A gene variation evidence mining method and system based on multi-stage trusted reasoning

Abstract

The application discloses a kind of gene variation evidence mining method and system based on multi-stage trusted inference, the method includes: obtaining academic literature to be parsed, literature is structured and semantically blocked;Respectively extract gene variation experimental evidence in table data and text text, construct dual-source evidence set;Based on gene variation identification, complete heterogeneous evidence fusion, and evidence is standardized processing;Combined with original text accurate matching and semantic checking mechanism, evidence is carried out traceability checking, abnormal identification and no illusion accurate correction;Further, cross-fragment evidence is globally aligned, deduplicated and fused, to generate standardized, high-precision, traceable gene variation functional evidence dataset;The application can effectively solve the problems of unstable long text analysis, evidence missed detection, data distortion and model illusion in the prior art, and improve the accuracy and engineering application capability of gene variation evidence mining.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics gene data mining technology, specifically to a gene variation evidence mining method and system based on multi-stage credible reasoning. It is mainly used to automatically extract, verify, correct, and integrate experimental evidence of gene variation function from medical academic literature, and can be widely applied to scenarios such as gene function annotation, clinical interpretation of hereditary diseases, and genomic research data analysis. Background Technology

[0002] Functional evidence of gene variations is core foundational data for genomic research, clinical diagnosis of genetic diseases, and gene function annotation. Currently, the industry generally relies on manual review of academic literature to screen and organize experimental data and functional evidence corresponding to gene variations. This manual processing method is not only labor-intensive, time-consuming, and costly, but also highly susceptible to the influence of operator experience and subjective judgment, resulting in problems such as missing evidence, annotation bias, and poor data consistency, which cannot meet the needs of large-scale, standardized gene data analysis.

[0003] Existing automated data mining solutions mostly adopt a single-step model reasoning mode, directly extracting evidence from the document text, which has several technical shortcomings in practical applications:

[0004] First, the parsing stability of long academic papers is poor, and long text inference is prone to timeouts and loss of contextual information. Second, it only recognizes the main text content, ignoring key experimental data such as bibliographic tables, figure captions, and supplementary materials, resulting in a high rate of missed evidence. Third, the model's autonomous inference is prone to generating false information, such as experimental numerical deviations, missing experimental conditions, and incorrect associations of variant sites. Fourth, it lacks a sound evidence verification and error correction mechanism, resulting in redundant, messy, and unstandardized output data that cannot be traced back to the original literature. Fifth, after long texts are segmented, experimental evidence of the same variant is scattered across multiple text fragments, making it difficult to achieve cross-fragment alignment and fusion, and thus failing to simultaneously guarantee evidence recall and data accuracy.

[0005] In summary, existing technologies suffer from shortcomings such as timeout anomalies, missed evidence detection, data distortion, redundancy and disorder, and the inability to correct errors in manual annotation, necessitating technological improvements. Currently, there is a lack of a stable, multi-source data-covering, traceable, and error-correctable automated solution for mining gene variation evidence. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a method and system for mining gene mutation evidence based on multi-stage credible reasoning. This invention solves the problems of traditional technologies such as failure of long text parsing, missed detection of multi-source evidence, false data generated by models, inability to trace the source of evidence, data redundancy and disorder, and errors in manual annotation by using a complete set of technical solutions, including hierarchical document fragmentation, parallel extraction of multi-source data, evidence fusion standardization, multi-level verification and error correction, and cross-fragment global alignment and fusion. This enables standardized, automated, and high-precision mining of gene mutation evidence from academic literature.

[0007] The technical solution adopted in this invention is as follows:

[0008] In a first aspect, the present invention provides a method for mining gene variation evidence based on multi-stage credible reasoning, comprising the following steps:

[0009] S1: Obtain the academic document to be parsed, extract the full text of the document, and use a chapter-aware adaptive semantic segmentation strategy to segment the full text to obtain structured text blocks.

[0010] S2: Extract table-type variation evidence and text-type variation evidence from the structured text blocks respectively to obtain the first type of gene variation evidence set and the second type of gene variation evidence set;

[0011] S3: For the first and second sets of gene variation evidence, generate an initial list of gene variation evidence based on gene variation identifiers;

[0012] S4: Standardize the initial list of evidence for gene mutations to generate a standardized list of evidence;

[0013] S5: Verify each piece of evidence in the standardized evidence list using a dual evidence verification mechanism, screen and mark abnormal evidence, and generate a standardized evidence verification report;

[0014] S6: Based on the evidence verification report, locate all abnormal evidence, perform accurate correction of non-hallucination evidence, and obtain a list of gene mutation evidence after completing the repair of all abnormal evidence.

[0015] S7: Collect a list of gene variation evidence corresponding to all structured text blocks, perform global grouping based on gene variation sites, align multiple blocks and multiple sources of evidence corresponding to the same gene variation site, and output a multi-source aligned list of gene variation evidence.

[0016] S8: Based on a multi-source aligned list of gene variation evidence, global deduplication and fusion are performed by combining semantic equivalence and numerical equivalence rules, unifying the output format of all evidence, and finally generating a gene variation functional evidence dataset.

[0017] Furthermore, in step S1, the chapter-aware adaptive semantic segmentation strategy includes: relying on the standardized chapter structure of academic literature, distinguishing different functional chapters of the literature, prioritizing chapter boundaries as segmentation nodes, and ensuring that each text block independently carries a single complete biological experimental data or chapter content.

[0018] The segmentation process specifically involves: dividing the text into blocks based on the document's chapter structure, controlling the number of characters in each block within a preset threshold range; during the segmentation process, avoiding text truncation issues for tables and figures to ensure the semantic integrity of gene variation-related experiments; for short text blocks with insufficient character length, merging adjacent texts to complete semantics; and configuring a unique identifier, chapter attribute, and start and end position information for each text block after segmentation to obtain the structured text blocks.

[0019] Further, in step S2, the extraction of tabular variant evidence specifically involves: traversing all structured text blocks, identifying various types of tabular data within the structured text blocks, mining the gene variant information, experimental attribute descriptions, and quantitative experimental values ​​contained in the tables, preserving the table row structure and the corresponding relationship between variant sites and experimental data, and constructing a first type of gene variant evidence set derived from tables; wherein, the various types of tabular data include native structured tables and image parsing tables.

[0020] The extraction of text-based mutation evidence specifically involves: performing semantic recognition on each structured text block to screen for various gene mutation identifiers contained in the text; using the statements containing the gene mutation identifiers as the core unit, bidirectionally expanding the text boundary based on contextual semantic association rules, matching the corresponding experimental environment, experimental conditions, and experimental result information, generating complete experimental-level text evidence fragments, and constructing a second type of gene mutation evidence set originating from the main text.

[0021] Further, in step S3, generating the initial gene variation evidence list specifically involves: aggregating the first type of gene variation evidence set and the second type of gene variation evidence set; associating and grouping the heterogeneous evidence scattered in the table and text based on the gene variation identifier; integrating the multi-dimensional experimental evidence corresponding to the same gene variation; retaining the original literature source information of each piece of evidence; and obtaining the initial gene variation evidence list after unifying the data format.

[0022] Furthermore, in step S4, the standardization and unification process specifically involves: converting all gene variation identifiers into the HGVS standard format, and unifying the measurement units and expression methods of quantitative data; simultaneously, combining semantic similarity algorithms and experimental data equivalence judgment rules to eliminate invalid evidence with semantic repetition and data redundancy; wherein, the quantitative data includes experimental fold change and percentage.

[0023] Furthermore, in step S5, the dual evidence verification mechanism specifically involves: combining the precise matching of the original document text with the fuzzy semantic verification mechanism to verify the evidence from four dimensions: source traceability effectiveness, mutation matching accuracy, experimental data consistency, and experimental scenario adaptability.

[0024] The dual evidence verification mechanism includes verification of the original literature, verification of the unique source of gene variation, verification of precise matching of quantitative experimental data, and verification of consistency of experimental scenario.

[0025] The standardized evidence verification report includes the anomaly type and anomaly details.

[0026] Furthermore, in step S6, the hallucination-free accurate evidence correction only corrects and supplements the existing information in the original text, without automatically generating or expanding semantic content.

[0027] Secondly, the present invention provides a gene variation evidence mining system based on the method of multi-stage credible reasoning, including a document segmentation module, a dual-source evidence extraction and fusion module, an evidence standardization module, an evidence verification and correction module, and a global alignment and fusion module that are connected in sequence.

[0028] The document segmentation module is used to complete adaptive semantic segmentation of academic documents, output structured and traceable structured text segments, and avoid long text reasoning failures.

[0029] The dual-source evidence extraction and fusion module is connected to the document segmentation module to simultaneously extract the table experimental evidence and the text experimental evidence within the structured text segment, construct a dual-source evidence set, and then aggregate the dual-source heterogeneous evidence to complete the association grouping of homologous variation evidence and the retention of source information.

[0030] The evidence standardization module is used to unify the gene mutation identification format and data standards, and to clean up redundant evidence.

[0031] The evidence verification and correction module is used to verify the validity of evidence from multiple dimensions, output an anomaly verification report, and accurately correct abnormal evidence based on the verification report to avoid model illusion.

[0032] The global alignment and fusion module is used to complete evidence alignment, global deduplication and fusion across blocks, and output a standardized, high-quality gene variation functional evidence dataset.

[0033] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the gene variation evidence mining method based on multi-stage credible reasoning described above.

[0034] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the gene variation evidence mining method based on multi-stage credible reasoning.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] 1. This invention adopts an adaptive semantic segmentation method that adapts to chapters, replacing the traditional one-time full-text reasoning mode. It completely solves the industry problems of long document parsing timeout and context loss, reduces the document parsing anomaly rate from 10.77% to 0%, and greatly improves the stability and adaptability of document parsing. It can be adapted to medical academic papers of various lengths.

[0037] 2. This invention simultaneously covers two core data sources: document tables and main text, enabling simultaneous extraction and fusion of dual-source evidence. It overcomes the shortcomings of traditional schemes that only identify main text and omit key experimental data from tables, fully preserving core details such as experimental methods, experimental systems, and quantitative results, and effectively reducing the evidence miss rate.

[0038] 3. This invention constructs a three-level trustworthy control system of "relaxed extraction - strict verification - precise correction". First, it maximizes the collection of potential experimental evidence to ensure data coverage. Then, it screens abnormal data through multi-dimensional source tracing and verification. Finally, it accurately corrects errors based on the original text. This not only avoids the problem of the model autonomously generating false data, but also corrects the biases and errors caused by manual annotation, and significantly improves the overall data quality of gene variation evidence.

[0039] 4. This invention solves the problems of evidence fragmentation and redundancy caused by text fragmentation processing by cross-fragment evidence alignment and global deduplication and fusion. While ensuring a high evidence recall rate of 95%, it completes global deduplication and unified alignment, taking into account both evidence integrity and data simplicity, and adapts to the needs of large-scale gene data analysis.

[0040] 5. All evidence extraction, correction, and fusion operations in this invention can be traced back to the original literature. There is no self-invented fabrication or arbitrary expansion. The data has high credibility and can meet the high-rigor application scenarios such as clinical gene interpretation, authoritative gene database construction, and scientific research data analysis. It has extremely strong engineering application value. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating the overall process architecture of the present invention.

[0043] Figure 2 This is an architecture diagram of the dual-source evidence extraction and fusion module of the present invention;

[0044] Figure 3 This is an architecture diagram of the evidence verification and correction module of the present invention;

[0045] Figure 4 This is a diagram of the global alignment and fusion module architecture of the present invention;

[0046] Figure 5 This is a structural diagram of the electronic device provided by the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] It should be noted that, unless otherwise specified, the features in the following embodiments and implementation methods can be combined with each other.

[0049] Firstly, this invention provides a method for mining gene variation evidence based on multi-stage credible reasoning, such as... Figure 1 As shown, it includes the following steps:

[0050] S1. Structured semantic segmentation of documents:

[0051] The academic PDF document to be parsed is obtained, and the full text of the document is extracted. The full text is segmented using a chapter-aware adaptive semantic segmentation strategy combined with the chapter structure of the paper. After segmentation, a unique identifier, the chapter attribute to which it belongs, and the start and end position information of the text are configured for each text block, resulting in multiple structural text blocks that are complete in structure and independent in meaning.

[0052] Specifically, the chapter-aware adaptive fragmentation strategy is as follows: based on the fixed chapter structure of academic literature, including abstract, experimental methods, experimental results, and discussion summary, different functional chapters of the literature are distinguished, and chapter boundaries are given priority as fragmentation nodes to ensure that a single text block carries only the content of a single chapter or a complete set of biological experimental data, thus ensuring the integrity and independence of experimental semantics.

[0053] Specifically, the segmentation process involves: dividing the text into blocks according to the structure of the literature chapters, and controlling the number of characters in each text block within a preset threshold range; during the segmentation process, avoiding text truncation issues for tables and figures to ensure the semantic integrity of gene variation-related experiments; and merging adjacent texts to complete semantic completion for short text blocks with insufficient character length.

[0054] S2, Extraction of Evidence of Mutation

[0055] The structured text blocks obtained in step S1 are subjected to table-type variation evidence extraction and text-type variation evidence extraction respectively to obtain the first type of gene variation evidence set and the second type of gene variation evidence set.

[0056] Specifically, the extraction of tabular variant evidence is as follows:

[0057] like Figure 2 As shown, all structured text blocks are traversed to identify the original structured tables and image parsing tables within the structured text blocks. Gene mutation sites, experimental attributes, functional characterizations, and quantitative experimental values ​​recorded in the original structured tables and image parsing tables are extracted, preserving the complete table row structure and the corresponding relationships between mutation sites and experimental data, thus forming a first-type gene mutation evidence set derived from the table data.

[0058] The extraction of text-based mutation evidence specifically involves:

[0059] Semantic recognition is performed on each structured text block to screen for various gene mutation markers, such as HGVS canonical format, protein format, and rsID format. Using sentences containing these gene mutation markers as the core, the text scope is expanded bidirectionally using contextual semantic association rules. This involves forward matching of pre-existing conditions such as experimental system, equipment, and environment, and backward extraction of post-existing information such as experimental results and functional characterization. Single short sentences are expanded into complete experimental-level textual evidence fragments, which are then aggregated to form a second type of gene mutation evidence set derived from the main text.

[0060] S3. Fusion of multi-source heterogeneous evidence:

[0061] For the first type of gene variation evidence set and the second type of gene variation evidence set, an initial gene variation evidence list is generated based on the gene variation identifier.

[0062] Specifically, the first and second sets of gene variation evidence obtained in step S2 are aggregated, and the heterogeneous evidence scattered in the tables and text is grouped and associated based on the gene variation identifiers. Multi-dimensional experimental data corresponding to the same variation site are integrated, the original literature source information corresponding to each piece of evidence is fully preserved, and an initial list of gene variation evidence is generated after unifying the data format.

[0063] S4. Evidence Standardization:

[0064] The initial gene variation evidence list is standardized and unified to output a concise, unified, and standardized evidence list.

[0065] Specifically, the standardization and unification process involves: converting all gene variation identifiers into the HGVS standard format, unifying the units of measurement and expression of quantitative data such as experimental fold changes and percentages; and simultaneously eliminating invalid evidence with semantic repetition and data redundancy by combining semantic similarity algorithms and experimental data equivalence judgment rules.

[0066] S5. Comprehensive Evidence Tracing and Verification:

[0067] like Figure 3 As shown, for each piece of evidence in the standardized evidence list, a dual evidence verification mechanism is used to verify it, screen and mark abnormal evidence such as lack of original text support, numerical deviation, semantic distortion, and incorrect site association, count the types and details of abnormalities, and generate a standardized evidence verification report.

[0068] Furthermore, the dual evidence verification mechanism specifically combines precise matching of the original literature with fuzzy semantic verification to verify the evidence from four dimensions: source traceability validity, variant matching accuracy, experimental data consistency, and experimental scenario adaptability. Specifically, the dual evidence verification mechanism includes original literature verification, unique source traceability verification of gene variants, precise matching verification of quantitative experimental data, and experimental scenario consistency verification. This comprehensive screening and elimination of invalid, distorted, and mismatched abnormal evidence ensures the credibility of the output data.

[0069] S6, Correction of precise evidence without hallucinations:

[0070] Based on the standardized evidence verification report, all abnormal evidence was located, and the evidence of no hallucination was accurately corrected by strictly adhering to the original text of the literature, resulting in a list of evidence of gene mutation.

[0071] The hallucination-free, precise evidence correction method only checks and supplements the existing information in the original text, without generating or expanding semantic content, thus completely avoiding the problem of data fabrication. After completing the correction of all abnormal evidence, a precise and effective list of gene mutation evidence is obtained.

[0072] S7, Cross-shard evidence alignment:

[0073] like Figure 4 As shown, the algorithm compiles a list of gene variant evidence corresponding to all structured text blocks and groups them globally according to gene variant sites. It aligns experimental evidence of the same gene variant site scattered across multiple blocks and sources, outputting a multi-source aligned list of gene variant evidence.

[0074] S8, Global Deduplication and Merging:

[0075] Based on the multi-source aligned gene variation evidence list, global deduplication and fusion are completed by combining semantic equivalence and numerical equivalence rules, unifying the output format of all evidence, and finally generating a standardized, highly accurate, and comprehensive gene variation functional evidence dataset.

[0076] Furthermore, this method is also equipped with a system fault tolerance optimization mechanism: automatic retry logic is set for model inference timeout failures; adaptive repair is completed based on preset standard data templates for structured data parsing errors; and semantic compensation is completed through cross-segment semantic association matching to address the context semantic breakage problem caused by text fragmentation, thus ensuring the overall stability of the system operation.

[0077] Secondly, the present invention provides a gene variation evidence mining system based on multi-stage credible reasoning, used to implement the gene variation evidence mining method. The system includes a document segmentation module, a dual-source evidence extraction and fusion module, an evidence standardization module, an evidence verification and correction module, and a global alignment and fusion module, which are sequentially connected in communication.

[0078] The document segmentation module is used to complete adaptive semantic segmentation of academic documents, output structured and traceable structured text segments, avoid long text reasoning failures, and solve the problems of long text parsing timeout and context loss.

[0079] The dual-source evidence extraction and fusion module is connected to the document segmentation module to simultaneously extract the table experimental evidence and the main text experimental evidence within the structured text segmentation, construct a dual-source evidence set, collect and integrate dual-source heterogeneous evidence, complete the grouping and association of homogeneous variant evidence, and retain the source information of all evidence.

[0080] The evidence standardization module is used to unify the gene variation identifier format and experimental data units, clean up redundant and duplicate evidence, and complete the data standardization process.

[0081] The evidence verification and correction module is used to verify the validity and accuracy of evidence from multiple dimensions, identify abnormal data and generate a verification report, and accurately repair abnormal evidence based on the verification report, supplement missing experimental details, and prevent the model from generating false data.

[0082] The global alignment and fusion module is used to complete cross-segment evidence alignment, global deduplication and fusion, and output a standardized, high-quality gene variation functional evidence dataset.

[0083] The technical solution of the present invention will be described more clearly and completely below with reference to specific embodiments.

[0084] Example:

[0085] This embodiment discloses a gene variation evidence mining method based on multi-stage credible reasoning, which is applied to the automated analysis of variation function evidence in clinical medical gene research papers. The specific implementation steps are as follows:

[0086] 1. Document segmentation and processing:

[0087] Import PDF papers on clinical gene mutation research, extract the full text, split it into semantic blocks based on the chapter structure, keep the number of characters in each block to around 20,000 characters, ensure that the experimental data in tables and mutation descriptions are not truncated, and generate structured text blocks with chapter and position identifiers.

[0088] 2. Extraction and fusion of multi-source evidence:

[0089] The experimental tables and main text within each structured text block are analyzed separately to identify variant sites such as I363L, R409H, and P94L corresponding to genes such as SLC26A4 and CYP11B1. Quantitative experimental data such as enzyme activity, transport capacity, and transcription level are extracted. The experimental evidence of gene variant sites in each block is collected and integrated from the tables and text, and the homologous variant evidence is grouped and associated to retain the source information of all evidence.

[0090] 3. Evidence standardization:

[0091] Standardized evidence is obtained by unifying the variant format of tabular and textual experimental evidence for the same variant site according to the HGVS standard, standardizing the units of percentage and fold change of experimental results, and removing semantically redundant evidence.

[0092] 4. Evidence verification and correction:

[0093] Each standardized piece of evidence was verified against its original source, identifying issues such as numerical mismatches and missing details. Based on the original text, the evidence content was accurately corrected to obtain a list of gene mutation evidence, supplementing key information such as experimental cell systems and experimental methods, while also correcting manually labeled data errors.

[0094] 5. Global deduplication and output merging:

[0095] By integrating all the gene variation evidence lists in the blocks, aligning and merging multiple experimental evidences for the same gene variation, removing duplicate records, and finally outputting a standardized, traceable, and highly accurate gene variation functional evidence dataset.

[0096] Through actual literature testing and verification, this embodiment mines and analyzes 1,000 standard gene variation experimental evidences from 65 articles, achieving an evidence recall rate of up to 95%. It can effectively screen out 41.61% of invalid and abnormal evidence, batch correct literature analysis biases and manual annotation errors, and completely solve the practical problems of missed detections, false detections, data redundancy, and model fraud in traditional mining schemes. Its stability and accuracy meet the standards for commercial application.

[0097] like Figure 5 As shown, this application provides an electronic device including a memory 101 for storing one or more programs and a processor 102. When the one or more programs are executed by the processor 102, they implement the method as described in any of the first aspects above.

[0098] The system also includes a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected directly or indirectly to each other to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, and the processor 102 executes various functional applications and data processing by executing the software programs and modules stored in the memory 101. The communication interface 103 can be used for signaling or data communication with other node devices.

[0099] The memory 101 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0100] The processor 102 can be an integrated circuit chip with signal processing capabilities. The processor 102 can be a general-purpose processor 102, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0101] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can also be implemented in other ways. The method and system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0102] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0103] On the other hand, embodiments of this application provide a computer-readable storage medium storing a computer program thereon. When executed by processor 102, the computer program implements the methods described in any of the first aspects above. If the functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0104] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only.

[0105] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for mining gene variation evidence based on multi-stage credible reasoning, characterized in that, Includes the following steps: S1: Obtain the academic document to be parsed, extract the full text of the document, and use a chapter-aware adaptive semantic segmentation strategy to segment the full text to obtain structured text blocks. S2: Extract table-type variation evidence and text-type variation evidence from the structured text blocks respectively to obtain the first type of gene variation evidence set and the second type of gene variation evidence set; S3: For the first and second sets of gene variation evidence, generate an initial list of gene variation evidence based on gene variation identifiers; S4: Standardize the initial list of evidence for gene mutations to generate a standardized list of evidence; S5: Verify each piece of evidence in the standardized evidence list using a dual evidence verification mechanism, screen and mark abnormal evidence, and generate a standardized evidence verification report; S6: Based on the evidence verification report, locate all abnormal evidence, perform accurate correction of non-hallucination evidence, and obtain a list of gene mutation evidence after completing the repair of all abnormal evidence. S7: Collect a list of gene variation evidence corresponding to all structured text blocks, perform global grouping based on gene variation sites, align multiple blocks and multiple sources of evidence corresponding to the same gene variation site, and output a multi-source aligned list of gene variation evidence. S8: Based on a multi-source aligned list of gene variation evidence, global deduplication and fusion are performed by combining semantic equivalence and numerical equivalence rules, unifying the output format of all evidence, and finally generating a gene variation functional evidence dataset.

2. The gene variation evidence mining method based on multi-stage credible reasoning according to claim 1, characterized in that, In step S1, the chapter-aware adaptive semantic segmentation strategy includes: relying on the standardized chapter structure of academic literature, distinguishing different functional chapters of the literature, prioritizing chapter boundaries as segmentation nodes, and ensuring that each text block independently carries a single complete biological experimental data or chapter content. The segmentation process specifically involves: dividing the text into blocks based on the document's chapter structure, controlling the number of characters in each block within a preset threshold range; during the segmentation process, avoiding text truncation issues for tables and figures to ensure the semantic integrity of gene variation-related experiments; for short text blocks with insufficient character length, merging adjacent texts to complete semantics; and configuring a unique identifier, chapter attribute, and start and end position information for each text block after segmentation to obtain the structured text blocks.

3. The gene variation evidence mining method based on multi-stage credible reasoning according to claim 1, characterized in that, In step S2, the extraction of tabular variant evidence specifically involves: traversing all structured text blocks, identifying various types of tabular data within the structured text blocks, mining the gene variant information, experimental attribute descriptions, and quantitative experimental values ​​contained in the tables, preserving the table row structure and the corresponding relationship between variant sites and experimental data, and constructing a first-type gene variant evidence set originating from tables; wherein, the various types of tabular data include native structured tables and image parsing tables; the extraction of textual variant evidence specifically involves: performing semantic recognition on each structured text block, screening for various gene variant identifiers contained in the text; using the statements containing the gene variant identifiers as core units, bidirectionally expanding the text boundaries based on contextual semantic association rules, matching the corresponding experimental environment, experimental conditions, and experimental result information, generating complete experimental-level textual evidence fragments, and constructing a second-type gene variant evidence set originating from the main text.

4. The gene variation evidence mining method based on multi-stage credible reasoning according to claim 1, characterized in that, In step S3, generating the initial gene variation evidence list specifically involves: aggregating the first type of gene variation evidence set and the second type of gene variation evidence set; associating and grouping the heterogeneous evidence scattered in the table and text based on the gene variation identifier; integrating multi-dimensional experimental evidence corresponding to the same gene variation; retaining the original literature source information of each piece of evidence; and obtaining the initial gene variation evidence list after unifying the data format.

5. The gene variation evidence mining method based on multi-stage credible reasoning according to claim 1, characterized in that, In step S4, the standardization and unification process specifically involves: converting all gene variation identifiers into the HGVS standard format, and unifying the units of measurement and expression of quantitative data; simultaneously, combining semantic similarity algorithms and experimental data equivalence judgment rules to eliminate invalid evidence with semantic repetition and data redundancy; wherein, the quantitative data includes experimental fold change and percentage.

6. The gene variation evidence mining method based on multi-stage credible reasoning according to claim 1, characterized in that, In step S5, the dual evidence verification mechanism specifically involves combining the precise matching of the original document text with the fuzzy semantic verification mechanism to verify the evidence from four dimensions: source traceability effectiveness, mutation matching accuracy, experimental data consistency, and experimental scenario adaptability. The dual evidence verification mechanism includes verification of the original literature, verification of the unique source of gene variation, verification of precise matching of quantitative experimental data, and verification of consistency of experimental scenario. The standardized evidence verification report includes the anomaly type and anomaly details.

7. The gene variation evidence mining method based on multi-stage credible reasoning according to claim 1, characterized in that, In step S6, the hallucination-free precise evidence correction only corrects and supplements the existing information in the original text, without automatically generating or expanding semantic content.

8. A gene mutation evidence mining system based on multi-stage credible reasoning according to any one of claims 1 to 7, characterized in that, It includes a document segmentation module, a dual-source evidence extraction and fusion module, an evidence standardization module, an evidence verification and correction module, and a global alignment and fusion module, which are connected in sequence. The document segmentation module is used to complete adaptive semantic segmentation of academic documents, output structured and traceable structured text segments, and avoid long text reasoning failures. The dual-source evidence extraction and fusion module is connected to the document segmentation module to simultaneously extract the table experimental evidence and the text experimental evidence within the structured text segment, construct a dual-source evidence set, and then aggregate the dual-source heterogeneous evidence to complete the association grouping of homologous variation evidence and the retention of source information. The evidence standardization module is used to unify the gene mutation identification format and data standards, and to clean up redundant evidence. The evidence verification and correction module is used to verify the validity of evidence from multiple dimensions, output an anomaly verification report, and accurately correct abnormal evidence based on the verification report to avoid model illusion. The global alignment and fusion module is used to complete evidence alignment, global deduplication and fusion across blocks, and output a standardized, high-quality gene variation functional evidence dataset.

9. An electronic device comprising a memory and a processor, characterized in that, The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the gene variation evidence mining method based on multi-stage credible reasoning as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the gene variation evidence mining method based on multi-stage credible reasoning as described in any one of claims 1-7.

Images