment bubble plot), and generate a Markdown report. In addition to this text content, the FASTQ raw sequencing files (single-end or double-end, multiple samples), species reference genome FASTA and gene annotation GTF, sample_sheet.csv (sample name, grouping, batch, etc. metadata) are uploaded synchronously. The internal processing process of the workflow agent includes:

[0148] (2.1) Plan generation (Plan Agent): The natural language "do DE and generate a report" is disassembled into the following PLAN (excerpt) and saved as plan.json: ① Quality control and pruning (fastp / FastQC → generate quality control report and pruned FASTQ); ② Quantification (both paths are supported) (alignment path: STAR / HISAT2 → BAM → featureCounts count table, quasi-mapping path: Salmon / Kallisto → transcript-level quantification → tximport summarized to gene level); ③ Normalization and differential analysis (use DESeq2 (or edgeR) for standardization, differential test, output differential gene table (with log2FC, p-value, FDR)); ④ Enrichment analysis (GO / KEGG / GSEA); ⑤ Visualization (PCA, sample clustering heatmap, MA / volcano plot, enrichment bubble plot). PLAN.json will specify the input / output / tools / key parameters of each step, such as "featureCounts-a annot.gtf-g gene_id-T16-p", and write the product path (for easy breakpoint recovery and review).

[0149] (2.2) Execution and Debugging (Task / Debug / Check): Task Agent: Refer to RAG (RAG has relevant operation instructions or cases) to automatically generate step_01.sh…step_05.sh and execute them in order; Debug Agent: Read the script and execute all contents, judge correctly and move to check agent, if error occurs (such as GTF chromosome naming and FASTA inconsistency), automatically identify common errors such as "chr / no chr prefix mismatch", propose standardization scheme (unify prefix, convert script) and retry, rewrite new sh file and execute; Check Agent: Judge whether the output file exists and is not empty. Avoid debug agent error judgment execution success. If the target file exists and is not empty, jump to the next step. If it is empty, back to Debug Agent and explain the reason.

[0150] (2.3) Interpretation and reporting (Q&A): When all tasks are completed, use Chat Agent to generate a project report according to all results. The system summarizes and generates output / <session_id>_report.md, including task execution success rate, expected file completion rate, meaning of each file, and generates pictures for display.

[0151] During execution, new questions can be input at any time in the dialog box of "Chat Agent", such as: "Why are these two samples outliers?" (The system will quote PCA, QC indicators and design matrix explanation), "If it is single-end sequencing, how to change the parameters?" (The system uses RAG to call out tool manual fragments to give replacement commands).

[0152] (3) Modeling analysis agent is usually used only in tasks similar to population genetics that need to be modeled, which is not needed in this embodiment. Specifically, assuming that the above task is the analysis of population genetics, complex analysis data is obtained, and Analysis Agent integrates the data to obtain a priori data.

[0153] At the same time, input in the dialog box of "Analysis Agent": "The current data involves multiple sample populations, I want to establish a separation model based on the existing data and analysis", and upload necessary files, such as: FASTQ raw sequencing file (single-end or double-end, multiple samples); species reference genome FASTA and gene annotation GTF; sample_sheet.csv (sample name, grouping, batch, etc. Metadata).

[0154] Analysis agent will build a model according to these data, such as when did the Europeans and Asians walk out of Africa? When did the Asians and Europeans separate? When did the northern and southern Chinese separate and settle down, get the analysis result diagram as shown in Figure 5 Figure 5 Only as an example, but not limited.

[0155] In this embodiment, an intelligent method for assisting multi-step genomic data analysis is provided, which can be used in the intelligent system for assisting multi-step genomic data analysis described above, although the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in an order different from here. Figure 6 is a flowchart of the intelligent method for assisting multi-step genomic data analysis according to an embodiment of the present application, as shown in Figure 6 The flowchart includes the following steps:

[0156] Step S101, generating a corresponding answer according to a user question, or reading an analysis plan file generated by a workflow agent, and generating an analysis interpretation text for the analysis plan file, the analysis plan file including a structured task execution plan generated by the workflow agent.

[0157] Step S102, generating a structured task execution plan according to a to-be-executed analysis task, and executing the to-be-executed analysis task according to the structured task execution plan.

[0158] Step S103, generating a configuration file and a script based on a user request, building a model, and combining the workflow agent to generate an analysis result corresponding to the user request.

[0159] Further description of each of the above steps is the same as the embodiment of the corresponding module described above, and will not be repeated here.

[0160] The embodiment of the present application also provides a computer device having the intelligent system for assisting multi-step genomic data analysis described above. Figure 1

[0161] Please refer to Figure 7 , Figure 7 is a structural diagram of a computer device provided by an optional embodiment of the present application, as shown in Figure 7 ​​As shown, the computer device includes one or more processors 10, memory 20, and interfaces 30 for the various components to communicate with one another. The various components communicate through one or more buses, and can be mounted on a common motherboard or in other manners as appropriate. The processor 10 can execute instructions, for example, stored in the memory 20 to display graphical information for a GUI on an external input / output device, such as a display device coupled to the interface. In some optional implementations, multiple processors and / or multiple buses can be employed as appropriate, such as about the memory 20. Also, multiple computers can be connected, with each computer providing portions of the necessary operations (e.g., as a server array or a group of blade servers, or multiple processors). Figure 7 The processor 10 is taken as an example in the embodiments.

[0162] The processor 10 can be a central processing unit, a network processor, or a combination thereof. The processor 10 can further include a hardware chip. The hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device can be a complex programmable logic device, a field programmable logic device, a general array logic, or any combination thereof.

[0163] The memory 20 stores instructions that are executable by the at least one processor 10, so as to enable the at least one processor 10 to perform the method shown in the above embodiments.

[0164] The memory 20 can include a program region and a data region. The program region can store an operating system and an application program required by at least one function. The data region can store data created according to the use of the computer device, and the like. In addition, the memory 20 can include a high-speed random access memory, and can further include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some optional implementations, the memory 20 can optionally include a memory that is remotely arranged with respect to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0165] The memory 20 can include a volatile memory, such as a random access memory, and can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid state disk. The memory 20 can further include a combination of the above-mentioned kinds of memories.

[0166] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0167] While embodiments of the application have been described in connection with the preferred embodiments of the various figures, those of ordinary skill in the art will appreciate that various modifications and changes can be made without departing from the spirit and scope of the application, and that such modifications and changes fall within the scope of the appended claims.

Claims

1. An intelligent system to assist multi-step genomic data analysis, characterized in that, The system comprises a multi-agent cooperation module, the multi-agent cooperation module comprising a dialogue agent, a workflow agent, and a modeling analysis agent, wherein The dialogue agent is configured to generate a corresponding answer according to a user question, or read an analysis plan file generated by the workflow agent and generate an analysis interpretation text for the analysis plan file, the analysis plan file comprising a structured task execution plan generated by the workflow agent; The workflow agent is configured to generate a structured task execution plan according to a to-be-executed analysis task, and execute the to-be-executed analysis task according to the structured task execution plan; The modeling analysis agent is configured to generate a configuration file and a script based on a user request, build a model, and generate an analysis result corresponding to the user request in combination with the workflow agent.

2. The system of claim 1, wherein, The system further comprises a front-end interaction module, a back-end processing module, a knowledge enhancement module, and a memory module, wherein The front-end interaction module is configured to create a project and a session, configure parameters, obtain a user question or a user request, and view an answer or an analysis result; The back-end processing module is in communication connection with the front-end interaction module and is configured to dispatch the multi-agent cooperation module to execute a corresponding task according to the user question or the user request, and feed back a task execution status and an analysis result to the front-end interaction module; The knowledge enhancement module is configured to generate a vector knowledge base in the field of genomics, perform semantic retrieval on the vector knowledge base to obtain a knowledge fragment according to the user question or the user request, and add the knowledge fragment to the multi-agent cooperation module to execute a corresponding task; The memory module is configured to store historical user interaction information for a long time, and assist the multi-agent cooperation module in understanding a user intent and generating a result meeting a user demand according to the historical user interaction information.

3. The system of claim 1, wherein, The generation of a corresponding answer according to a user question comprises: configuring a service interface by using a locally deployed large model interface service address, and selecting a target large language model; constructing a dialogue template comprising a system task instruction, a typical question and a question and answer example, and a real-time user input; obtaining a user question based on the dialogue template, submitting the user question to the target large language model for inference calculation based on the service interface, and automatically converting a text content output by an output parser into a text content as a corresponding answer to the user question.

4. The system of claim 1, wherein, The reading of an analysis plan file generated in the background and the generation of an analysis interpretation text for the analysis plan file comprise: reading an analysis plan file generated by the workflow agent, and extracting file parameters therefrom, the file parameters comprising a task step description, an input file, an output file, and a tool used; retrieving a script file for back-end task execution, and parsing commands and parameters in the script file; analyzing a user question, an execution plan, and a current task step based on the file parameters and the commands and parameters in the script file, and generating an analysis interpretation text.

5. The system of claim 1, wherein, The workflow agent comprises a plan agent, a task agent, a correction agent, and a check agent. The generation of a structured task execution plan according to a to-be-executed analysis task and the execution of the to-be-executed analysis task according to the structured task execution plan comprise: The planned agent obtains an analysis task to be executed, and finds relevant knowledge from a vector knowledge base in the field of genomics, splits the analysis task to be executed into specific steps, and executes the specific steps in sequence; During the execution of the specific steps in sequence, the task agent finds relevant knowledge from the tool library and the tool knowledge base to select a target tool, and generates a script corresponding to each specific step using the target tool; The scripts corresponding to each specific step are executed in sequence, and a correction checking process is performed in a loop until the checking results of all scripts meet the preset standard, the correction checking process including: if the execution is unsuccessful, the correction agent modifies the code according to the execution result and executes again until the execution is successful or the number of repeated executions exceeds a threshold value, the checking agent checks the execution result, and if the checking result does not meet the preset standard, the process of modifying the code according to the execution result by the correction agent is returned.

6. The system of claim 5, wherein, If the execution is unsuccessful, the correction agent modifies the code according to the execution result and executes again, including: If the execution is unsuccessful, the correction agent analyzes the reason for the unsuccessful execution based on the execution result; Based on the reason for the unsuccessful execution, the correction agent finds relevant knowledge from the tool library and the tool knowledge base, and modifies the code based on the relevant knowledge and executes again.

7. The system of claim 1, wherein, Based on the user request, a configuration file and a script are generated, a model is built, and the analysis result corresponding to the user request is generated in combination with the workflow agent, including: Based on the user request, the task target and the data list are parsed, and the parameter agent is called to generate a standardized parameter set of the population genetic analysis tool; Based on the standardized parameter set, the configuration file generation agent is called to generate a standard configuration file; The script generation agent is called to generate a script file for executing the population genetic modeling tool, and an execution model corresponding to the user request is established by executing the script file; The analysis result corresponding to the user request is generated using the execution model and the standard configuration file.

8. The system of claim 7, wherein, The modeling analysis agent is also used to interpret and analyze the results, including: The analysis plan file generated by the workflow agent is read, and the latest execution step is extracted therefrom; Based on the user request, the knowledge enhancement module is used to obtain relevant knowledge; Based on the latest execution step and the relevant knowledge, a result interpretation text and a model analysis explanation text corresponding to the user request are generated.

9. An intelligent method of assisting multi-step genomic data analysis, characterized in that, The method is applied to the intelligent system for assisting multi-step genomic data analysis according to any one of claims 1-8, and the method includes: According to the user's question, a corresponding answer is generated, or the analysis plan file generated by the workflow agent is read, and an analysis interpretation text is generated for the analysis plan file, the analysis plan file including the structured task execution plan generated by the workflow agent; A structured task execution plan is generated based on the analysis task to be executed, and the analysis task to be executed is executed according to the structured task execution plan; Based on the user request, a configuration file and a script are generated, a model is built, and the analysis result corresponding to the user request is generated in combination with the workflow agent.

10. A computer device, comprising: Including: A memory and a processor, which are connected to each other in communication, the memory storing computer instructions, and the processor executing the computer instructions to perform the method of claim 9.

Images