A low-code multi-group bioinformatics analysis method and analysis system

Abstract

The application discloses a low-code multi-omics bioinformatics analysis method and system. The analysis method supports a dual-path creation analysis process of drag-and-drop visual construction and natural language intelligent generation. The intelligent process generation adopts a dual-channel hybrid strategy of AI generation priority and rule matching fallback. The task monitoring adopts a dual-state determination mechanism combining process survival detection and log marking detection. The unified architecture supports full-process analysis of multiple types of omics data, and a process template sharing and reuse system with an audit mechanism. The analysis system includes a visual process arrangement module, an intelligent process generation module, a task execution and state monitoring module, a multi-omics unified architecture support module, and an analysis template sharing module. The application can effectively reduce the technical threshold of bioinformatics analysis, improve analysis efficiency and process reusability, and realize one-stop, visual, and traceable low-code analysis of multi-omics data.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics data analysis and low-code workflow orchestration technology, and in particular to a low-code multi-omics bioinformatics analysis method and system, which can be applied to the standardization, visualization, and reusable analysis of multi-omics data such as conventional transcriptome sequencing, single-cell transcriptome sequencing, chromatin accessibility sequencing, and chromatin immunoprecipitation sequencing. Background Technology

[0002] With the rapid development of high-throughput sequencing technology and the continuous decline in sequencing costs, biomedical research has fully entered the era of multi-omics big data. Conventional transcriptome sequencing (Bulk RNA-seq), single-cell RNA sequencing (scRNA-seq), chromatin accessibility sequencing (ATAC-seq), and chromatin immunoprecipitation sequencing (ChIP-seq / CUT&Tag) are widely used in core areas such as tumor heterogeneity research, developmental biology, and drug target discovery. However, the rapid increase in sequencing data output speed has widened the gap between the data analysis capabilities of biomedical researchers. Current bioinformatics analysis models mainly fall into three categories, each with corresponding application limitations: 1. Traditional Command-Line Analysis Mode: The vast majority of bioinformatics analyses rely on Linux command-line tools and Shell, R, and Python scripting, requiring researchers to have both a biological background and programming skills. The analysis workflow requires manually connecting multiple tools, adapting data formats, and debugging analysis parameters, resulting in a high barrier to entry and a long debugging cycle. The input and output formats of different tools are fragmented, data flow relies on manual management, is prone to errors, and is difficult to reproduce. Long-running analysis tasks lack refined methods for tracking operational status, requiring regular manual log checks, and it is difficult to quickly locate the problem node after a failure. Furthermore, the analysis protocols lack a standardized and consolidation mechanism, leading to widespread duplication of similar workflows within research groups, resulting in a large amount of repetitive work.

[0003] 2. Open-source bioinformatics workflow platforms (represented by Galaxy): These platforms provide graphical tool configuration and workflow management capabilities, lowering the operational threshold to some extent, but still have significant shortcomings: tools and workflows are configured in list and form formats, lacking an intuitive drag-and-drop workflow arrangement canvas; the task scheduling mechanism based on job queues can only perceive the overall status of the task queue and cannot provide feedback on the real-time running status of individual analysis nodes; it does not support natural language-driven intelligent workflow generation, and workflow construction still requires manual configuration node by node; the platform deployment and tool integration are highly complex and require professional operation and maintenance personnel.

[0004] 3. Commercial workflow management platforms (represented by Nextflow Tower): These platforms have strong distributed task scheduling capabilities, but the process definition relies on a proprietary DSL scripting language, and users still need to have coding skills; they do not provide the ability to automatically generate analysis processes from natural language input; the interface is designed for professional bioinformatics engineers, with a steep learning curve, and biologists without a programming background find it difficult to use independently.

[0005] Furthermore, existing technologies are all designed around software systems / platforms, with accompanying operating procedures serving only as supplementary instructions. They lack a complete, standardized, and practical end-to-end analytical methodology. Existing systems are mostly designed for single-omics data, requiring different omics types to be analyzed in different software or platforms, lacking a unified entry point and consistent user experience. The learning and switching costs for cross-omics analysis are also high. Overall, there is currently a lack of a one-stop bioinformatics analysis solution that is centered on standardized analytical methods, supported by a complete system for end-to-end implementation, designed for users with no programming experience, covers multiple omics scenarios, possesses both visual orchestration and intelligent generation capabilities, and supports refined status monitoring and standardized template reuse. Summary of the Invention

[0006] The purpose of this invention is to provide a low-code method and system for multi-omics bioinformatics analysis. This invention enables drag-and-drop construction and intelligent natural language generation in a dual-path workflow, precise task status monitoring, and workflow template reuse, thereby lowering the technical threshold for bioinformatics analysis, improving analysis efficiency and workflow reusability, and achieving one-stop, visualized, and traceable low-code analysis of multi-omics data.

[0007] The technical solution of this invention: a low-code multi-group bioinformatics analysis method, comprising the following steps: Step S1: Create a visualization canvas and receive drag and drop operations and connection operations of bioinformatics analysis tool nodes. Extract the unique identifier prefix of the analysis node to be connected for group matching, block the connection that does not conform to the prefix grouping rule, and generate process configuration data; or receive natural language description, use a dual-channel strategy of AI generation priority and rule matching backoff to generate process configuration data, and render it as analysis tool nodes and connections on the canvas. Step S2: Receive user adjustments to parameters and input file selections for each analysis node, and save the complete process execution configuration data; Step S3: Asynchronously start the bioinformatics analysis task according to the process configuration data. The task running status is obtained in real time through a dual status determination mechanism that combines process survival detection and log marking detection, and the status data is sent back to the canvas to update the node status indication. Step S4: When the task is determined to be successful, output multiple sets of bioinformatics analysis results; save the analysis process verified by the task as a template, review it and include it in the shared template library for subsequent manual process construction and intelligent process generation.

[0008] The aforementioned low-code multi-group bioinformatics analysis method, in step S1, receives the user's natural language description and automatically generates process configuration data using a dual-channel strategy of AI-driven generation and rule-matching fallback. Specifically, this includes: S11. Receive user natural language instructions and prioritize calling the RAG intelligent generation channel: retrieve relevant document fragments from the pre-built bioinformatics analysis knowledge base as context, call the large language model to generate a structured process plan and output confidence scores; S12. Determine whether the confidence score is higher than the preset threshold. If so, convert the structured process plan into renderable process configuration data; otherwise, switch to the rule matching channel. S13. The rule matching channel identifies the process type corresponding to the user's instruction through keyword matching, performs weighted scoring matching on the templates in the shared template library to select the best matching template, and extracts the parameter information in the user's instruction. After normalization, it is injected into the template configuration to generate the final process configuration data.

[0009] In the aforementioned low-code multi-group bioinformatics analysis method, the specific process of the dual-state determination mechanism in step S3 is as follows: S31. During the task startup phase, the bioinformatics analysis tool is started in an independent asynchronous process session, and the corresponding process identifier and log file path are recorded and persistently stored. S32. When receiving a status query request, first check if the corresponding log file exists. If it does not exist, return the inactive status. S33. If the log file exists, verify the liveness status of the corresponding process through the system process query interface; S34. If the process is in a running state, return directly to the running state; S35. If the process has terminated, read the complete contents of the corresponding log file and search for the predefined task completion marker string; if the completion marker string is detected, return a successful running status; otherwise, return a failed running status.

[0010] In the aforementioned low-code multi-group bioinformatics analysis method, in step S4, the analysis process information that has completed the running verification in the canvas is serialized and saved as a user's private template in a standardized format; after the user submits a template release application, the template approved by the administrator is included in the platform's shared template library; the shared template library also serves as a candidate data source for intelligent process generation, providing an executable process benchmark for dual-channel generation processes.

[0011] A low-code, multi-group bioinformatics analysis system includes: The visual workflow orchestration module provides a web-based visual orchestration canvas, enabling drag-and-drop analysis workflow setup, connection validity verification, operation history management, node parameter configuration and file selection, and outputting workflow configuration data. The intelligent process generation module communicates with the visual process orchestration module. It generates process configuration data based on the user's natural language description and transmits the process configuration data to the visual process orchestration module, which then renders it onto the canvas. The task execution and status monitoring module is bidirectionally connected to the visualization process orchestration module. It is used to asynchronously start analysis tasks based on process configuration data, obtain task running status through a dual status determination mechanism, and send status data back to the visualization process orchestration module to update canvas node status indicators. The multi-omics unified architecture support module provides unified underlying operation support for the visualization process orchestration module, intelligent process generation module, task execution and status monitoring module, and analysis template sharing module. It is used to organize and operate the support system with unified tools to support the full-process analysis of multiple omics data types. The analysis template sharing module communicates with the visualization process orchestration module and the intelligent process generation module respectively. It is used to realize the template-based saving, review and release, and sharing and reuse of analysis processes, and to provide a shared template library for intelligent process generation.

[0012] The aforementioned low-code multi-group bioinformatics analysis system includes a visualization workflow orchestration module comprising a node menu unit, a canvas editing unit, a history management unit, and an interactive zoom unit. The node menu unit is used to construct a hierarchical classification tree according to function and omics type, and displays multi-omics analysis tool nodes including quality control, sequence alignment, quantitative analysis and differential analysis. Each node has preset default analysis parameters and is managed uniformly through a unique identifier. The canvas editing unit is built on a visual canvas rendering engine and is used to render analysis nodes with status indicators. It receives drag-and-drop connection operations through the input and output endpoints at the end of the node and automatically blocks invalid connections in the same group through pre-grouping verification rules. The history management unit adopts a command mode to build a stack-based operation record system, which is used to store and manage all canvas editing operations and supports multi-step undo and redo. The interactive scaling unit integrates a canvas interactive control component to realize multi-level scaling and panning of the canvas, and synchronizes the scaling factor to the canvas rendering engine.

[0013] The aforementioned low-code multi-group bioinformatics analysis system, wherein the intelligent process generation module adopts a dual-channel hybrid strategy of AI generation priority and rule matching fallback, including a rule matching unit and a RAG intelligent generation unit; The rule matching unit is used to identify the process type input by the user through keyword matching, perform multi-dimensional weighted scoring on each template in the shared template library to select the best matching result, and inject the parameters input by the user into the corresponding node configuration through parameter extraction and normalization mapping, and output process configuration data. The RAG intelligent generation unit is used to construct a retrieval-generation link. It retrieves relevant document fragments from the bioinformatics analysis knowledge base as context through a keyword retrieval device, calls a large language model, and generates a structured process plan containing process intent, template selection, node parameters, and confidence scores through structured output constraints. The intelligent process generation module is preset with a confidence threshold. When the confidence of the AI-generated result is lower than the threshold, it automatically switches to the rule matching unit to output process configuration data.

[0014] The aforementioned low-code multi-group bioinformatics analysis system includes a task execution and status monitoring module comprising a task initiation unit and a dual status detection unit. The task initiation unit is used to receive parameter configuration and input file path from the front end, construct bioinformatics analysis command line, start the analysis tool in an independent asynchronous process, and persistently store runtime metadata. The dual-state detection unit adopts a dual-determination logic that combines process liveness detection and log marking detection, with log existence verification as the first judgment: first, it checks whether the task log file exists. If the log does not exist, it is determined to be in the not started state; if the log exists, it checks the process liveness status. If the process is alive, it is determined to be in the running state; if the process has terminated, it is determined to be in the running state by marking the log completion status; a total of four task states are output: not started, running, running successfully, and running failed. The front end obtains real-time status data through a timed polling mechanism, which drives the status indicators of the canvas nodes to update and display synchronously.

[0015] The aforementioned low-code multi-omics bioinformatics analysis system, wherein the multi-omics unified architecture support module covers four types of omics analysis: conventional transcriptome sequencing, single-cell RNA sequencing, chromatin accessibility sequencing, and chromatin immunoprecipitation sequencing. The backend adopts a modular architecture, with each omics type independently registering functional interfaces while sharing a unified process management framework, file management mechanism, and status detection logic; The visualization workflow orchestration module is based on a multi-omics unified architecture support module. It organizes all omics analysis tool nodes with a unified hierarchical classification tree and uses a unified parameter component template to render configuration forms, providing a consistent end-to-end operation experience for all omics analyses.

[0016] The aforementioned low-code multi-group bioinformatics analysis system includes an analysis template sharing module comprising a template saving unit, a template review unit, and a template loading unit. The template saving unit is used to serialize and save the process configuration data built in the canvas into a user private template in a standardized format; The template review unit is used to review the templates submitted by users. Templates that pass the review are changed to a publicly released status and included in the platform's shared template library. The template loading unit provides users with template browsing and one-click loading functions, while using the shared template library as a candidate space for the intelligent process generation module to ensure the executable nature of the intelligently generated analysis process.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention takes a standardized low-code multi-omics bioinformatics analysis method as its core, and relies on a supporting system consisting of five modules: visual workflow orchestration, intelligent workflow generation, task execution and status monitoring, unified multi-omics architecture support, and shared analysis templates. This analysis method supports the creation of analysis workflows through two paths: visual drag-and-drop construction and intelligent natural language generation. Combined with asynchronous task dual status monitoring and full-process control of workflow template review and reuse, it effectively reduces the technical threshold of bioinformatics analysis, improves the efficiency of workflow construction and analysis execution, and solves the problems of high analysis threshold, invisible status, difficult workflow reuse, and scattered multi-omics platforms in the traditional mode. It realizes standardized, visualized, and traceable low-code analysis of multi-omics data. Furthermore, this invention enhances editing tolerance and traceability while ensuring the biological rationality of the process through pre-grouping verification rules and a stack-based operation history system. It also employs a dual-channel hybrid strategy of AI-driven generation and rule-matching fallback to balance the flexibility of intelligent generation in adapting to different scenarios with the reliability of rule-matching results, ensuring the accuracy and stability of process generation. The invention overcomes the limitations of single detection methods by combining process liveness detection with log tagging detection in a two-level judgment logic, improving the accuracy and robustness of long-running task status determination. Through a modular backend registration architecture and a unified frontend component system, this invention facilitates the subsequent expansion of new omics types while ensuring consistent operational experience across different omics scenarios. Finally, this invention manages the quality of shared templates through a template review mechanism and uses the shared template library as a candidate space for intelligent process generation, further ensuring the executability and analytical standardization of automatically generated processes. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall architecture of the low-code multi-group bioinformatics analysis system described in this invention; Figure 2 This is a schematic diagram of the low-code multi-group bioinformatics analysis method of the present invention. Detailed Implementation

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the present invention.

[0020] Example 1: A low-code multi-group bioinformatics analysis system. The low-code multi-group bioinformatics analysis system described in this example adopts a front-end and back-end separated B / S architecture and can be deployed on a cloud server (such as Alibaba Cloud ECS). The front-end is based on Vue.js3, Vuex4, jsPlumb2.9.2, and panzoom8.1.0 to build a visual interactive interface. The back-end is based on the Python Flask framework with a Blueprint modular architecture to implement business logic. User, process, and template data are stored through MySQL. The intelligent process generation module is based on the LangChain framework and a large language model to achieve retrieval enhancement generation.

[0021] like Figure 1 As shown, the platform as a whole consists of a visual workflow orchestration module, a parameter configuration and management module, an intelligent workflow generation module, a task execution and status monitoring module, a multi-omics unified architecture support module, an analysis template sharing module, a file management and result browsing module, and a front-end status management and authentication authorization module. The specific implementation methods of each module are as follows: The visualization workflow orchestration module features a hierarchical classification tree on the left-hand node menu, organized by function and omics type, covering categories such as quality control, sequence alignment, quantitative analysis, differential analysis, enrichment analysis, and cluster analysis. Each analysis tool node has a unique identifier and preset default analysis parameters. The canvas editing unit uses jsPlumb to render analysis nodes with status indicators. Nodes have target input endpoints at the top and source output endpoints at the bottom. Prefix grouping verification is performed via the beforeDrop hook function, extracting the prefix part of the node ID for group matching and automatically blocking invalid connections within the same group to ensure the biological rationality of the workflow. The history management unit uses a command pattern to build a stack-based operation record system, defining an abstract base class Command and seven specific command subclasses: Add Node, Delete Node, Add Connection, Delete Connection, Move Node, Rename Node, and Paste Node. Command history is managed using a stack structure, supporting multi-step undo and redo. The interactive zoom unit uses panzoom to achieve multi-level scaling and panning of the canvas, and synchronizes the scaling factor to the canvas rendering engine to ensure the accuracy of node dragging and connection positioning.

[0022] Parameter Configuration and Management Module: When a user clicks a canvas node, the right panel dynamically renders the corresponding parameter configuration component, which includes four functional areas: a status bar, which intuitively displays the four operating states of the node using gray / blue rotation / green checkmark / red cross; a parameter form, which renders controls such as text input, drop-down selection, and numerical input according to the node type, and automatically binds them to default parameters; a file management area, which provides functions such as chunked upload, server file browsing, and management of the uploaded file list; and an operation button area, which provides operation entry points such as saving parameters, running analysis, and clearing output results.

[0023] The intelligent process generation module employs a dual-channel hybrid strategy of AI-driven generation and rule-matching fallback. The rule-matching unit identifies the process type input by the user through keyword matching and performs multi-dimensional weighted scoring on templates in the shared template library: +6 points for a matching template name, +3 points for a matching template description, +2 points for keyword weights based on the omics type and analysis steps, and +2 points for each matching node name. The best matching template is selected in descending order of scores. Then, the parameters input by the user are extracted using regular expressions, normalized, and mapped (species name → standard identifier, yes / no → boolean value, numeric string → numeric type) before being injected into the corresponding node configuration. The RAG intelligent generation unit uses the BM25 retrieval tool to recall the top-5 relevant document fragments from the bioinformatics analysis knowledge base as context. It calls a large language model and uses PydanticOutputParser to structure the output constraints, setting temperature=0 to ensure output determinism. It generates a structured process plan that includes process intent, template selection, node parameters, and confidence score. The module has a preset confidence threshold of 0.6. When the confidence score of the AI-generated result is lower than the threshold, it automatically switches to the rule matching unit to output process configuration data.

[0024] The task execution and status monitoring module includes a task startup unit and a dual-state detection unit. The task startup unit receives parameter configurations and input file paths from the front end, constructs a bioinformatics analysis command line, starts the analysis tool as an independent asynchronous process, redirects standard output and standard error to a log file, and persistently stores runtime metadata such as process PID, log path, and input / output directories in a server-side runtime information file. The dual-state detection unit uses a combined logic of process liveness detection and log flag detection: first, it checks the process's liveness status; if the process is alive, it is determined to be running; if the process has terminated, it searches for a preset completion flag string in the log file; if the flag is detected, the process is considered successful; otherwise, it is considered a failure. Including the scenario of no startup due to a missing log file, four types of task statuses are output. The front end polls the status interface at 1200ms intervals, driving the status indicators of the canvas nodes to update synchronously.

[0025] The multi-omics unified architecture support module covers four types of omics analysis: conventional transcriptome sequencing, single-cell RNA sequencing, chromatin accessibility sequencing, and chromatin immunoprecipitation sequencing. Each omics type in the backend has an independent Blueprint registration function interface, sharing a unified process management framework, file management mechanism, and status detection logic. The frontend organizes all omics analysis tool nodes with a unified hierarchical classification tree and uses a unified parameter component template to render configuration forms, providing a consistent operating experience for all omics analyses.

[0026] The template sharing module comprises a template saving unit, a template review unit, and a template loading unit. The template saving unit serializes the completed process configuration data in the canvas using standardized JSON format and saves it as a user's private template. The template review unit performs administrator review on user-submitted templates; once approved, the template is changed to a publicly released status and included in the platform's shared template library. The template loading unit provides users with template browsing and one-click loading functions, while the shared template library serves as a candidate space for the intelligent process generation module, ensuring the executability of intelligently generated processes.

[0027] The file management and results browsing module uses a three-level directory structure on the server side: {BASE_DATA_DIR} / files / admin / temp / {processId} / {NodeName} / {upload,output} / . The upload directory stores user-uploaded files, and the output directory stores analysis output results. The front end implements a server file browser in the form of a recursive directory tree, and prevents directory traversal attacks through path security verification. It supports large file chunk uploads and MD5 integrity verification. The output results are displayed in the form of a file tree, and can be downloaded after valid path verification.

[0028] Front-end state management and authentication / authorization module: The front-end uses Vuex4 for global state management, storing core states such as node parameter configurations, unique process identifiers, and single-cell intermediate data caching; it uses VueRouter to implement single-page application routing management, and the designer page supports state caching to avoid loss during switching; user authentication uses the JWT mechanism, where a token is obtained after login and automatically injected into the request header by the request interceptor, and the back-end verifies the token's validity through decorators; user passwords are stored using hashing, and the model includes permission fields to distinguish between administrator and ordinary user roles.

[0029] Example 2: A low-code multi-group bioinformatics analysis method. This example is based on the low-code multi-group bioinformatics analysis system described in Example 1, and includes the following steps: Step S1: Receive drag and drop operations of bioinformatics analysis tool nodes and connect them through the Web visualization canvas, extract the unique identifier prefix of the analysis node to be connected, perform group matching, block connections that do not conform to the prefix grouping rules, and generate process configuration data; or receive natural language descriptions, use a dual-channel strategy of AI generation priority and rule matching fallback to generate process configuration data, and render it as analysis tool nodes and connections on the canvas.

[0030] This step describes the process of generating workflow configuration data using natural language, employing a dual-channel strategy of AI-driven generation followed by rule-matching fallback. Specifically, it includes: S11. Receive user natural language instructions and prioritize calling the RAG intelligent generation channel: retrieve relevant document fragments from the pre-built bioinformatics analysis knowledge base as context, call the large language model to generate a structured process plan and output confidence scores; S12. Determine whether the confidence score is higher than the preset threshold. If so, convert the structured process plan into renderable process configuration data; otherwise, switch to the rule matching channel. S13. The rule matching channel identifies the process type corresponding to the user's instruction through keyword matching, performs weighted scoring matching on the templates in the shared template library to select the best matching template, and extracts the parameter information in the user's instruction. After normalization, it is injected into the template configuration to generate the final process configuration data.

[0031] Step S2: Receive user adjustments to parameters and input file selections for each analysis node, and save the complete process running configuration data.

[0032] Step S3: Asynchronously start the bioinformatics analysis task according to the process configuration data. The task running status is obtained in real time through a dual status determination mechanism that combines process survival detection and log marking detection, and the status data is sent back to the canvas to update the node status indication.

[0033] The specific process of the dual-state determination mechanism is as follows: S31. During the task startup phase, the bioinformatics analysis tool is started in an independent asynchronous process session, and the corresponding process identifier and log file path are recorded and persistently stored. S32. When receiving a status query request, first check if the corresponding log file exists. If it does not exist, return the inactive status. S33. If the log file exists, verify the liveness status of the corresponding process through the system process query interface; S34. If the process is in a running state, return directly to the running state; S35. If the process has terminated, read the complete contents of the corresponding log file and search for the predefined task completion marker string; if the completion marker string is detected, return a successful running status; otherwise, return a failed running status.

[0034] Step S4: When the task is determined to be successful, output multiple sets of bioinformatics analysis results; save the analysis process verified by the task as a template, review it, and include it in the shared template library for subsequent manual process construction and intelligent process generation.

[0035] The specific process of this step is as follows: the analysis process information that has been verified in the canvas is serialized and saved as a user's private template in a standardized format; after the user submits a template release application, the template approved by the administrator is included in the platform's shared template library; the shared template library also serves as a candidate data source for intelligent process generation, providing an executable process benchmark for dual-channel generation processes, and ensuring that the automatically generated analysis process complies with bioinformatics analysis specifications.

[0036] Example 3: Manual Construction and Execution of Conventional Transcriptome (BulkRNA-seq) Differential Expression Analysis Workflow. This example is a specific application of the analysis method described in Example 2 in a conventional transcriptome sequencing scenario. It focuses on the differential expression and functional enrichment analysis of BulkRNA-seq data in clinical research, and fully executes the entire workflow from visualization setup, parameter configuration, task execution, and result output. The specific steps are as follows: Step 1: Visual Workflow Setup: Users access the platform designer page via a web browser. From the BulkRNA category in the left-hand node menu, drag and drop five analysis nodes—PreProcess, Correlation, DEG, Enrichment, and Clustering—to the canvas. Drag a connection from the output endpoint of the first node to the input endpoint of the next node. The system automatically verifies the validity of the connection using node ID prefix grouping rules, sequentially completing the workflow chain from PreProcess to Correlation to DEG to Enrich to Clustering. All editing operations are automatically recorded in the history stack, allowing users to undo or redo operations at any time to correct errors.

[0037] Step 2, Parameter Configuration and File Upload: The user clicks the PreProcess node, and the corresponding configuration items are dynamically rendered in the parameter panel on the right. The user uploads the original FASTQ sequencing file through the fragment upload function, or selects existing sequencing data through the server file browser. The reference genome is selected as hg38, and the number of analysis threads is set to 4. Click the DEG node and configure the differential expression screening threshold: foldchange=2, corrected p-value threshold=0.05. The other nodes retain the platform's default parameters. After completing the configuration, save the complete workflow configuration data.

[0038] Step 3: Asynchronous Task Execution and Status Monitoring: Users initiate analysis tasks node by node. Upon receiving parameters, the backend sequentially constructs command lines for bioinformatics tools such as Trimmomatic, HISAT2, FeatureCounts, and DESeq2, starting the analysis as independent asynchronous processes. Simultaneously, it persistently stores metadata such as process PID, log file path, and input / output directories. The frontend polls the backend status interface at fixed intervals of 1200ms, obtaining node status in real-time through a dual-state determination mechanism: when a process is alive, the canvas node displays a blue "running" status; when the process terminates and a preset completion marker is detected in the log file, the node displays a green "successful" status; if the process terminates abnormally and no completion marker is detected, the node displays a red "failed" status. Users can visually view the running status of each node through the canvas, and when a task fails, they can directly locate the problematic node and retrieve logs to investigate the cause.

[0039] Step 4, Results Acquisition and Template Deposit: After all nodes have run successfully, users can download the differential gene list, differential expression volcano plot, GO / KEGG enrichment analysis bubble chart, co-expression network heatmap, and other analysis results from the output results panel of each node. After confirming that the process is stable and usable, users can save the current canvas process as a private template or submit a template release application. After being reviewed and approved by the administrator, it will be included in the shared template library for other users in the team to reuse directly.

[0040] Example 4: Intelligent Generation and Execution of Single-Cell RNA Sequencing (scRNA-seq) Analysis Workflow. This example is a specific application of the analysis method described in Example 2 in a single-cell RNA sequencing scenario. It uses natural language to drive the intelligent generation of a single-cell analysis workflow, verifying the effectiveness of the dual-channel intelligent generation strategy. The specific steps are as follows: Step 1, Natural Language Command Input: The user enters the command at the platform's intelligent generation entry: "Generate a single-cell RNA-seq analysis workflow, including cell clustering, cell type annotation, marker gene screening, and using the human reference genome."

[0041] Step 2, Dual-channel intelligent generation workflow: The system prioritizes the RAG intelligent generation channel, using the BM25 retrieval tool to recall Top-5 relevant document fragments from the bioinformatics analysis knowledge base as context. It then calls the large language model and generates a structured workflow plan through Pydantic structured output constraints. This plan includes the workflow intent of single-cell transcriptome analysis, matching corresponding standard templates, and default parameters for each node, with an output confidence score of 0.88. Since this confidence score is higher than the preset threshold of 0.6, the system converts the structured workflow plan into renderable workflow configuration data. It automatically renders four nodes—data preprocessing (PreProcess_scRNA), cluster analysis (Cluster_scRNA), cell annotation (Annotation_scRNA), and marker gene analysis (Marker_scRNA)—along with their corresponding connections on the canvas, and automatically injects parameter configurations such as the human reference genome.

[0042] If the confidence level of the AI-generated result is lower than the threshold, the system automatically switches to the rule matching channel: the process type is identified by keywords such as "single cell", "clustering" and "Marker", multi-dimensional weighted scoring is performed on the single cell templates in the shared template library, and the standard single cell analysis template with the highest score is selected; at the same time, the "human reference genome" parameter is extracted from the user's instructions, normalized and mapped (species names are uniformly converted into standard identifiers) and injected into the corresponding node configuration to generate the final process configuration data.

[0043] Step 3, Parameter Adjustment and Task Execution: Users verify the workflow structure and parameters in the canvas, and can adjust configurations such as cluster resolution and cell annotation database as needed. After confirmation, the analysis task is started node by node. The backend starts analysis tools such as CellRanger and Seurat in independent asynchronous processes, and transmits the running status in real time through a dual status determination mechanism. Users can intuitively grasp the progress of each step through the status indicators of the canvas nodes.

[0044] Step 4, Results Output and Template Iteration: After the analysis is completed, users can download result files such as cell clustering UMAP diagrams, cell type annotation results, and marker gene expression heatmaps. The validated and optimized analysis workflow can be saved as a new template, which will be included in the shared template library after being reviewed by the administrator, providing more executable workflow benchmarks for subsequent intelligent workflow generation.

[0045] Example 5: Chromatin Accessibility Sequencing (ATAC-seq) Template Reuse Analysis. This example is a specific application of the analysis method described in Example 2 in a chromatin accessibility sequencing scenario. Based on a shared template, the entire ATAC-seq workflow analysis is quickly completed, verifying the reuse value of the template sharing system. The specific steps are as follows: Step 1, One-click template loading: Users enter the platform's template center, select the approved and published "ATAC-seq Standard Analysis Template" from the shared template library, and click one-click loading. The complete node layout, connection relationships, and standardized parameters contained in the template are automatically rendered to the current canvas. The process covers five core nodes: data preprocessing (PreProcess_ATAC), quality control (QC_ATAC), peak calling (PeakCall_ATAC), motif analysis (Motif_ATAC), and IGV visualization (IGV_ATAC).

[0046] Step 2, Sample parameter configuration: The user uploads the original FASTQ file of ATAC-seq from mice in the PreProcess_ATAC node, modifies the reference genome to mm10, and retains the template preset standardized analysis parameters in the other nodes.

[0047] Step 3, Full-Process Execution and Monitoring: The user initiates the full-process operation. The system executes the analysis tasks of each node sequentially according to the connection topology, calling bioinformatics tools such as Trim-Galore, Bowtie2, MACS2, HOMER, and deepTools in sequence. During task execution, a dual-state detection mechanism provides real-time feedback on the running status of each node. The user can intuitively grasp the progress of the entire process through the status indicators of the canvas nodes. After the QC_ATAC node is completed, the user can directly view key quality indicators such as the inserted fragment length distribution and TSS enrichment score.

[0048] Step 4, Results Download and Template Optimization: After all nodes have run successfully, users can download analysis products such as peak interval files, Motif enrichment results, and genomic region IGV snapshots. If users have optimized and adjusted the process parameters for this sample, they can save the modified process as a new private template or submit a release application. Once approved, the shared template library will have a wider range of application scenarios.

[0049] Example 6: Chromium Immunoprecipitation Sequencing (ChIP-seq / CUT&Tag) Analysis Implementation. This example demonstrates the application of the analysis method described in Example 2 within a chromatin immunoprecipitation sequencing scenario, verifying the platform's unified support capability for ChIP-seq / CUT&Tag omics types. The specific process is as follows: Users can manually build workflows by dragging and dropping nodes such as data preprocessing, sequence alignment, peak calling, enrichment analysis, and visualization through the ChIP-seq category in the left-hand node menu; alternatively, they can trigger the intelligent generation channel by using the natural language command "Generate ChIP-seq standard analysis workflow, using the human reference genome," automatically generating a standardized analysis workflow. During workflow execution, the backend calls corresponding analysis tools such as Bowtie2, MACS2, and deepTools, monitoring task progress through a dual status detection mechanism consistent with other omics. After analysis, peak files, enrichment heatmaps, and genome visualization results can be downloaded. This omics type is registered with an independent backend Blueprint module, reusing unified process management, file management, and status detection logic. The frontend uses parameter component templates consistent with other omics, ensuring a consistent user experience.

[0050] In summary, this invention lowers the technical threshold for bioinformatics analysis, improves analysis efficiency and process reusability, and achieves one-stop, visualized, and traceable low-code analysis of multi-omics data by using a visual drag-and-drop process construction, natural language-driven intelligent process generation, asynchronous task monitoring with dual state determination, multi-omics unified architecture support, and a process template sharing system with an audit mechanism.

Claims

1. A low-code, multi-group bioinformatics analysis method, characterized in that, Includes the following steps: Step S1: Create a visualization canvas and receive drag and drop operations and connection operations of bioinformatics analysis tool nodes. Extract the unique identifier prefix of the analysis node to be connected for group matching, block the connection that does not conform to the prefix grouping rule, and generate process configuration data; or receive natural language description, use a dual-channel strategy of AI generation priority and rule matching backoff to generate process configuration data, and render it as analysis tool nodes and connections on the canvas. Step S2: Receive user adjustments to parameters and input file selections for each analysis node, and save the complete process execution configuration data; Step S3: Asynchronously start the bioinformatics analysis task according to the process configuration data. The task running status is obtained in real time through a dual status determination mechanism that combines process survival detection and log marking detection, and the status data is sent back to the canvas to update the node status indication. Step S4: When the task is determined to be successful, output multiple sets of bioinformatics analysis results; save the analysis process verified by the task as a template, review it and include it in the shared template library for subsequent manual process construction and intelligent process generation.

2. The low-code multi-group bioinformatics analysis method according to claim 1, characterized in that, In step S1, the user's natural language description is received, and a dual-channel strategy of AI-driven generation and rule-matching fallback is used to automatically generate process configuration data, specifically including: S11. Receive user natural language instructions and prioritize calling the RAG intelligent generation channel: retrieve relevant document fragments from the pre-built bioinformatics analysis knowledge base as context, call the large language model to generate a structured process plan and output confidence scores; S12. Determine whether the confidence score is higher than the preset threshold. If so, convert the structured process plan into renderable process configuration data; otherwise, switch to the rule matching channel. S13. The rule matching channel identifies the process type corresponding to the user's instruction through keyword matching, performs weighted scoring matching on the templates in the shared template library to select the best matching template, and extracts the parameter information in the user's instruction. After normalization, it is injected into the template configuration to generate the final process configuration data.

3. The low-code multi-group bioinformatics analysis method according to claim 1, characterized in that, In step S3, the specific process of the dual-state determination mechanism is as follows: S31. During the task startup phase, the bioinformatics analysis tool is started in an independent asynchronous process session, and the corresponding process identifier and log file path are recorded and persistently stored. S32. When receiving a status query request, first check if the corresponding log file exists. If it does not exist, return the inactive status. S33. If the log file exists, verify the liveness status of the corresponding process through the system process query interface; S34. If the process is in a running state, return directly to the running state; S35. If the process has terminated, read the complete contents of the corresponding log file and search for the predefined task completion marker string; if the completion marker string is detected, return a successful running status; otherwise, return a failed running status.

4. The low-code multi-group bioinformatics analysis method according to claim 1, characterized in that, In step S4, the analysis process information that has completed the verification process in the canvas is serialized and saved as a user-private template in a standardized format; After a user submits a template release application, the templates approved by the administrator are included in the shared template library; the shared template library also serves as a candidate data source for intelligent process generation, providing an executable process benchmark for dual-channel process generation.

5. A low-code, multi-group bioinformatics analysis system, characterized in that, include: The visual workflow orchestration module provides a web-based visual orchestration canvas, enabling drag-and-drop analysis workflow setup, connection validity verification, operation history management, node parameter configuration and file selection, and outputting workflow configuration data. The intelligent process generation module communicates with the visual process orchestration module. It generates process configuration data based on the user's natural language description and transmits the process configuration data to the visual process orchestration module, which then renders it onto the canvas. The task execution and status monitoring module is bidirectionally connected to the visualization process orchestration module. It is used to asynchronously start analysis tasks based on process configuration data, obtain task running status through a dual status determination mechanism, and send status data back to the visualization process orchestration module to update canvas node status indicators. The multi-omics unified architecture support module provides unified underlying operation support for the visualization process orchestration module, intelligent process generation module, task execution and status monitoring module, and analysis template sharing module. It is used to organize and operate the support system with unified tools to support the full-process analysis of multiple omics data types. The analysis template sharing module communicates with the visualization process orchestration module and the intelligent process generation module respectively. It is used to realize the template-based saving, review and release, and sharing and reuse of analysis processes, and to provide a shared template library for intelligent process generation.

6. The low-code multi-group bioinformatics analysis system according to claim 5, characterized in that, The visualization process orchestration module includes a node menu unit, a canvas editing unit, a history management unit, and an interactive zoom unit; The node menu unit is used to construct a hierarchical classification tree according to function and omics type, and displays multi-omics analysis tool nodes including quality control, sequence alignment, quantitative analysis and differential analysis. Each node has preset default analysis parameters and is managed uniformly through a unique identifier. The canvas editing unit is built on a visual canvas rendering engine and is used to render analysis nodes with status indicators. It receives drag-and-drop connection operations through the input and output endpoints at the end of the node and automatically blocks invalid connections in the same group through pre-grouping verification rules. The history management unit adopts a command mode to build a stack-based operation record system, which is used to store and manage all canvas editing operations and supports multi-step undo and redo. The interactive scaling unit integrates a canvas interactive control component to realize multi-level scaling and panning of the canvas, and synchronizes the scaling factor to the canvas rendering engine.

7. The low-code multi-group bioinformatics analysis system according to claim 5, characterized in that, The intelligent process generation module adopts a dual-channel hybrid strategy of AI generation priority and rule matching fallback, including a rule matching unit and a RAG intelligent generation unit; The rule matching unit is used to identify the process type input by the user through keyword matching, perform multi-dimensional weighted scoring on each template in the shared template library to select the best matching result, and inject the parameters input by the user into the corresponding node configuration through parameter extraction and normalization mapping, and output process configuration data. The RAG intelligent generation unit is used to construct a retrieval-generation link. It retrieves relevant document fragments from the bioinformatics analysis knowledge base as context through a keyword retrieval device, calls a large language model, and generates a structured process plan containing process intent, template selection, node parameters, and confidence scores through structured output constraints. The intelligent process generation module has a preset confidence threshold. When the confidence of the AI-generated result is lower than the threshold, it automatically switches to the rule matching unit to output process configuration data.

8. The low-code multi-group bioinformatics analysis system according to claim 5, characterized in that, The task execution and status monitoring module includes a task initiation unit and a dual status detection unit; The task initiation unit is used to receive parameter configuration and input file path from the front end, construct bioinformatics analysis command line, start the analysis tool in an independent asynchronous process, and persistently store runtime metadata. The dual-state detection unit adopts a dual-determination logic that combines process liveness detection and log marking detection, with log existence verification as the first judgment: first, it checks whether the task log file exists. If the log does not exist, it is determined to be in the not started state; if the log exists, it checks the process liveness status. If the process is alive, it is determined to be in the running state; if the process has terminated, it is determined to be in the running state by marking the log completion status; a total of four task states are output: not started, running, running successfully, and running failed. The front end obtains real-time status data through a timed polling mechanism, which drives the status indicators of the canvas nodes to update and display synchronously.

9. The low-code multi-group bioinformatics analysis system according to claim 5, characterized in that, The multi-omics unified architecture support module covers four types of omics analysis: conventional transcriptome sequencing, single-cell RNA sequencing, chromatin accessibility sequencing, and chromatin immunoprecipitation sequencing. The backend adopts a modular architecture, with each omics type independently registering functional interfaces while sharing a unified process management framework, file management mechanism, and status detection logic; The visualization workflow orchestration module is based on a multi-omics unified architecture support module. It organizes all omics analysis tool nodes with a unified hierarchical classification tree and uses a unified parameter component template to render configuration forms, providing a consistent end-to-end operation experience for all omics analyses.

10. The low-code multi-group bioinformatics analysis system according to claim 5, characterized in that, The analysis template sharing module includes a template saving unit, a template review unit, and a template loading unit; The template saving unit is used to serialize and save the process configuration data built in the canvas into a user private template in a standardized format; The template review unit is used to review user-submitted templates by administrators. Templates that pass the review are changed to publicly released status and included in the shared template library. The template loading unit is used to provide users with template browsing and one-click loading functions; The shared template library serves as a candidate space for the intelligent process generation module, ensuring that the intelligently generated analysis process is executable.

Images