Multi-omics data analysis method, system, electronic device, and storage medium

Abstract

The application provides a multi-omics data analysis method, system, electronic equipment and storage medium, and belongs to the technical field of bioinformatics and artificial intelligence cross-application, and the method comprises the following steps: generating a running plan based on an analysis request and meta-information of multi-omics data and a plurality of analysis skills; executing the running plan to obtain member products of each candidate analysis member; determining a plurality of target analysis members from all candidate analysis members based on a first screening rule, wherein the first screening rule is determined based on the maximum category proportion of each output label of each candidate analysis member; determining a plurality of basic analysis members from all target analysis members based on a second screening rule, wherein the second screening rule is screening according to a composite score of a consistency index and a quality index of each target analysis member; and generating an analysis result based on each output label of all basic analysis members. The application improves the accuracy and reproducibility of data analysis by setting constraint conditions.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary applications of bioinformatics and artificial intelligence, and in particular to a multi-omics data analysis method, system, electronic device, and storage medium. Background Technology

[0002] Multi-omics data analysis typically involves a variety of complex data types, such as spatial transcriptomics, single-cell omics, genomics, and proteomics. In practical analysis scenarios, researchers often need to frequently switch between various programming languages ​​and command-line tools, repeatedly and manually select algorithms such as preprocessing, clustering, and spatial domain identification, and spend a lot of time organizing parameter configurations and result files.

[0003] However, existing multi-omics data analysis workflows have many limitations: the selection of analysis methods relies heavily on the researchers' human experience, and different algorithms vary greatly in input requirements, output formats, and evaluation metrics, making it difficult to compare the results of multiple algorithms, lacking a unified and verifiable consensus merging path, and resulting in poor accuracy; although AI-assisted tools have been introduced in recent years, AI is prone to parameter illusion when automatically generating analysis steps, and its planned execution path is often uncertain and has poor reproducibility; finally, existing desktop visualization interfaces are usually disconnected from the underlying actual operation and lack consistent and auditable operation records. Summary of the Invention

[0004] This invention provides a multi-omics data analysis method, system, electronic device, and storage medium to address the shortcomings of poor accuracy and reproducibility in existing multi-omics data analysis methods.

[0005] This invention provides a multi-omics data analysis method, comprising: Based on analysis requests and metadata from multi-omics data, as well as multiple registered analysis skills, an execution plan is generated, which includes multiple candidate analysis members. Execute the run plan to obtain the member output of each candidate analysis member, and each member output includes the output label of each observation object of the multi-omics data; Based on the first screening rule, multiple target analysis members are determined from all the candidate analysis members. The first screening rule is determined based on the maximum category proportion of each output label of each candidate analysis member. The maximum category proportion is the ratio of the number of output labels belonging to the single largest category to the total number of observed objects. Based on the second screening rule, multiple basic analysis members are determined from all the target analysis members. The second screening rule includes determining a composite score based on the consistency index and quality index of each target analysis member, and screening all the target analysis members based on the composite score. The analysis results of the multi-omics data are generated based on the output labels of all the aforementioned basic analysis members.

[0006] In some embodiments, generating the analysis results of the multi-omics data based on the output labels of all the basic analysis members includes: Align the output labels of each of the basic analysis members to obtain aligned output labels; The deterministic consensus operator is invoked to generate the analysis results of the multi-omics data based on the aligned output labels.

[0007] In some embodiments, the invocation of the deterministic consensus operator to generate the analysis results of the multi-omics data based on the aligned output labels includes: The total number of members in the basic analysis is counted; If the total number is determined to be greater than a preset threshold, a deterministic consensus operator is invoked to determine a consensus label based on the aligned output label and generate the analysis result. The deterministic consensus operator is any one of the following: observation-by-observation mode consensus, majority consensus weighted by member composite score, or latent category analysis consensus.

[0008] In some embodiments, the analysis request and metadata based on multi-omics data, as well as multiple registered analysis skills, generate an execution plan, including: The analysis request is matched with multiple analysis skills to obtain matching results; If a match is successful, the target analytical skill is determined based on the matching result; The necessary data prerequisites and registration information for obtaining the target analysis skill are as follows: the registration information includes skill description, method set, and parameter constraints. Based on the meta-information, check whether the multi-omics data meets the necessary data prerequisites. If it does, input the analysis request, the meta-information, and the registration information of the target analysis skill into the pre-trained planning model. Based on the analysis request and the metadata, the planning model recommends multiple candidate analysis methods and corresponding key parameters under the constraints of the registration information, and outputs the operation plan.

[0009] In some embodiments, determining the target analytical skill based on the matching result includes: If, based on the matching results, it is determined that there is a uniquely matching analysis skill among the multiple analysis skills, then the uniquely matching analysis skill is determined as the target analysis skill; If, based on the matching results, multiple matching analysis skills are determined among the multiple analysis skills, a structured pre-detection question is generated based on the multiple matching analysis skills, the structured pre-detection question is sent to the user terminal, confirmation information is received from the user terminal, and the target analysis skill is determined from the multiple matching analysis skills based on the confirmation information.

[0010] In some embodiments, executing the run plan to obtain the member output of each candidate analysis member includes: Obtain the registered workflow template, which is used to define the execution rules of the candidate analysis members and the reading method of the member outputs; Based on the workflow template, the operation plan is converted into workflow steps for each of the candidate analysis members; Multiple workflow steps are executed in parallel to obtain the member output of each candidate analysis member.

[0011] In some embodiments, the workflow template is pre-registered in the following manner: Define the consensus type and source attribute of the workflow template. The consensus type includes output label consensus, and the source attribute includes verification path and exploration path. Register at least one source entry for the workflow template, the source entry being bound to member skills, member artifact readers, member planners, and analysis domains.

[0012] In some embodiments, determining multiple target analysis members from all the candidate analysis members based on a first screening rule includes: The maximum category proportion of each output label of each candidate analysis member is compared with a preset proportion threshold. From the multiple candidate analysis members, remove the candidate analysis members whose largest category proportion is greater than the preset proportion threshold to obtain multiple target analysis members.

[0013] In some embodiments, determining multiple basic analysis members from all the target analysis members based on a second screening rule includes: Determine the first weight of the consistency index and the second weight of the quality index; Based on the first weight and the second weight, the consistency index and the quality index of each target analysis member are weighted and summed to obtain the composite score of each target analysis member; Based on the composite score and / or user input, multiple basic analysis members are selected from the multiple target analysis members.

[0014] In some embodiments, after generating the analysis results of the multi-omics data based on the output labels of all the aforementioned basic analysis members, the method further includes: An analysis record of the multi-omics data is generated, the analysis record including the operation plan, the consistency index of each target analysis member, the quality index and the composite score, and the analysis results; If the analysis results include output labels of a unified category, then the analysis results are determined to be verification consensus results, and the analysis results are isolated and written into the verification namespace. If the analysis results do not include output labels of a uniform category, then the analysis results are determined to be exploratory synthesis results, and the analysis results are isolated and written into the exploratory namespace.

[0015] This invention also provides a multi-omics data analysis system, comprising: The plan generation module is used to generate an operation plan based on analysis requests and metadata from multi-omics data, as well as multiple registered analysis skills. The operation plan includes multiple candidate analysis members. An execution module is used to execute the operation plan and obtain the member product of each candidate analysis member, wherein each member product includes the output label of each observation object of the multi-omics data; A filtering module is used to determine multiple target analysis members from all the candidate analysis members based on a first filtering rule, wherein the first filtering rule is determined based on the maximum category proportion of each output label of each candidate analysis member, and the maximum category proportion is the ratio of the number of output labels belonging to the single most numerous category to the total number of observed objects; The scoring and screening module is used to determine multiple basic analysis members from all the target analysis members based on a second screening rule. The second screening rule includes determining a composite score based on the consistency index and quality index of each target analysis member, and screening all the target analysis members based on the composite score. The results generation module is used to generate analysis results of the multi-omics data based on the output labels of all the basic analysis members.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multi-omics data analysis method as described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-omics data analysis method as described above.

[0018] This invention provides a multi-omics data analysis method, system, electronic device, and storage medium. Based on analysis requests and metadata from multi-omics data, and multiple registered analysis skills, it generates an execution plan, confining intelligent planning within the boundaries of the registered skills, thus improving the controllability of the execution plan. Executing the execution plan yields the outputs of each candidate analysis member, improving the execution efficiency of data analysis. Based on a first screening rule, multiple target analysis members are determined from all candidate analysis members. This first screening rule is based on the maximum category proportion of each candidate analysis member's output labels. Based on a second screening rule, multiple basic analysis members are determined from all target analysis members. This second screening rule is based on a composite score determined by consistency and quality indicators for each target analysis member, ensuring high accuracy and reliability of the selected members. Based on the output labels of all basic analysis members, analysis results are generated, improving the accuracy and reproducibility of data analysis. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the multi-omics data analysis method provided by the present invention.

[0021] Figure 2 This is a schematic diagram of the process for generating analysis results of multi-omics data provided by the present invention.

[0022] Figure 3 This is a schematic diagram of the structure of the multi-omics data analysis system provided by the present invention.

[0023] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] Figure 1 This is a flowchart illustrating the multi-omics data analysis method provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps: Step 110: Based on the analysis requests and metadata of multi-omics data, as well as multiple registered analysis skills, generate an operation plan, which includes multiple candidate analysis members.

[0026] This multi-omics data includes, but is not limited to, spatial transcriptomics data, single-cell omics data, genomics data, proteomics data, and metabolomics data. Analysis requests can be text-based requests in natural language initiated by users through desktop interfaces, command lines, or messaging entry points.

[0027] Optionally, the data path of the multi-omics data can be obtained, and the metadata of the multi-omics data can be determined based on the data path. The data path includes paths stored locally or controlled remote paths, and the corresponding data file types can be h5ad files, matrix files, sequencing files, table files, or result directories of partially analyzed data, etc.

[0028] The metadata of multi-omics data includes at least a data structure summary, field names, modality information, dimension information, and lightweight information such as necessary runtime status. Specifically, when extracting metadata, direct reading and uploading of the original multi-omics matrix should be avoided as much as possible to ensure data security.

[0029] Among these, "multiple analysis skills" refers to a variety of pre-maintained bioinformatics analysis skills within the system, covering bioinformatics analysis algorithms or tools such as clustering, spatial domain identification, and differential analysis. "Candidate analysis members" refers to instances of analysis tasks to be executed.

[0030] Specifically, by combining the acquired analysis requests, metadata, and various analysis skills registered within the system, intelligent planning can be performed within defined skill boundaries to generate execution plans for multiple candidate analysis members targeting different algorithm instances. The execution plans specify the analysis methods and key parameter configurations to be executed for each candidate analysis member.

[0031] Step 120: Execute the run plan to obtain the member outputs of each candidate analysis member. Each member output includes the output labels of each observation object in the multi-omics data.

[0032] Optionally, multiple candidate analysis members can be executed in parallel.

[0033] The outputs of each member can include not only runtime status, time consumption records, or error information, but also core output labels. Output labels are classification identifiers or prediction results assigned to the observed objects, such as the spatial domain label for each observation point output by the spatial domain identification member, or the clustering label for each cell output by the single-cell clustering member.

[0034] In this context, the observation object refers to the basic unit that is independently analyzed and labeled in multi-omics data; for spatial transcriptomics data, the observation object is the capture point; for single-cell omics data, the observation object is a single cell; and for batch ribonucleic acid (RNA) sequencing data, the observation object is a single sample.

[0035] Step 130: Based on the first screening rule, determine multiple target analysis members from all candidate analysis members. The first screening rule is determined based on the maximum category proportion of each output label of each candidate analysis member. The maximum category proportion is the ratio of the number of output labels belonging to the single most numerous category to the total number of observed objects.

[0036] Specifically, the distribution of the number of each category in the output labels of each candidate analysis member is statistically analyzed, and the specific proportion of the category with the largest proportion in the total is calculated, i.e., the maximum category proportion. To prevent a single category from dominating the entire analysis result due to improper parameter selection in an analysis algorithm, multiple candidate analysis members are coarsely filtered based on this maximum category proportion, and abnormal members with excessively high maximum category proportions are removed. The remaining candidate analysis members after filtering are then used as target analysis members and retained for the subsequent scoring stage.

[0037] Step 140: Based on the second screening rule, determine multiple basic analysis members from all target analysis members. The second screening rule includes determining a composite score based on the consistency index and quality index of each target analysis member, and screening all target analysis members based on the composite score.

[0038] Among them, consistency metrics are used to measure the degree of cross-method consistency between the output labels of the current target analysis member and the output labels of other members; for example, normalized mutual information can be used as a reference. Quality metrics are used to measure the intrinsic quality of the algorithm of the member's output results; for example, for single-cell clustering, the silhouette coefficient can be used as a reference, and for spatial analysis, metrics such as spatial local purity or spatial continuity can be used as references.

[0039] Specifically, a weighted calculation is performed by combining consistency and quality indicators to obtain a composite score for each target analysis member. Based on this composite score, basic analysis members who meet the requirements are selected from multiple target analysis members, for example, several members with top scores are selected.

[0040] Step 150: Based on the output labels of all basic analysis members, generate the analysis results of multi-omics data.

[0041] The analysis results can be either merged structured data reports or consensus tag files with unified identifiers, enabling the integration of heterogeneous outputs from various algorithms and providing researchers with intuitive and clear data references.

[0042] In this embodiment of the invention, by acquiring analysis requests and metadata and generating run plans for candidate analysis members based on registered skills, the algorithm scheduling is constrained within a defined boundary. Subsequently, the run plan is executed to obtain member products containing output labels. Candidate members are hard-filtered based on the largest category proportion to eliminate algorithm products with distorted distributions. Furthermore, the composite score of the target analysis members is calculated by combining consistency and quality indicators to screen out high-quality basic analysis members. Finally, the analysis results of multi-omics data are generated based on the output labels of the basic analysis members. This transforms the discrete results generated independently by multiple tools into a unified product that has undergone objective quantitative screening, improving the accuracy of multi-algorithm collaborative merging and the reproducibility of the data processing path.

[0043] In some embodiments, step 110, based on the analysis requests and metadata of multi-omics data, and multiple registered analysis skills, generates an execution plan, including: Step 111: Match the analysis request with multiple analysis skills to obtain the matching results.

[0044] Optionally, by using natural language processing techniques such as semantic similarity calculation or keyword extraction, the analysis request is compared with the features of each registered analysis skill, and the resulting comparison score or association mapping result is the matching result.

[0045] Step 112: If the match is successful, determine the target analysis skills based on the matching results.

[0046] Specifically, the matching results determine whether a skill category matches the user's intent. If a match is successful, the specific analytical skill that the user's intent points to is identified based on the matching degree or mapping relationship, and this skill is designated as the target analytical skill. For example, it can be determined whether the user's intent is to perform spatial niche identification or single-cell clustering.

[0047] In some embodiments, determining the target analytical skills based on the matching results includes: If, based on the matching results, a uniquely matching analytical skill is found among multiple analytical skills, then the uniquely matching analytical skill is identified as the target analytical skill. If the matching results indicate that there are multiple matching analysis skills among multiple analysis skills, a structured pre-detection question is generated based on the multiple matching analysis skills. The structured pre-detection question is sent to the user terminal, and the user terminal sends a confirmation message. Based on the confirmation message, the target analysis skill is determined from the multiple matching analysis skills.

[0048] It should be noted that when the analysis request is very specific, and only one pre-registered analysis skill is highly relevant to its semantic features or invocation conditions, that skill is the uniquely matched analysis skill. In this case, the uniquely matched analysis skill is identified as the target analysis skill, and the subsequent data pre-verification stage can proceed without manual intervention.

[0049] Specifically, when an analysis request has a broad intent or semantic ambiguity, resulting in the matching results being associated with two or more analysis skills simultaneously, multiple matching analysis skills exist. To avoid the system blindly executing multiple feasible paths, the core differences between the multiple matching analysis skills are extracted and combined to generate a structured pre-detection question for interactive inquiry. A structured pre-detection question is a text-based question with focus options or guiding prompts. This structured pre-detection question is pushed to user terminals such as desktop interfaces for display. After the user selects according to their research needs on the user terminal, the user terminal sends corresponding confirmation information back to the system. Upon receiving this confirmation information, and based on the true intent indicated by the confirmation information, the system accurately selects the target analysis skill from the multiple matching analysis skills.

[0050] Understandably, by automatically determining the target skill when a match is unique, and extracting differences to generate structured pre-detection questions for user confirmation when multiple matches occur, the efficiency of automated operation is balanced with the accuracy of expert human intervention. This prevents the planning model from arbitrarily assigning execution paths when faced with ambiguous instructions, and improves the accuracy of multi-omics analysis intent recognition.

[0051] Step 113: Obtain the necessary data prerequisites and registration information for the target analysis skill. The registration information includes the skill description, method set, and parameter constraints.

[0052] Optionally, the underlying configuration file bound to the target analysis skill can be invoked to extract the necessary data prerequisites and registration information. Necessary data prerequisites refer to the minimum requirements for multi-omics data upon which the analysis skill depends for successful operation; for example, spatial transcriptome analysis may require the data to contain spatial coordinate fields or embeddings. Registration information provides deterministic constraints on the skill's capabilities, specifically covering the skill description in standardized documentation form, the set of methods consisting of various underlying algorithm frameworks supported by the skill, and parameter constraints consisting of the parameter types and numerical ranges allowed for each algorithm.

[0053] Step 114: Based on meta-information, check whether the multi-omics data meets the necessary data prerequisites. If it does, input the analysis request, meta-information, and registration information of the target analysis skills into the pre-trained planning model.

[0054] Specifically, the process involves verifying whether the data structure summary or field dimensions in the metadata cover the necessary data prerequisites required for the target analysis skill. After confirming that the execution conditions are met, the user's original analysis request, the data metadata, and the registration information of the target analysis skill defining the execution boundaries are all converted into contextual prompts and input into the pre-trained planning model. The pre-trained planning model is an intelligent large language model or a dedicated planning module with natural language understanding and task decomposition capabilities.

[0055] Optionally, the planning model is trained based on the analysis request samples, meta-information samples, and registration information samples corresponding to the target analysis skills samples of the multi-omics data samples, as well as the operation plan labels of multiple candidate analysis members.

[0056] Step 115: Based on the analysis request and metadata, the planning model recommends multiple candidate analysis methods and corresponding key parameters under the constraints of the registration information, and outputs the operation plan.

[0057] Specifically, after receiving input information, the planning model performs reasoning and decision-making within the skill pool, method set, and parameter constraints defined by the registration information. Based on the focus of the analysis request and the characteristics of the meta-information, the planning model selects multiple candidate analysis methods that meet the task requirements from the method set, and assigns key parameters that conform to the parameter constraint rules to each method, thereby generating multiple specific and independent execution plans for candidate analysis members.

[0058] In this embodiment of the invention, the target analysis skill is determined by matching the analysis request with the pre-registered skill, and the relevant limiting information is input into the pre-trained planning model after verifying the necessary data. The planning model recommends analysis methods and parameters under the strict constraints of the registration information, which effectively avoids the risk of artificial intelligence directly generating uncontrolled analysis steps or fictitious execution parameters, and improves the rigor of the intelligent analysis plan and the determinism of the underlying execution.

[0059] In some embodiments, step 120 executes the run plan to obtain the member outputs for each candidate analysis member, including: Step 121: Obtain the registered workflow template. The workflow template is used to define the execution rules for candidate analysis members and the reading method of member outputs.

[0060] Specifically, analytical tasks that can undergo consensus processing are abstracted into specific structures for management, and workflow templates are structured abstract definitions of the analytical consensus execution process. Member execution rules include underlying operational control logic such as upper limits for concurrent execution, timeout settings, and cancellation signal propagation. The member output reading method specifies how the system should parse and extract specific result files from the output directories of each candidate analytical member.

[0061] In some embodiments, workflow templates are pre-registered in the following ways: Define the consensus type and source attributes of the workflow template. The consensus type includes output label consensus, and the source attributes include verification path and exploration path. Register at least one source entry for the workflow template. The source entry is bound to member skills, member artifact readers, member planners, and analysis domains.

[0062] Specifically, during workflow template pre-registration, the consensus type is first defined, i.e., the mathematical or synthetic form of the consensus result is determined, for example, specifying it as an output classification label type consensus. Simultaneously, source attributes are defined, categorizing those capable of rigorous mathematical alignment and yielding deterministic results as validation paths, and those unable to reach a unified standard and only capable of descriptive synthesis as exploratory paths. Based on this, the second level of the two-level registration structure is established, which involves adding corresponding source entries to the workflow template. These source entries, as specific assembly units, tightly bind together the member skills responsible for the specific analysis algorithm, the member product reader responsible for parsing the skill's output, the member planner responsible for generating the member parameters, and the analytical domain to which the task belongs, such as spatial transcriptomics or single-cell clustering.

[0063] Step 122: Based on the workflow template, convert the run plan into workflow steps for each candidate analysis member.

[0064] Each workflow step explicitly includes the name of the member to be called, the specific skill name to be run, and the set of configured parameters.

[0065] Step 123: Execute multiple workflow steps in parallel to obtain the member output of each candidate analysis member.

[0066] Optionally, multiple transformed workflow steps can be started and executed simultaneously in an asynchronous and parallel manner within the same process or through a distributed scheduling node. Each candidate analysis member runs independently in a separate subprocess without interfering with others, and each writes its running status, time consumption records, error information, and bioinformatics analysis results to an independent output directory, thereby summarizing the individual outputs of each candidate analysis member.

[0067] In this embodiment of the invention, execution and reading rules are defined by introducing a two-level registration structure of workflow templates and source entries, and the running plan is transformed into workflow steps for parallel execution. This decouples the underlying system orchestration logic from the specific bioinformatics analysis domain. When adding new analysis domains in the future, only the corresponding source entries need to be added without rewriting the overall core code, which improves the flexibility of system expansion and enhances the execution efficiency and security of workflow steps.

[0068] In some embodiments, step 130 involves determining multiple target analysis members from all candidate analysis members based on a first screening rule, including: Step 131: Compare the maximum category proportion of each output label of each candidate analysis member with the preset proportion threshold.

[0069] The preset ratio threshold can be flexibly set according to the specific omics data characteristics and the researchers' historical experience, for example, its value can be set to 80%.

[0070] Step 132: Remove candidate analysis members whose largest category proportion is greater than a preset proportion threshold from multiple candidate analysis members to obtain multiple target analysis members.

[0071] Specifically, if the maximum category proportion of a candidate analysis member exceeds a preset threshold, it means that due to improper key parameter settings or limitations of the algorithm itself, the algorithm has classified most of its analyzed observations into the same category, resulting in severe distortion of clustering or spatial domain identification results. A hard filtering operation is performed on these abnormal candidate analysis members, removing them from the subsequent evaluation process. After filtering, the remaining candidate analysis members with a relatively balanced and reasonable label distribution are determined as the target analysis members.

[0072] In this embodiment of the invention, by comparing the largest category proportion with a preset proportion threshold and removing candidate analysis members whose proportion exceeds the threshold, individual algorithm products with extremely distorted output distribution are directly excluded, preventing a single large category of data from swallowing the overall result in subsequent merging and ensuring the quality of algorithm members entering the subsequent quantitative scoring stage.

[0073] In some embodiments, step 140 involves determining multiple basic analysis members from all target analysis members based on a second screening rule, including: Step 141: Determine the first weight of the consistency index and the second weight of the quality index.

[0074] Optionally, the first and second weights can be determined based on historical experience data or user input.

[0075] Step 142: Based on the first weight and the second weight, perform a weighted summation of the consistency index and quality index of each target analysis member to obtain the composite score of each target analysis member.

[0076] Specifically, for each member of the target analysis, the specific value of their consistency index is multiplied by the first weight, and the specific value of their quality index is multiplied by the second weight. Then, the two products are added together to obtain the composite score.

[0077] Optionally, the formula for calculating the composite score is as follows: ; ; in, Indicates the first Composite scores of individual target analysis members; Indicates the first The consensus index of the target analysis members, namely the average normalized mutual information. It is the first weight; Indicates the first The target analysis member's quality index refers to the normalized quality index. As the second weight; This indicates the total number of members in the target analysis. Indicates the first The output labels of each target analysis member Indicates the first The output labels of each target analysis member express and Normalized mutual information.

[0078] Step 143: Based on composite scores and / or user input, select multiple basic analysis members from multiple target analysis members.

[0079] Specifically, the composite score-based filtering is mainly aimed at non-interactive scenarios or default execution processes. Multiple target analysis members can be sorted in descending order based on the composite score, and the top K basic members, where K is a positive integer greater than 1, can be selected from the sorting results.

[0080] User input refers to the information of specific members that researchers specify by manually selecting or issuing commands after viewing the rating and ranking table in a command-line interactive scenario or a desktop graphical interface.

[0081] In this embodiment of the invention, a composite score is obtained by weighting and summing the consistency index and the quality index by setting a first weight and a second weight. The quantitative score result is combined with the input instructions of expert users to screen basic analysis members. This takes into account both the external consensus between algorithms and the evaluation quality of the algorithms. It is also compatible with two business scenarios: fully automatic batch execution and manual review and intervention. This makes the final basic analysis members not only have objective and rigorous quantitative basis, but also accurately meet the actual professional needs of researchers, thereby improving the scientificity and confidence of subsequent data analysis results.

[0082] Figure 2 This is a schematic diagram illustrating the process for generating analysis results from multi-omics data, as provided by the present invention. Figure 2 As shown, in some embodiments, step 150 generates analysis results for multi-omics data based on the output labels of all basic analysis members, including: Step 151: Align the output labels of each basic analysis member to obtain the aligned output labels.

[0083] Because different multi-omics data analysis algorithms often use different naming rules or numbering methods for their output classification labels when performing clustering or classification, alignment is necessary to unify these labels. In practice, the output labels of each basic analysis member can be normalized by frequency. Then, the member with the most clusters of output labels is selected as the reference member. Algorithms such as Hungarian matching are then used to uniformly transform and map the output labels of other non-reference members into the label space of the reference member, resulting in aligned output labels. This ensures that the labels generated by each basic analysis member are clearly comparable in the same dimension.

[0084] Step 152: Invoke the deterministic consensus operator and generate analysis results for multi-omics data based on the aligned output labels.

[0085] Specifically, after aligning the labels, a pre-defined deterministic consensus operator can be invoked to fuse the results of each member. A deterministic consensus operator is an algorithmic operator that can derive stable and consistent output results under the same input data and parameter configuration.

[0086] In some embodiments, a deterministic consensus operator is invoked to generate analysis results for multi-omics data based on the aligned output labels, including: If the total number is greater than a preset threshold, the deterministic consensus operator is invoked to determine the consensus label based on the aligned output label and generate the analysis result. Among them, the deterministic consensus operator is any one of the following: mode consensus per observation object, majority consensus weighted by composite score of members, or latent category analysis consensus.

[0087] Specifically, before proceeding to consensus computation, the total number of basic analysis members who successfully generated readable output labels is counted and compared with a preset threshold. This preset threshold can be set according to the reliability requirements of the actual task, such as setting it to two to ensure that at least two members participate in the consensus. If the threshold is met, the consensus label is determined by invoking the observation-by-observation mode consensus, the majority consensus weighted by member composite scores, or the latent category analysis consensus.

[0088] Taking a majority consensus based on member composite scores as an example, for each observation object, the sum of the composite scores of the basic analysis members corresponding to each candidate label after alignment is calculated, and the label with the largest cumulative score is selected as the consensus label for that observation object. If weights and ties occur during the statistical process, a fixed column order or predetermined weighting rules can be used to break the ties, thereby generating the final consensus label, which is then compiled into the analysis results of multi-omics data. If the total number of basic analysis members obtained from the statistics does not reach the preset threshold, an error can be triggered to terminate the consensus merging.

[0089] In this embodiment of the invention, by uniformly aligning the output classification labels, the barrier of inconsistent label numbers between different algorithms is eliminated. Under the premise that the statistics meet the preset quantity threshold, deterministic consensus operators, including mode consensus for each observation object, are called to derive and determine the consensus labels. This ensures that the outputs of heterogeneous algorithms can be effectively merged in accordance with rigorous and controllable mathematical calculation rules, eliminating randomness in the merging process and improving the reliability and comparability of the results of multi-algorithm collaborative analysis.

[0090] In some embodiments, after generating the analysis results of multi-omics data based on the output labels of all basic analysis members, the method further includes: The analysis record generates multi-omics data, which includes the operation plan, consistency indicators, quality indicators and composite scores for each target analysis member, and analysis results.

[0091] If the analysis results include output labels of a unified category, then the analysis results are determined to be verification consensus results, and the analysis results are isolated and written into the verification namespace. If the analysis results do not include output labels of a uniform category, then the analysis results are determined to be exploratory synthesis results, and the analysis results are isolated and written into the exploratory namespace.

[0092] Optionally, the analysis results can be visualized on a desktop interface.

[0093] The desktop interface primarily serves to provide researchers with a comprehensive view of information, including a chat-style task entry point, multi-omics data files, a list of currently available analysis skills, the concurrent execution progress of each candidate analysis member, pending tasks, structured pre-detection questions generated by the system, generated member outputs and analysis results, historical analysis records, and the operational status of remote system nodes. The desktop interface maintains real-time communication with the underlying multi-omics data analysis runtime system through local services, continuously receiving streaming events generated during the underlying process and simultaneously rendering and displaying them to the user. It is crucial to emphasize that this desktop interface acts solely as a front-end interaction medium and does not independently interfere with or alter the underlying deterministic consensus algorithm logic.

[0094] Specifically, if the analysis results include uniform classification labels, it indicates that the analysis results have high comparability and rigorous mathematical derivation logic, thus confirming the analysis results as verification consensus results. To ensure data independence and accurate traceability, a dedicated storage identifier can be used to store the verification consensus results in an independent verification namespace for isolated storage.

[0095] Specifically, for certain specialized combinations of analytical skills, the output may merely be a narrative extraction of conclusions and points of conflict from various reports, which cannot be transformed into alignable labeled data on the same observation object. This means that the analytical results do not include unified classification labels. In such cases, instead of forcibly aligning, the results are classified as exploratory synthesis results, using a separate set of dedicated identifiers and written into a dedicated exploratory namespace, thus physically and logically isolating them from the outputs of the validation path.

[0096] In this embodiment of the invention, by generating an analysis record containing detailed process information such as the operation plan and multidimensional index scores, and by isolating the verification consensus results and exploratory synthesis results into independent namespaces based on whether they contain a unified classification label, not only does this ensure that every step of the multi-omics data analysis operation and evidence has a traceable and verifiable review basis, but it also strictly distinguishes between deterministic statistical consensus and descriptive textual synthesis, preventing researchers from mistaking intelligent textual summaries for rigorous mathematical consensus results, and improving the scientific rigor and audit compliance of the system output.

[0097] The multi-omics data analysis system provided by this invention is described below. The multi-omics data analysis system described below can be referred to in correspondence with the multi-omics data analysis method described above.

[0098] Figure 3 This is a schematic diagram of the structure of the multi-omics data analysis system provided by the present invention. Figure 3 As shown, the present invention also provides a multi-omics data analysis system 300, comprising: The plan generation module 310 is used to generate an operation plan based on analysis requests and metadata from multi-omics data, as well as multiple registered analysis skills. The operation plan includes multiple candidate analysis members. Execution module 320 is used to execute the run plan and obtain the member output of each candidate analysis member. Each member output includes the output label of each observation object in the multi-omics data. The filtering module 330 is used to determine multiple target analysis members from all candidate analysis members based on a first filtering rule. The first filtering rule is determined based on the maximum category proportion of each output label of each candidate analysis member. The maximum category proportion is the ratio of the number of output labels belonging to the single most numerous category to the total number of observed objects. The scoring and screening module 340 is used to determine multiple basic analysis members from all target analysis members based on a second screening rule. The second screening rule includes determining a composite score based on the consistency index and quality index of each target analysis member, and screening all target analysis members based on the composite score. The results generation module 350 is used to generate analysis results for multi-omics data based on the output labels of all basic analysis members.

[0099] Optionally, based on the output labels of all basic analysis members, analysis results of multi-omics data are generated, including: Align the output labels of each member of the basic analysis to obtain the aligned output labels; The deterministic consensus operator is invoked to generate analysis results for multi-omics data based on the aligned output labels.

[0100] Optionally, a deterministic consensus operator is invoked to generate analysis results for multi-omics data based on the aligned output labels, including: The total number of members in the basic statistical analysis; If the total number is greater than a preset threshold, the deterministic consensus operator is invoked to determine the consensus label based on the aligned output label and generate the analysis result. Among them, the deterministic consensus operator is any one of the following: mode consensus per observation object, majority consensus weighted by composite score of members, or latent category analysis consensus.

[0101] Optionally, based on analysis requests and metadata from multi-omics data, as well as multiple registered analysis skills, an execution plan is generated, including: The analysis request is matched with multiple analysis skills to obtain the matching results; If a match is successful, the target analytical skills will be determined based on the matching results. The necessary data prerequisites and registration information for acquiring target analysis skills are required. The registration information includes skill descriptions, method sets, and parameter constraints. Based on meta-information, check whether the multi-omics data meets the necessary data prerequisites. If it does, input the analysis request, meta-information, and registration information of the target analysis skills into the pre-trained planning model. Based on the analysis request and metadata, the planning model recommends multiple candidate analysis methods and corresponding key parameters under the constraints of the registration information, and outputs an operation plan.

[0102] Optionally, based on the matching results, target analysis skills can be determined, including: If, based on the matching results, a uniquely matching analytical skill is found among multiple analytical skills, then the uniquely matching analytical skill is identified as the target analytical skill. If the matching results indicate that there are multiple matching analysis skills among multiple analysis skills, a structured pre-detection question is generated based on the multiple matching analysis skills. The structured pre-detection question is sent to the user terminal, and the user terminal sends a confirmation message. Based on the confirmation message, the target analysis skill is determined from the multiple matching analysis skills.

[0103] Optionally, the run plan is executed to obtain the member outputs for each candidate analysis member, including: Retrieve the registered workflow template. The workflow template is used to define the execution rules for candidate analysis members and the way member artifacts are read. Based on the workflow template, the run plan is converted into workflow steps for each candidate analysis member; Multiple workflow steps are executed in parallel to obtain the member outputs of each candidate analysis member.

[0104] Optionally, workflow templates can be pre-registered in the following ways: Define the consensus type and source attributes of the workflow template. The consensus type includes output label consensus, and the source attributes include verification path and exploration path. Register at least one source entry for the workflow template. The source entry is bound to member skills, member artifact readers, member planners, and analysis domains.

[0105] Optionally, based on the first screening rule, multiple target analysis members are determined from all candidate analysis members, including: The maximum category proportion of each output label of each candidate analysis member is compared with the preset proportion threshold; From multiple candidate analysis members, those whose largest category proportion is greater than a preset proportion threshold are removed, resulting in multiple target analysis members.

[0106] Optionally, based on the second screening rule, multiple basic analysis members are identified from all target analysis members, including: Determine the first weight for the consistency indicator and the second weight for the quality indicator; Based on the first and second weights, the consistency index and quality index of each target analysis member are weighted and summed to obtain the composite score of each target analysis member; Based on composite scores and / or user input, multiple basic analysis members are selected from multiple target analysis members.

[0107] Optionally, the multi-omics data analysis system 300 also includes a storage module for performing the following steps after generating the analysis results of the multi-omics data based on the output labels of all basic analysis members: Generate analysis records of multi-omics data, including the operation plan, consistency indicators, quality indicators and composite scores for each target analysis member, and analysis results; If the analysis results include output labels of a unified category, then the analysis results are determined to be verification consensus results, and the analysis results are isolated and written into the verification namespace. If the analysis results do not include output labels of a uniform category, then the analysis results are determined to be exploratory synthesis results, and the analysis results are isolated and written into the exploratory namespace.

[0108] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 4As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communications bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other through the communications bus 440. The processor 410 can call logical instructions in the memory 430 to execute a multi-omics data analysis method. This method includes: generating an execution plan based on analysis requests and metadata of the multi-omics data, as well as multiple registered analysis skills. The execution plan includes multiple candidate analysis members. Executing the execution plan yields the member outputs of each candidate analysis member, each member output including the output label of each observation object in the multi-omics data. Based on a first screening rule, multiple target analysis members are determined from all candidate analysis members. The first screening rule is determined based on the maximum category proportion of each output label of each candidate analysis member, where the maximum category proportion is the ratio of the number of output labels belonging to the most numerous single category to the total number of observation objects. Based on a second screening rule, multiple basic analysis members are determined from all target analysis members. The second screening rule includes determining a composite score based on the consistency and quality indicators of each target analysis member, and screening all target analysis members based on the composite score. Finally, based on the output labels of all basic analysis members, the analysis results of the multi-omics data are generated.

[0109] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the multi-omics data analysis method provided by the methods described above. This method includes: generating an execution plan based on analysis requests and metadata of the multi-omics data, and multiple registered analysis skills; executing the execution plan to obtain the member output of each candidate analysis member, each member output including the output label of each observation object in the multi-omics data; determining multiple target analysis members from all candidate analysis members based on a first screening rule, the first screening rule being determined based on the maximum category proportion of each output label of each candidate analysis member, the maximum category proportion being the ratio of the number of output labels belonging to the most numerous single category to the total number of observation objects; determining multiple basic analysis members from all target analysis members based on a second screening rule, the second screening rule including determining a composite score based on the consistency index and quality index of each target analysis member, and screening all target analysis members based on the composite score; and generating analysis results of the multi-omics data based on the output labels of all basic analysis members.

[0111] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0112] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-omics data analysis method, characterized in that, include: Based on analysis requests and metadata from multi-omics data, as well as multiple registered analysis skills, an execution plan is generated, which includes multiple candidate analysis members. Execute the run plan to obtain the member output of each candidate analysis member, and each member output includes the output label of each observation object of the multi-omics data; Based on the first screening rule, multiple target analysis members are determined from all the candidate analysis members. The first screening rule is determined based on the maximum category proportion of each output label of each candidate analysis member. The maximum category proportion is the ratio of the number of output labels belonging to the single largest category to the total number of observed objects. Based on the second screening rule, multiple basic analysis members are determined from all the target analysis members. The second screening rule includes determining a composite score based on the consistency index and quality index of each target analysis member, and screening all the target analysis members based on the composite score. The analysis results of the multi-omics data are generated based on the output labels of all the aforementioned basic analysis members.

2. The multi-omics data analysis method according to claim 1, characterized in that, The analysis results of the multi-omics data generated based on the output labels of all the aforementioned basic analysis members include: Align the output labels of each of the basic analysis members to obtain aligned output labels; The deterministic consensus operator is invoked to generate the analysis results of the multi-omics data based on the aligned output labels.

3. The multi-omics data analysis method according to claim 2, characterized in that, The invocation of the deterministic consensus operator, based on the aligned output labels, generates the analysis results of the multi-omics data, including: The total number of members in the basic analysis is counted; If the total number is determined to be greater than a preset threshold, a deterministic consensus operator is invoked to determine a consensus label based on the aligned output label and generate the analysis result. The deterministic consensus operator is any one of the following: observation-by-observation mode consensus, majority consensus weighted by member composite score, or latent category analysis consensus.

4. The multi-omics data analysis method according to claim 1, characterized in that, The analysis requests and metadata based on multi-omics data, along with multiple registered analysis skills, generate an execution plan, including: The analysis request is matched with multiple analysis skills to obtain matching results; If a match is successful, the target analytical skill is determined based on the matching result; The necessary data prerequisites and registration information for obtaining the target analysis skill are as follows: the registration information includes skill description, method set, and parameter constraints. Based on the meta-information, check whether the multi-omics data meets the necessary data prerequisites. If it does, input the analysis request, the meta-information, and the registration information of the target analysis skill into the pre-trained planning model. Based on the analysis request and the metadata, the planning model recommends multiple candidate analysis methods and corresponding key parameters under the constraints of the registration information, and outputs the operation plan.

5. The multi-omics data analysis method according to claim 4, characterized in that, The step of determining the target analysis skill based on the matching result includes: If, based on the matching results, it is determined that there is a uniquely matching analysis skill among the multiple analysis skills, then the uniquely matching analysis skill is determined as the target analysis skill; If, based on the matching results, multiple matching analysis skills are determined among the multiple analysis skills, a structured pre-detection question is generated based on the multiple matching analysis skills, the structured pre-detection question is sent to the user terminal, confirmation information is received from the user terminal, and the target analysis skill is determined from the multiple matching analysis skills based on the confirmation information.

6. The multi-omics data analysis method according to claim 1, characterized in that, The execution of the run plan, yielding the member outputs for each candidate analysis member, includes: Obtain the registered workflow template, which is used to define the execution rules of the candidate analysis members and the reading method of the member outputs; Based on the workflow template, the operation plan is converted into workflow steps for each of the candidate analysis members; Multiple workflow steps are executed in parallel to obtain the member output of each candidate analysis member.

7. The multi-omics data analysis method according to claim 6, characterized in that, The workflow template is pre-registered in the following way: Define the consensus type and source attribute of the workflow template. The consensus type includes output label consensus, and the source attribute includes verification path and exploration path. Register at least one source entry for the workflow template, the source entry being bound to member skills, member artifact readers, member planners, and analysis domains.

8. The multi-omics data analysis method according to claim 1, characterized in that, The determination of multiple target analysis members from all the candidate analysis members based on the first screening rule includes: The maximum category proportion of each output label of each candidate analysis member is compared with a preset proportion threshold. From the multiple candidate analysis members, remove the candidate analysis members whose largest category proportion is greater than the preset proportion threshold to obtain multiple target analysis members.

9. The multi-omics data analysis method according to claim 1, characterized in that, Based on the second screening rule, multiple basic analysis members are determined from all the target analysis members, including: Determine the first weight of the consistency index and the second weight of the quality index; Based on the first weight and the second weight, the consistency index and the quality index of each target analysis member are weighted and summed to obtain the composite score of each target analysis member; Based on the composite score and / or user input, multiple basic analysis members are selected from the multiple target analysis members.

10. The multi-omics data analysis method according to any one of claims 2-9, characterized in that, After generating the analysis results of the multi-omics data based on the output labels of all the aforementioned basic analysis members, the method further includes: An analysis record of the multi-omics data is generated, the analysis record including the operation plan, the consistency index of each target analysis member, the quality index and the composite score, and the analysis results; If the analysis results include output labels of a unified category, then the analysis results are determined to be verification consensus results, and the analysis results are isolated and written into the verification namespace. If the analysis results do not include output labels of a uniform category, then the analysis results are determined to be exploratory synthesis results, and the analysis results are isolated and written into the exploratory namespace.

11. A multi-omics data analysis system, characterized in that, include: The plan generation module is used to generate an operation plan based on analysis requests and metadata from multi-omics data, as well as multiple registered analysis skills. The operation plan includes multiple candidate analysis members. An execution module is used to execute the operation plan and obtain the member product of each candidate analysis member, wherein each member product includes the output label of each observation object of the multi-omics data; A filtering module is used to determine multiple target analysis members from all the candidate analysis members based on a first filtering rule, wherein the first filtering rule is determined based on the maximum category proportion of each output label of each candidate analysis member, and the maximum category proportion is the ratio of the number of output labels belonging to the single most numerous category to the total number of observed objects; The scoring and screening module is used to determine multiple basic analysis members from all the target analysis members based on a second screening rule. The second screening rule includes determining a composite score based on the consistency index and quality index of each target analysis member, and screening all the target analysis members based on the composite score. The results generation module is used to generate analysis results of the multi-omics data based on the output labels of all the basic analysis members.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the multi-omics data analysis method as described in any one of claims 1 to 10.

13. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the multi-omics data analysis method as described in any one of claims 1 to 10.

Images