Gene mutation point screening method and device based on clinical phenotype, terminal and medium

Abstract

The present invention provides a method, apparatus, terminal, and medium for screening gene mutation points based on clinical phenotypes. The method includes: identifying candidate genes based on gene detection results, clinical text to be analyzed, and a multi-source database; calculating the basic weights of a first pathway; extracting candidate mutation sites within the genomic regions of the candidate genes and obtaining external evidence corresponding to the candidate mutation sites; performing phenotypic extraction on the clinical text to be analyzed to obtain a structured phenotype; inputting the structured phenotype, contextual information of the candidate mutation sites, and external evidence into a trained mutation site screening model to obtain intermediate semantic analysis results; determining the basic weights of a second pathway based on the intermediate semantic analysis results; obtaining a comprehensive weight for each candidate mutation site based on the basic weights of the first and second pathways; and screening the candidate mutation sites based on the comprehensive weights to obtain screening results. This application can perform site-level screening, improving the accuracy of gene mutation point screening.

Description

Technical Field

[0001] This invention relates to the field of gene detection technology, and in particular to a method, device, terminal, and medium for screening gene mutation points based on clinical phenotypes. Background Technology

[0002] With the continuous development of gene testing technology, its applications are becoming increasingly widespread. Especially with the application of technologies such as high-throughput sequencing, whole-exome sequencing, and whole-genome sequencing, hundreds or even thousands of candidate mutation sites can typically be detected in subject samples. However, not all of these candidate mutation sites are directly related to the subject's current disease phenotype or health status; a significant portion may be background variations, low-relevance variations, or variations with unclear clinical significance. Therefore, how to efficiently and accurately screen candidate mutation sites closely related to the subject's clinical manifestations from a large number of test results has become a key technical issue in gene testing result analysis.

[0003] Existing phenotype-based screening methods generally remain at the gene-level screening level, outputting candidate genes or gene-level association scores. However, in actual clinical applications, the same gene often contains multiple candidate mutation sites, and different sites may have significant differences in functional impact, inheritance patterns, literature evidence, and clinical relevance. Therefore, completing gene-level screening alone is often insufficient to directly support the refined determination of pathogenic mutation points, thereby reducing the accuracy of gene mutation point screening.

[0004] Therefore, existing technologies have shortcomings and need to be improved and developed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, device, terminal and medium for screening gene mutation points based on clinical phenotype, in order to address the above-mentioned deficiencies of the prior art and solve the problem of low accuracy in gene mutation point screening in the prior art.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows: A method for screening gene mutation points based on clinical phenotype, comprising: The system acquires the gene detection results and clinical text to be analyzed, identifies candidate genes based on the gene detection results, clinical text to be analyzed, and a pre-constructed multi-source database, and calculates the basic weight of the first pathway for each candidate gene. Obtain gene sequencing data corresponding to the clinical text to be analyzed; based on the gene sequencing data, extract candidate mutation sites within the genomic region corresponding to the candidate gene, and obtain external evidence corresponding to the candidate mutation sites. Phenotypic extraction is performed on the clinical text to be analyzed to obtain a structured phenotype. The structured phenotype, the contextual information of the candidate mutation sites, and external evidence are input into a trained mutation site screening model to obtain intermediate semantic analysis results. The basic weights of the second pathway are determined based on the intermediate semantic analysis results. The comprehensive weight of each candidate mutation site is obtained based on the basic weight of the first pathway and the basic weight of the second pathway. The candidate mutation sites are then screened based on the comprehensive weight to obtain the screening results.

[0007] In one embodiment of this application, the process involves acquiring the gene detection results and the clinical text to be analyzed, identifying candidate genes based on the gene detection results, the clinical text, and a pre-constructed multi-source database, and calculating the basic weight of the first pathway for each candidate gene, including: The system acquires the gene detection results and clinical text to be analyzed, and loads a pre-constructed multi-source database to obtain the gene-disease correspondence and gene-phenotype correspondence. Extract the disease information and phenotypic information to be analyzed from the clinical text to be analyzed; Each mutated gene in the gene detection results is traversed, and the associated disease information and associated phenotype information corresponding to each mutated gene are obtained from the gene-disease correspondence and gene-phenotype correspondence. If the phenotypic information to be analyzed matches the associated phenotypic information, the mutated gene will be used as a candidate gene. The phenotypic information associated with each candidate gene is compared with the phenotypic information to be analyzed by calculating the string similarity to obtain a phenotypic similarity score. The disease information associated with each candidate gene is compared with the disease information to be analyzed by calculating the string similarity to obtain a disease similarity score. The phenotypic similarity score and the disease similarity score are weighted and fused to obtain the first pathway basic weight for each candidate gene.

[0008] In one embodiment of this application, the phenotypic similarity score and the disease similarity score are weighted and fused to obtain the first pathway basic weight for each candidate gene, including: Based on the relationship between disease similarity scores and preset scoring thresholds, the first weight corresponding to the disease similarity score is determined; If the associated phenotypic information and associated disease information are found to be related based on the preset phenotypic-disease correspondence, then the phenotypic similarity score is assigned a second weight. If the associated phenotypic information and associated disease information obtained based on the preset phenotypic-disease correspondence are not related, then a third weight is assigned to the phenotypic similarity score. The disease similarity score is weighted based on the first weight to obtain a first weighted value, and the phenotypic similarity score is weighted based on the second weight or the third weight to obtain a second weighted value. The sum of the first weighted value and the second weighted value is used as the first pathway base weight for each candidate gene; The second weight is greater than the third weight.

[0009] In one embodiment of this application, determining the basic weights of the second path based on the intermediate semantic analysis results includes: Based on preset semantic matching rules and evidence level evaluation mechanism, the intermediate semantic analysis results are mapped into semantic support scores and evidence levels. The semantic support score and the evidence level are multiplied together to obtain the basic weight of the second pathway.

[0010] In one embodiment of this application, based on preset semantic matching rules and an evidence level evaluation mechanism, the intermediate semantic analysis results are mapped to semantic support scores and evidence levels, including: Based on preset semantic matching rules, the current semantic support level in the intermediate semantic analysis results is determined; Find the correspondence between the preset semantic support level and the semantic support score to obtain the semantic support score corresponding to the current semantic support level; The evidence level is determined based on the evidence level evaluation mechanism to determine the support strength and verifiability of external evidence in the intermediate semantic analysis results, and the evidence level is obtained based on the support strength and verifiability.

[0011] In one embodiment of this application, the training steps of the mutation site screening model include: We collect standard gene mutation points and their corresponding standardized phenotypic terms, select whole genome sequences of healthy individuals from publicly available population genome data as genetic background templates, and insert target gene mutation points into the corresponding genomic coordinates to construct virtual patient whole genome samples.

[0012] Based on the virtual patient whole genome sample, the pre-trained large language model is fine-tuned to obtain a trained mutation site screening model. The training tasks of the mutation site screening model include: gene-level association learning tasks and site-level training tasks.

[0013] In one embodiment of this application, the contextual information of the candidate mutation site includes at least one of the following: the gene where the candidate mutation site is located, genomic coordinates, mutation type, amino acid change, transcript information, protein functional domain location, information related to the predetermined genetic pattern, population frequency information, functional impact prediction results, and database annotation information. The external evidence includes at least one of the following: literature evidence fragments, case summaries, database entries, descriptions of functional experiments, descriptions of disease mechanisms, and family information; The intermediate semantic analysis results include at least one of the following: candidate mutation site identification, core phenotype support, auxiliary phenotype support, consistency of disease mechanism or genetic pattern, semantic contradiction information, summary of supporting reasons, explanation of evidence sources, and preliminary support level.

[0014] In one embodiment of this application, the comprehensive weight of each candidate mutation site is obtained based on the first pathway basic weight and the second pathway basic weight, including: The first pathway base weight of each candidate gene is assigned to all candidate mutation sites located on the candidate gene, so that the candidate mutation sites have the same first pathway base weight as their respective genes. The basic weights of the first and second pathways corresponding to each candidate mutation site are added together to obtain the comprehensive weight of each candidate mutation site.

[0015] In one embodiment of this application, the method further includes: Based on the mutation site screening model, the structured results, matching criteria, evidence summaries and ranking results generated during the screening process are semantically integrated and organized into natural language to generate a gene interpretation report in a predetermined format. The gene interpretation report is used to present the basis, evidence sources and reasoning logic for the ranking results.

[0016] This application also provides a gene mutation point screening device based on clinical phenotype, comprising: The first weight determination module is used to obtain the gene detection results and clinical text to be analyzed, determine candidate genes based on the gene detection results, clinical text to be analyzed and a pre-constructed multi-source database, and calculate the first pathway basic weight of each candidate gene. The extraction module is used to acquire gene sequencing data corresponding to the clinical text to be analyzed, and based on the gene sequencing data, extract candidate mutation sites in the genomic region corresponding to the candidate gene, and obtain external evidence corresponding to the candidate mutation sites. The input module is used to extract the phenotype of the clinical text to be analyzed to obtain a structured phenotype. The structured phenotype, the contextual information of the candidate mutation sites, and external evidence are input into a trained mutation site screening model to obtain intermediate semantic analysis results. The second weight determination module is used to determine the basic weights of the second path based on the intermediate semantic analysis results. The sorting module is used to obtain a comprehensive weight for each candidate mutation site based on the basic weights of the first and second pathways, and to filter the candidate mutation sites based on the comprehensive weights to obtain the screening results.

[0017] This application also provides a terminal, including: a memory, a processor, and a clinical phenotype-based gene mutation screening program stored in the memory and executable on the processor, wherein the clinical phenotype-based gene mutation screening program, when executed by the processor, implements the steps of the clinical phenotype-based gene mutation screening method as described above.

[0018] This application also provides a computer-readable storage medium storing a computer program that can be executed to implement the steps of the clinical phenotype-based gene mutation point screening method as described above.

[0019] The present invention provides a method, apparatus, terminal, and medium for screening gene mutation points based on clinical phenotypes. The method includes: acquiring gene detection results and clinical text to be analyzed; determining candidate genes based on the gene detection results, the clinical text, and a pre-constructed multi-source database, and calculating the basic weight of a first pathway for each candidate gene; acquiring gene sequencing data corresponding to the clinical text; extracting candidate mutation sites within the genomic region corresponding to the candidate genes based on the gene sequencing data, and acquiring external evidence corresponding to the candidate mutation sites; performing phenotypic extraction on the clinical text to obtain a structured phenotype; inputting the structured phenotype, contextual information of the candidate mutation sites, and external evidence into a trained mutation site screening model to obtain intermediate semantic analysis results; determining the basic weight of a second pathway based on the intermediate semantic analysis results; obtaining a comprehensive weight for each candidate mutation site based on the first and second pathway basic weights; and screening candidate mutation sites based on the comprehensive weights to obtain screening results. This application, by screening candidate mutation sites, completes site-level screening, which can support the refined determination of pathogenic mutation points, thereby improving the accuracy of gene mutation point screening. Attached Figure Description

[0020] Figure 1 This is a flowchart of a preferred embodiment of the gene mutation point screening method based on clinical phenotype in this invention.

[0021] Figure 2 This is an overall framework diagram of a preferred embodiment of the gene mutation point screening method based on clinical phenotype in this invention.

[0022] Figure 3 This is a schematic diagram illustrating the logical principle of a preferred embodiment of the gene mutation point screening method based on clinical phenotype in this invention.

[0023] Figure 4 This is a functional principle block diagram of a preferred embodiment of the gene mutation point screening device based on clinical phenotype in this invention.

[0024] Figure 5 This is a functional principle block diagram of the terminal in this invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0026] Current phenotype-based screening methods still have significant shortcomings. On the one hand, existing methods typically rely heavily on pre-constructed gene-phenotype databases, disease-phenotype databases, or knowledge graphs. The knowledge content in these databases often has long update cycles, making it difficult to incorporate newly published literature, new case reports, and the latest discovered gene-phenotype associations in a timely manner. When faced with rapidly evolving real-world clinical scenarios, these methods are prone to insufficient knowledge coverage and reduced recall capabilities. On the other hand, existing technologies largely remain at the gene-level screening level, primarily outputting candidate genes or gene-level association scores. However, in actual clinical applications, a single gene often contains multiple candidate mutation sites, and different sites may exhibit significant differences in functional impact, inheritance patterns, literature evidence, and clinical relevance. Therefore, simply completing gene-level screening is often insufficient to directly support the refined determination of pathogenic mutations. Furthermore, the output of existing technologies is mostly based on candidate lists or ranking scores, lacking a systematic integration of screening criteria, evidence sources, and judgment logic, making it difficult to further develop gene interpretation reports suitable for clinical use. Meanwhile, although some technologies have begun to incorporate large language models in recent years to enhance knowledge supplementation and semantic analysis capabilities, the performance of general-purpose large language models in specific tasks such as candidate variant screening, site-level determination, and clinical interpretation remains limited due to the lack of high-quality clinical cohort data that matches specific clinical tasks for training or adaptation. Their stability and interpretability also need improvement. In particular, in real-world clinical scenarios, high-quality labeled samples often fail to simultaneously meet conditions such as "clear pathogenic sites, complete phenotypic information, available candidate interference sites, and labels usable for supervised training." Furthermore, due to privacy protection and data sharing restrictions, the amount of case data directly usable for model training is relatively limited. On the other hand, real-world cases commonly involve multiple coexisting candidate sites, incomplete or non-standard phenotypic expression, and lagging external knowledge updates, making it difficult for general-purpose models, even after minor adjustments with general data, to directly possess the fine-grained discrimination capabilities required for candidate mutation site screening tasks.

[0027] Specifically, regarding the screening and identification of pathogenic genes or mutations related to hereditary diseases, existing technologies can be broadly divided into gene detection technologies centered on experimental testing or sequencing, and computational analysis technologies that, based on test results and combined with patient clinical phenotypic information, screen, match, and rank candidate genes. These include: First, traditional gene detection and screening technologies. Traditional gene detection technologies mainly include polymerase chain reaction (PCR), gene chips, and high-throughput sequencing. PCR technology typically amplifies and detects known target sites or fragments, offering advantages such as high sensitivity, strong specificity, and a mature operational procedure, making it suitable for confirmatory testing of specific genes or sites. Gene chip technology can achieve parallel detection of multiple sites within a pre-set probe range, suitable for batch screening of known mutation sites or specific gene sets. High-throughput sequencing technologies, especially whole-exome sequencing and whole-genome sequencing, can obtain large-scale sequence information, discover candidate genes and candidate variant sites, and have become important means of genetic disease detection and variant discovery.

[0028] Second, gene screening technologies based on clinical phenotypes. In recent years, a class of computational analysis methods has emerged that use patient clinical phenotype information to further screen, prioritize, and assist in interpreting candidate genes or variants. These technologies typically use patient symptom descriptions, physical signs, clinical diagnoses, examination results, or standardized phenotypic terminology as input. They match the patient's phenotype with existing gene-phenotype, disease-phenotype, or gene-disease association knowledge to narrow down the candidate pool and improve the screening efficiency of pathogenic genes or variants. Typically, based on existing gene-phenotype databases, disease-phenotype databases, or knowledge graphs, they compare the similarity between the patient's clinical phenotype and the set of phenotypes associated with candidate genes, calculate the degree of matching between the patient's phenotype and the target gene or disease, and prioritize candidate genes accordingly. Some schemes combine patient medical record information, standardized phenotypic terminology, and association information from the knowledge base into the matching model to output a phenotypic consistency score or a list of candidate genes. These technologies can, to some extent, alleviate the problem of too many candidate genes or candidate sites and high manual interpretation costs after high-throughput testing.

[0029] However, despite the positive role played by the aforementioned existing technologies in screening for genes or variants related to genetic diseases, both traditional detection techniques and phenotype-based screening and ranking techniques still have significant limitations, mainly in the following aspects: First, traditional gene detection technologies focus more on "detection and discovery," but suffer from high costs and low efficiency in large-scale screening and subsequent interpretation. PCR technology typically requires the design of specific primers for each target gene or site. When the screening scope expands to a large number of candidate genes, the time and resource consumption for primer design, experimental optimization, and detection implementation increase significantly. Furthermore, PCR results can be affected by experimental conditions such as temperature, pH, and primer concentration, leading to false positives or false negatives. While gene chip technology enables parallel detection of multiple sites, chip design and fabrication are technically demanding and costly, and typically rely on pre-known probe sets, thus limiting its ability to detect unknown genes, rare genes, or newly discovered variants. High-throughput sequencing technology, while capable of acquiring large amounts of sequence information at once, usually requires expensive equipment, specialized technical personnel, and complex data analysis workflows. Its results primarily provide information on "which candidate variants were detected." For complex diseases or cases, further integration with clinical information and bioinformatics analysis methods is needed to identify the gene mutations truly relevant to the patient's disease from a vast pool of candidate results. Therefore, traditional gene testing technology is more focused on "data acquisition" and "candidate discovery," and it is difficult to directly meet the needs of "efficiently screening, interpreting and prioritizing candidate results in combination with clinical phenotypes" in complex cases.

[0030] Second, existing phenotype-driven screening methods heavily rely on static knowledge bases, making it difficult to effectively address the challenges of knowledge updates and complex phenotypic expressions. Most existing phenotype-based gene or variant screening methods rely on pre-built gene-phenotype databases, disease-phenotype databases, or knowledge graphs for matching and ranking. While these methods can be effective when the knowledge base is comprehensive and patient phenotypic expressions are relatively standardized, when patient phenotypic descriptions are complex, rare, or non-standardized, or when the existing knowledge base has not yet incorporated the latest research progress or gene-phenotype association information from newly published literature, the screening system often struggles to fully recall the most relevant candidate genes or loci for the patient, thus affecting subsequent ranking and interpretation accuracy. In other words, existing methods generally rely on a relatively static knowledge base, with knowledge updates typically dependent on manual compilation, annotation, or periodic database maintenance. This limited update speed makes it difficult to promptly absorb rapidly growing literature evidence, case reports, and new associations. Therefore, in clinical scenarios characterized by rapid knowledge iteration, scattered evidence, and inconsistent phenotypic descriptions, existing phenotype-driven screening methods still exhibit significant shortcomings in terms of the completeness and timeliness of knowledge coverage.

[0031] Third, existing phenotype-driven screening methods mostly remain at the gene-level screening level, offering insufficient support for refined identification at the mutation site level. In actual clinical testing, multiple candidate mutation sites may exist under the same gene, and different sites may differ significantly in terms of functional impact, inheritance pattern, existing literature reports, population frequency, and consistency with patient clinical phenotypes. Therefore, simply outputting candidate genes or providing gene-level association scores is often insufficient to directly support the accurate screening of pathogenic mutations.

[0032] Fourth, existing technologies still have limitations in terms of interpretability, reportable output, and clinical adaptability. While some current screening methods can output lists of candidate genes or ranking results of candidate variants, they often lack clear and systematic expression of the screening criteria, sources of evidence, relationships between different scoring items, and the process by which the ranking results are formed. This makes it difficult for clinical users to quickly determine the reliability of the system's recommendations when faced with complex cases, and it also hinders subsequent manual review, expert confirmation, and clinical decision support.

[0033] In summary, while existing technologies have made initial progress from gene detection to phenotype-assisted screening, they still have significant shortcomings. Therefore, there is an urgent need for a new technological solution that, based on patients' clinical phenotypes, can leverage large-scale models to enhance knowledge updates and semantic understanding capabilities, enabling further screening and interpretation from candidate genes to candidate mutation sites, and ultimately generating a gene interpretation report that is evidence-supported, logically clear, and easy to use in clinical practice.

[0034] Therefore, it is necessary to construct a training sample generation mechanism that more closely reflects the challenges of real-world clinical interpretation, and on this basis, to develop a model training and inference method for candidate mutation site screening tasks. This would improve the accuracy, stability, and verifiability of candidate variant screening, site-level determination, and result interpretation in complex clinical scenarios. Based on the above issues, the core technical problem addressed by this invention is not simply to achieve phenotype-based pathogenic gene screening, but rather to construct a closed-loop candidate mutation screening technology for real-world clinical interpretation scenarios.

[0035] Specifically, this invention constructs a virtual patient cohort by combining real disease loci and publicly available genomic data for training or adaptation of a large language model. Simultaneously, by incorporating structured knowledge resources, this closed-loop technology can utilize structured knowledge such as gene-phenotype and gene-disease pairs for stable and controllable initial recall of candidate variants. Furthermore, in situations where the knowledge base is insufficient, knowledge updates are lagging, or subject phenotypic expression is non-standard, a constrained large language model is introduced, combining the latest textual evidence, semantic understanding, and reasoning capabilities to supplement the analysis of candidate genes and candidate mutation sites, thereby improving adaptability to complex cases and new knowledge scenarios. The invention further elevates the screening results from the gene level to the mutation site level, fusing, resolving conflicts, and re-ranking multi-source evidence at the site level, ultimately outputting a more clinically relevant report of candidate mutation site results and their interpretive basis. Through the technical solution of this invention, the accuracy, recall, interpretability, and clinical applicability of gene testing results can be improved, providing technical support for precise disease diagnosis, assisted treatment decision-making, and personalized medicine applications. Furthermore, this invention can be applied to gene testing result analysis, auxiliary diagnosis of genetic diseases, bioinformatics interpretation platforms, auxiliary interpretation systems for hospital laboratories / genetic centers, precision medicine analysis platforms, and related scientific research and industrialization projects.

[0036] Gene mutation site screening has significant application value in precision medicine, genetic disease research, drug development, and personalized health management, and is of great importance for improving human health and quality of life. This invention proposes a candidate mutation screening technology driven by clinical phenotypes and combining structured knowledge with the semantic reasoning capabilities of a large language model. This scheme, through a dual-pathway screening and evidence fusion mechanism, achieves further determination from candidate genes to candidate mutation sites, and can generate a gene interpretation report containing candidate sites, relevant phenotypic evidence, literature / knowledge base evidence, and the rationale for the determination. This technological innovation overcomes the limitations of traditional methods, improves screening efficiency and accuracy, and provides strong support for precision medicine and personalized treatment.

[0037] The following description, with reference to the accompanying drawings, illustrates a method, apparatus, system, terminal, and storage medium for screening gene mutation points based on clinical phenotypes according to embodiments of this application. Addressing the issue of low accuracy in gene mutation point screening based on clinical phenotypes in the related technologies mentioned in the background section, this application provides a method for screening gene mutation points based on clinical phenotypes. In this method, the following steps are taken: acquiring the gene detection results and the clinical text to be analyzed; determining candidate genes based on the gene detection results, the clinical text, and a pre-constructed multi-source database, and calculating the basic weight of a first pathway for each candidate gene; acquiring gene sequencing data corresponding to the clinical text to be analyzed; extracting candidate mutation sites within the genomic region corresponding to the candidate genes based on the gene sequencing data, and acquiring external evidence corresponding to the candidate mutation sites; performing phenotypic extraction on the clinical text to be analyzed to obtain a structured phenotype; inputting the structured phenotype, the contextual information of the candidate mutation sites, and the external evidence into a trained mutation site screening model to obtain intermediate semantic analysis results; determining the basic weight of a second pathway based on the intermediate semantic analysis results; obtaining a comprehensive weight for each candidate mutation site based on the first and second pathway basic weights; and screening candidate mutation sites based on the comprehensive weights to obtain screening results. This application achieves site-level screening by screening candidate mutation sites, which supports the refined determination of pathogenic mutation points and thus improves the accuracy of gene mutation point screening.

[0038] Please see Figure 1 , Figure 1 This is a flowchart of the gene mutation point screening method based on clinical phenotype in this invention. For example... Figure 1 As shown in the embodiments of the present invention, the gene mutation point screening method based on clinical phenotype includes the following steps: Step S100: Obtain the gene detection results and clinical text to be analyzed. Based on the gene detection results, clinical text to be analyzed, and pre-constructed multi-source database, determine candidate genes and calculate the first pathway basic weight of each candidate gene.

[0039] First, obtain the subject's genetic testing results and related clinical texts to be analyzed. The genetic testing results may include candidate variant results obtained through whole-exome sequencing, whole-genome sequencing, targeted panel sequencing, or other genetic testing methods. These candidate variant results typically include multiple candidate genes and their corresponding one or more candidate mutation sites. The clinical texts to be analyzed may include, but are not limited to, symptoms, signs, past medical history, family history, laboratory test results, imaging findings, clinical diagnostic conclusions, and standardized phenotypic terminology.

[0040] The acquired clinical phenotype information of the subjects needs to be standardized and structured. Through text cleaning, synonym merging, terminology standardization, and structured field extraction, the free-text clinical descriptions are mapped to a pre-defined set of HPO phenotype terms and OMIM disease feature labels, thus completing the standardization process. This design improves the standardization and computability of phenotype input and reduces the impact of free-text, synonyms, and non-standard descriptions on subsequent screening results.

[0041] This invention uses a large language model to identify semantic and clinical relationships between phenotypes. These relationships may include, but are not limited to: relationships between primary and secondary phenotypes; relationships between core and secondary phenotypes; chronological, disease progression, or severity relationships; co-occurrence, causal, or hierarchical relationships between different phenotypes; and suggestive relationships between certain phenotypes and specific disease directions, systemic involvement directions, or genetic patterns. This design enhances semantic understanding in complex phenotype scenarios, enabling the system to utilize relationships between phenotypes for more clinically logical filtering rather than relying solely on individual phenotype terms. It also provides richer phenotype representations for subsequent structured knowledge matching and clearer contextual input for subsequent LLM inference, reducing uncertainty when the model directly processes raw, messy pathological text.

[0042] After processing, the gene testing results and clinical texts to be analyzed yield phenotypic input results that include at least the following: a standardized terminology set, an annotated summary of clinical phenotypes, semantic or structural relationships between phenotypes, and a set of core and auxiliary phenotypic features relevant to subsequent screening. These results can serve as the common input basis for subsequent structured knowledge screening pathways and large language model supplementary reasoning pathways.

[0043] This application uses pre-defined gene-disease and gene-phenotype databases as the structured knowledge foundation for screening candidate genes and mutation sites. By integrating publicly available phenotypic terms, gene-phenotype relationships, gene-disease relationships, and disease-phenotype relationships, a searchable, verifiable, and sustainably expandable multi-source database is formed, thereby enabling stable recall and preliminary screening of candidate genes and mutation sites in gene testing results. Unlike existing technologies that rely solely on a single database or knowledge graph, this invention employs a multi-source heterogeneous data collaborative knowledge base construction approach. On the one hand, it fully utilizes existing high-reliability relationships in mature databases to ensure the stability and verifiability of the screening process. On the other hand, addressing issues such as outdated knowledge updates, incomplete coverage, and lack of phenotypic records for some genes in existing databases, a large language model is introduced to supplement and annotate missing information, thereby improving the database's completeness, timeliness, and recall capabilities.

[0044] For example, the publicly available data resources collected in this invention include two parts: a phenotype database and a gene-phenotype database. The role of the phenotype database in this invention is mainly reflected in: providing a terminological basis for phenotype annotation and standardization in the subjects' free-text medical records; providing a unified phenotype representation for subsequent gene-phenotype matching; providing support for the mapping between Chinese and English medical phenotype terms; and providing a basis for subsequent standardization of phenotype information extracted by the large language model. The phenotype database constructed in this invention includes PhenoPro and HPO. PhenoPro is derived from the publicly available resource https: / / github.com / jumphone / PhenoPro. This database contains 11,896 phenotype entries, covering English and Chinese phenotype descriptions. The information format is relatively standardized, making it convenient to directly serve as an important source for phenotype standardization and Chinese-English comparison. This invention downloads, cleans, parses, and structures the PhenoPro data, extracting phenotype terms, Chinese names, English names, and their corresponding descriptive information to construct the first part of the phenotype database. Because the phenotype information in PhenoPro is relatively accurate and the terminology correspondence is relatively clear, this part serves as the main source of the phenotype database construction in this invention. The Human Phenotype Ontology (HPO), sourced from https: / / hpo.jax.org / app / , is a standardized vocabulary system for human disease phenotypic abnormalities. It contains over 13,000 terms and numerous annotation relationships related to hereditary diseases, making it a widely used foundational resource in clinical genetic disease and phenotypic standardization analysis. This invention downloaded and parsed phenotypic-related files from the official HPO website, ultimately obtaining 17,875 phenotypic terms / descriptions. Considering that the original HPO data is primarily provided in English, to meet the needs of this invention in a Chinese clinical setting, this invention further utilizes translation tools to map the relevant terms to Chinese, and incorporates the resulting Chinese terms, English terms, and corresponding descriptions into the phenotypic database. In this invention, HPO data primarily serves as an important supplementary source to PhenoPro data.

[0045] On the one hand, HPO provides a relatively standardized and hierarchical phenotypic vocabulary system, which helps to unify the representation of clinical phenotypes in the future; on the other hand, its wide range of terminology coverage can make up for the shortcomings of a single Chinese phenotype database in rare and complex phenotypes.

[0046] For example, the gene-phenotype database records known associations between genes and clinical phenotypes, and is the core database for this invention to perform structured recall and preliminary matching of candidate genes and candidate mutation sites. This database is constructed by integrating multiple publicly available sources, primarily including OMIM, HPO, H2GKBs, and ClinVar. OMIM (Online Mendelian Inheritance in Man) is a commonly used authoritative database in the field of genetic disease research, containing a large amount of textual descriptions of gene-disease associations, gene functions, and disease characteristics. This invention uses OMIM's entry pages for batch acquisition and parsing, for example, source URLs in the form of https: / / omim.org / entry / 131550, where the last number is the corresponding MIM Number. This invention crawled and organized a total of 17,073 records, and further parsed them to obtain 4,863 gene-disease association data. By organizing the disease descriptions and gene text information in OMIM, auxiliary support can be provided for subsequent gene-phenotype relationship inference and candidate gene background interpretation. This invention downloaded the "PHENOTYPE TO GENES" file from the official HPO data page (https: / / hpo.jax.org / app / data / annotations), obtaining 298,384 gene-phenotype relationship data entries. Subsequently, combined with the previously constructed phenotype database, the phenotype entries were standardized in terms of terminology, mapped to Chinese, and structured, ultimately forming standardized gene-phenotype relationship data. This data is characterized by its standardized source, large quantity, and relatively unified terminology system, and is a crucial core component of this invention's gene-phenotype database. H2GKBs (HPO2Gene KnowledgeBase) is a knowledge base resource built around the association between HPO and genes. This invention extracts data entries that can describe the relationship between genes and phenotypes by parsing relevant fields in its knowledge base file, standardizes them, and incorporates them into the gene-phenotype database to further enrich the existing gene-phenotype association coverage. ClinVar is an important database closely related to the clinical significance of variants and their associated phenotypic information. This invention downloaded the clinvar_GRCh37.vcf and clinvar_GRCh38.vcf files, parsed their fields, and extracted relationships to obtain 5103 gene-phenotype correlation data. This data can provide auxiliary support for site-level analysis and can also be used to supplement gene-phenotype association information from clinical variation records.

[0047] By integrating multiple sources such as OMIM, HPO, H2GKBs, and ClinVar, this invention forms a multi-source heterogeneous gene-phenotype database. This database includes direct gene-phenotype relationships, gene-disease relationships, disease description information, and auxiliary information related to the clinical significance of loci. The main functions of this database in this invention include: providing structured phenotypic associations for candidate genes; providing a search basis for preliminary matching between subject clinical phenotypes and candidate genes; providing background support for subsequent screening from gene-level to locus-level screening; and providing a traceable knowledge source for generating the final gene interpretation report.

[0048] In addition to the aforementioned publicly available databases, data generated and extracted based on a large language model is also an important component of the gene-phenotype database of this invention. Unlike existing solutions that rely solely on static knowledge bases, this invention further introduces a large language model to supplement phenotypic information for genes not yet fully covered in existing databases, thereby addressing issues such as knowledge gaps, delayed knowledge updates, and insufficient coverage.

[0049] In this invention, the entire human genome is first traversed and examined to identify genes that lack clear phenotypic records, have insufficient phenotypic records, lack detailed phenotypic descriptions, or have not yet established standardized gene-phenotype relationships in existing structured databases. These genes are then prioritized for supplementation to the large language model and completion of phenotypic information. This invention can utilize large language models, such as Wenxin Yiyan, ChatGLM, or other large models with medical text understanding, information extraction, and natural language generation capabilities, to perform clinical phenotypic extraction tasks on the target genes. Specifically, the gene name, gene aliases, related disease keywords, associated information from existing databases, and necessary prompt templates are used as input. The large language model outputs clinical phenotypic information related to the gene. The output includes, but is not limited to: possible clinical phenotypes related to the gene; disease manifestations or systemic involvement characteristics; common symptoms, typical signs, or clinical descriptions; source descriptions, contextual hints, or evidence summaries related to the above phenotypes.

[0050] After obtaining the model output, this invention structures and organizes the results, extracting candidate "gene-phenotype" relationships to construct supplementary gene-phenotype knowledge entries. Considering that the output of the large language model may contain free expression, diverse descriptions, or inconsistent granularity, this invention further standardizes the model output. Specifically, this includes: 1. Mapping the phenotypic descriptions output by the model to standardized phenotypic terms in the aforementioned phenotypic database; 2. Merging synonyms, near-synonyms, or expressions in different languages; 3. Filtering out results that clearly do not conform to medical phenotypic expression standards or lack sufficient evidence; 4. Adding source markers to retained results, indicating that the relationship was supplemented and generated by the large language model, and recording the model name, generation time, input prompts, or other traceable information; 5. Manually verifying the above traceability information, and adding manual confirmation markers to results with high reliability.

[0051] Through the above processing, the results generated by the large language model can be transformed into standardized gene-phenotype knowledge entries that are searchable, callable, and traceable, and incorporated into the knowledge system of this invention as a supplement to existing multi-source databases. Compared with existing technologies that rely solely on static knowledge bases for phenotype-driven screening, this invention, by introducing the large language model into the database supplementation process, achieves continuous expansion and dynamic updating of the structured knowledge base. This allows it to better adapt to the constantly emerging new knowledge, the lack of information on rare genes, and the interpretation needs of complex clinical phenotype scenarios, thereby enhancing the applicability and accuracy of this invention in candidate gene screening, candidate site determination, and subsequent gene interpretation.

[0052] For example, the screening of the entire gene mutation site in this application is as follows: Figure 2 As shown. After genetic testing, subjects will obtain hundreds or even thousands of mutation points. The primary task in generating a gene interpretation report is to screen these mutation points. An important part of this screening is based on the subject's clinical phenotype, identifying mutation points consistent with the clinical phenotype. This invention constructs a structured knowledge base. In practical applications, such as... Figure 3 As shown, the first pathway can prioritize the initial screening of candidate genes based on structured knowledge matching results and extract candidate mutation sites within the candidate gene range by combining subject sequencing data. The second pathway, based on LLM-G, further combines the contextual information of candidate mutation sites and supporting evidence materials to perform semantic support judgment, structured interpretation output, and site-level priority analysis on the candidate mutation sites. Thus, the first and second pathways respectively undertake the functions of gene-level coarse screening and site-level fine screening, and finally complete the comprehensive ranking at the candidate mutation site level.

[0053] In this embodiment of the application, step S100 specifically includes: Step S110: Obtain the gene detection results and clinical text to be analyzed, and load the pre-constructed multi-source database to obtain the gene-disease correspondence and gene-phenotype correspondence. Step S120: Extract the disease information and phenotype information to be analyzed from the clinical text to be analyzed; Step S130: Traverse each mutated gene in the gene detection results, and obtain the associated disease information and associated phenotype information corresponding to each mutated gene from the gene-disease correspondence and gene-phenotype correspondence; Step S140: If the phenotypic information to be analyzed matches the associated phenotypic information successfully, then the mutated gene is used as a candidate gene. Step S150: Calculate the string similarity between the associated phenotypic information of each candidate gene and the phenotypic information to be analyzed to obtain a phenotypic similarity score; and calculate the string similarity between the associated disease information of each candidate gene and the disease information to be analyzed to obtain a disease similarity score. Step S160: The phenotypic similarity score and the disease similarity score are weighted and fused to obtain the first pathway basic weight of each candidate gene.

[0054] Specifically, based on a pre-built or continuously maintained structured knowledge base, the standardized clinical phenotypes of subjects are screened using a first pathway. The multi-source database is a structured knowledge base, which may include gene-phenotype relationships, gene-disease relationships, disease-phenotype relationships, genetic pattern information, known locus annotation information, and other structured knowledge entries related to gene interpretation. By matching the clinical phenotypes of subjects with known association information in the structured knowledge base, a set of candidate genes related to the subject's phenotype is obtained, and a first pathway matching result is generated. Its purpose is to fully utilize the gene-phenotype, gene-disease, and disease-phenotype relationships already clearly recorded in existing mature knowledge bases to perform a controllable and verifiable preliminary screening of candidate results. Unlike schemes that rely solely on large models to directly output conclusions, this invention retains the fundamental pathway of structured knowledge screening to ensure high stability and interpretability within the scope of known knowledge coverage. Specifically, disease-phenotype relationships are used to determine whether a phenotype is directly related to the disease and to differentiate phenotype weights accordingly. Genetic pattern information, known locus annotation information, and other structured knowledge entries are mainly used for subsequent report generation.

[0055] For example, the input data for the gene screening algorithm includes: gene testing results, clinical text to be analyzed, and a multi-source database. The output data is a list of candidate genes. First, the gene-disease correspondence and gene-phenotype correspondence in the multi-source database are loaded. These correspondences can also exist in the form of a gene-disease database and a gene-phenotype database. Then, the phenotypic information of the subject is parsed, and each mutated gene in the gene testing results is traversed to obtain the associated disease information and associated phenotypic information of that mutated gene. It is then determined whether the associated phenotypic information of the disease matches the phenotypic information of the subject. If they match, the mutated gene is added to the candidate database. The phenotypes and diseases corresponding to the mutated gene are traversed, and fuzzy string matching is used to calculate the similarity between the phenotype and disease information of the subject. After the traversal is complete, the disease similarity score and phenotypic similarity score are obtained.

[0056] In one embodiment of this application, step S160 specifically includes: Step S161: Based on the relationship between the disease similarity score and the preset score threshold, determine the first weight corresponding to the disease similarity score; Step S162: If the associated phenotypic information and associated disease information are related based on the preset phenotypic-disease correspondence, then assign a second weight to the phenotypic similarity score. Step S163: If the associated phenotypic information and associated disease information obtained based on the preset phenotypic-disease correspondence are not related, then a third weight is assigned to the phenotypic similarity score. Step S164: Weight the disease similarity score based on the first weight to obtain a first weighted value; weight the phenotypic similarity score based on the second weight or the third weight to obtain a second weighted value. Step S165: The sum of the first weighted value and the second weighted value is used as the first pathway basic weight for each candidate gene. The second weight is greater than the third weight.

[0057] In the first pathway, for each candidate gene, the disease information and phenotypic information associated with it are compared with the disease information and phenotypic information extracted from the subject's clinical text to calculate the string similarity score. Phenotypic similarity score Similarity calculation can be implemented based on longest common substring matching, as shown in the following code: def longest_common_substring(s1, s2): m = [[0] * (1 + len(s2)) for i in range(1 + len(s1))] longest, x_longest = 0, 0 for x in range(1, 1 + len(s1)): for y in range(1, 1 + len(s2)): if s1[x - 1] == s2[y - 1]: m[x][y] = m[x - 1][y - 1] + 1 if m[x][y]>longest: longest = m[x][y] x_longest = x else: m[x][y] = 0 return len(s1[x_longest - longest: x_longest]) / len(s2) The above function can be used to calculate the similarity ratio between the string to be matched and the target string in the clinical text. Considering the different importance of disease information matching and phenotypic information matching in clinical interpretation, this invention further assigns different weights to the similarity score. If a gene mutation site matches the subject's disease information, a weight is assigned. If a gene mutation site matches the subject's phenotypic information, and the phenotypic information is directly related to the disease, then a weight is assigned. If phenotypic information is not directly related to the disease, then weights are assigned. .

[0058] For example, when assigning a first weight to the disease similarity score, if the disease similarity score is greater than a preset score threshold, it means that the gene mutation site matches the subject's disease information, and a preset weight, such as 3, is assigned; if the disease similarity score is less than or equal to the preset score threshold, the weight is set to 0.

[0059] In one specific embodiment of the present invention, The final string similarity score (i.e., the base weight of the first pathway) for each gene mutation site is recorded as follows: Therefore, the basic weight of the first pathway at candidate mutation site i on the candidate gene. It can ultimately be expressed as: ; in, Scoring for disease similarity The weight, Scoring phenotypic similarity The weights, where for or .

[0060] After calculating the disease similarity score and phenotypic similarity score for all candidate genes, the basic weight of the first pathway corresponding to each candidate result was obtained. Subsequently, the candidate results are sorted according to their primary pathway weights, with higher weights having higher priority and being ranked earlier, and lower weights having lower priority and being ranked later, thus forming a preliminary candidate gene list for the primary pathway, which is used for subsequent screening of the secondary pathway, comprehensive scoring, and manual interpretation.

[0061] like Figure 1 As shown, the gene mutation point screening method based on clinical phenotype further includes the following steps: Step S200: Obtain gene sequencing data corresponding to the clinical text to be analyzed. Based on the gene sequencing data, extract candidate mutation sites within the genomic region corresponding to the candidate gene and obtain external evidence corresponding to the candidate mutation sites.

[0062] like Figure 1 As shown, the gene mutation point screening method based on clinical phenotype further includes the following steps: Step S300: Phenotypic extraction is performed on the clinical text to be analyzed to obtain a structured phenotype. The structured phenotype, the contextual information of the candidate mutation sites, and external evidence are input into the trained mutation site screening model to obtain intermediate semantic analysis results.

[0063] In this embodiment of the application, the training steps of the mutation site screening model include: We collect standard gene mutation points and their corresponding standardized phenotypic terms, select whole genome sequences of healthy individuals from publicly available population genome data as genetic background templates, and insert target gene mutation points into the corresponding genomic coordinates to construct virtual patient whole genome samples.

[0064] Based on the virtual patient whole genome sample, the pre-trained large language model is fine-tuned to obtain a trained mutation site screening model. The training tasks of the mutation site screening model include: gene-level association learning tasks and site-level training tasks.

[0065] Specifically, this invention provides a screening method for fine-tuning a large language model based on a virtual patient cohort. This invention targets the clinical phenotype-driven candidate mutation site screening task, constructing a training sample of difficult virtual patients containing real genetic background, pathogenic site implantation, and natural background variations. Based on this, the model is trained to develop the ability to recall from the gene level to differentiate and re-rank at the site level. Simultaneously, in the model application stage, instead of directly outputting the final ranking score, it first outputs intermediate semantic analysis results in a fixed format based on structured phenotypes, candidate site context, and external evidence. Then, through preset rules, the intermediate results are mapped to semantic support scores and evidence levels for priority determination of candidate mutation sites.

[0066] Unlike methods that construct training samples solely based on the "phenotype-pathogenic gene" pairing, the virtual patient samples constructed in this invention are not merely for learning the coarse-grained correspondence between phenotypes and pathogenic genes. Instead, they are designed for real-world clinical tasks involving the screening of candidate mutation sites, constructing training samples that more closely approximate the difficulties in actual interpretation. Specifically, this invention collects gene mutations and their corresponding standardized phenotypic terms that have been experimentally verified or clinically agreed upon as pathogenic or potentially pathogenic from authoritative databases such as OMIM, HPO, H2GKBs, and ClinVar, as well as high-quality literature. Then, it selects whole-genome sequences of healthy individuals from publicly available population genomic data as genetic background templates. Using bioinformatics methods, it precisely inserts the target pathogenic or potentially pathogenic variants into the corresponding genomic coordinates to construct virtual patient whole-genome samples. This method, while preserving other natural variations in the background genome, can simulate the complex situation of multiple candidate sites coexisting in real cases. The above sample construction method can effectively overcome the problems of scarcity of high-quality labeled real clinical samples and the fact that the model can only complete gene-level coarse screening and is difficult to perform site-level differentiation. This enables the model to learn how to identify real pathogenic sites among multiple similar candidate sites during the training phase, and provides support for subsequent site-level priority determination, re-sorting and interpretation output.

[0067] Based on the aforementioned virtual patient dataset, this invention further performs supervised fine-tuning, instruction adaptation, or efficient parameter fine-tuning on a basic large language model with biomedical text understanding capabilities to obtain a specialized model LLM-G for screening candidate pathogenic variant sites. Preferably, model training includes at least the following two types of tasks.

[0068] The first category is gene-level association learning tasks. The input is a standardized phenotypic description of a virtual patient, and the output is the corresponding pathogenic gene label or candidate gene ranking result. Through this task, the model learns the association patterns between complex phenotypic combinations and pathogenic genes, thereby enhancing the model's basic understanding of phenotypic-gene relationships and providing prior knowledge support for subsequent semantic support judgment and structured interpretation output of candidate mutation sites. In a preferred embodiment, the initial screening of candidate genes in the practical application stage can be mainly completed by the first pathway, while the gene-level association learning task mainly serves as an auxiliary task for training the second pathway model.

[0069] The second category is a training task involving site-level semantic support judgment and structured interpretation output. Its input includes phenotypic information of a virtual patient, candidate mutation site information, and supporting evidence. Through this task, the model learns to perform semantic support judgment and structured interpretation output for candidate mutation sites according to preset rules, given the site context and supporting evidence, rather than directly generating final ranking scores or judgment conclusions arbitrarily.

[0070] The model outputs support labels for candidate mutation sites in the current case and intermediate semantic analysis results in a pre-formatted format.

[0071] In one implementation, the model training process employs supervised learning. Specifically, the two types of training samples are input into the basic large language model, with the corresponding target labels or target structured outputs serving as supervision signals. A cross-entropy loss function or other suitable optimization objective is used, and gradient descent-like algorithms are employed to update the model parameters, enabling the model to progressively learn the associations between phenotypes, candidate genes, candidate mutation sites, and multi-source evidence. If necessary, instruction template alignment or efficient parameter fine-tuning can also be used to improve the stability of the model's output format and its task adaptability.

[0072] The trained LLM-G is preferably used for the second pathway analysis. In practical applications, the first pathway can prioritize the initial screening of candidate genes based on structured knowledge matching results and extract candidate mutation sites within the candidate gene range by combining subject sequencing data; the second pathway, based on LLM-G, further combines the contextual information of candidate mutation sites and supporting evidence materials to perform semantic support judgment, structured interpretation output, and site-level priority analysis on the candidate mutation sites. Thus, the first and second pathways respectively undertake the functions of gene-level coarse screening and site-level fine screening, and finally complete the comprehensive ranking at the candidate mutation site level.

[0073] In the embodiments of this application, the contextual information of the candidate mutation site includes at least one of the following: the gene where the candidate mutation site is located, genomic coordinates, mutation type, amino acid change, transcript information, protein functional domain location, information related to the predetermined genetic pattern, population frequency information, functional impact prediction results, and database annotation information.

[0074] The external evidence includes at least one of the following: literature evidence fragments, case summaries, database entries, descriptions of functional experiments, descriptions of disease mechanisms, and family information.

[0075] The intermediate semantic analysis results include at least one of the following: candidate mutation site identification, core phenotype support, auxiliary phenotype support, consistency of disease mechanism or genetic pattern, semantic contradiction information, summary of supporting reasons, explanation of evidence sources, and preliminary support level.

[0076] For example, the external evidence may include excerpts of literature evidence, case summaries, database entries, descriptions of functional experiments, descriptions of disease mechanisms, pedigree information, and other supplementary evidence that can be used to support or refute the relevance of candidate mutation sites to the current case.

[0077] like Figure 1 As shown, the gene mutation point screening method based on clinical phenotype further includes the following steps: Step S400: Determine the basic weights of the second pathway based on the intermediate semantic analysis results.

[0078] In one embodiment of this application, step S400 specifically includes: Step S410: Based on the preset semantic matching rules and evidence level evaluation mechanism, the intermediate semantic analysis results are mapped into semantic support scores and evidence levels; Step S420: Multiply the semantic support score and the evidence level to obtain the basic weight of the second pathway.

[0079] For example, based on the intermediate semantic analysis results output by the second pathway, this invention further constructs semantic matching rules to obtain a score. For each candidate mutation site, the large language model can output a corresponding semantic support score S based on its ability to interpret the subject's disease information and clinical phenotype. i Furthermore, to further demonstrate the traceability and reliability of the model output, this invention also introduces an evidence level evaluation mechanism. The evidence level E... i It is used to characterize the strength and verifiability of external evidence related to candidate mutation sites, rather than simply indicating the existence of literature sources.

[0080] In one embodiment, the second pathway base weight L for each candidate mutation site calculated by the second pathway iExpressed as follows: ; This application can integrate the semantic interpretability of the model output and the verifiability of the evidence into a site-level evaluation system, thereby forming a site-level supplementary weight for subsequent comprehensive ranking.

[0081] In this embodiment of the application, step S410 specifically includes: Step S411: Based on preset semantic matching rules, determine the current semantic support level in the intermediate semantic analysis results; Step S412: Find the correspondence between the preset semantic support level and the semantic support score to obtain the semantic support score corresponding to the current semantic support level; Step S413: Determine the support strength and verifiability of external evidence in the intermediate semantic analysis results based on the evidence level evaluation mechanism, and obtain the evidence level based on the support strength and verifiability.

[0082] For example, the semantic support score does not directly adopt the original confidence level of the large language model, but is a task-specific score obtained by post-processing and mapping the intermediate semantic analysis results based on pre-constructed semantic matching rules. In one embodiment, the semantic support relationship between candidate mutation sites and subject phenotypes can be divided into strong support, moderate support, weak support, and no support or uncertainty, and mapped to 3 points, 2 points, 1 point, and 0 points, respectively. Among them, strong support means that the gene containing the candidate mutation site or the evidence related to the candidate mutation site can explain one or more core phenotypes, and the disease mechanism or genetic pattern is consistent with the case information, while there is no significant semantic contradiction; moderate support means that it can explain some core phenotypes or multiple auxiliary phenotypes, but the support range is incomplete or there are certain uncertainties; weak support means that there is only an indirect association with a few auxiliary phenotypes; no support or uncertainty means that it cannot explain the main phenotype, there is significant semantic contradiction, or the model cannot form a stable judgment.

[0083] In one embodiment, if only verifiable database entries, literature mentions, or calculated predictions exist as preliminary supporting information, then E i = 1; if there are localizable case reports, supporting functional tests, or clear, high-quality external evidence, then E i =2; if the source is unverifiable or the evidence is obviously contradictory, then E i = 0. By combining semantic support scores with evidence levels, the output of the second pathway can reflect not only whether the candidate mutation site is "supported" but also whether its "supporting evidence is traceable," thus providing a more reliable reference for subsequent comprehensive ranking and manual review.

[0084] like Figure 1As shown, the gene mutation point screening method based on clinical phenotype further includes the following steps: Step S500: Based on the basic weights of the first and second pathways, obtain the comprehensive weight of each candidate mutation site, and screen the candidate mutation sites based on the comprehensive weights to obtain the screening results.

[0085] After screening through the first and second pathways, this application performs a comprehensive scoring and ranking at the candidate mutation site level.

[0086] In this embodiment of the application, the comprehensive weight of each candidate mutation site is obtained based on the basic weight of the first pathway and the basic weight of the second pathway, including: The first pathway base weight of each candidate gene is assigned to all candidate mutation sites located on the candidate gene, so that the candidate mutation sites have the same first pathway base weight as their respective genes. The basic weights of the first and second pathways corresponding to each candidate mutation site are added together to obtain the comprehensive weight of each candidate mutation site.

[0087] Specifically, the first pathway first outputs a set of candidate genes and their corresponding gene-level basic weights based on structured knowledge matching results; then, combined with the subject's sequencing data, it extracts corresponding candidate mutation sites from the candidate gene range, and transfers the basic weights of the candidate genes to the corresponding candidate mutation sites as the first pathway basic weights W of the candidate mutation sites. i The second pathway then outputs site-level supplementary weights (i.e., the base weights of the second pathway) L for the candidate mutation sites. i Based on this, the comprehensive weight of each candidate mutation site was calculated. The expression is: ; Based on the comprehensive weight T i Candidate mutation sites are ranked, with those having higher overall weights ranked higher and lower. When outputting the ranking results, the system can simultaneously output disease matching items, phenotypic matching items, semantic support level of the large language model, inference summary, and evidence source information corresponding to each candidate mutation site, for use in generating subsequent gene interpretation reports. If the overall weights of all candidate genes or candidate mutation sites are low, and there is a lack of significant distinction between different candidate results, this situation is recorded, and a prompt indicating that further manual review or clinical validation is required is output.

[0088] This application integrates gene-level prior information provided by the first pathway with site-level semantic reasoning results provided by the second pathway at the candidate mutation site level through a comprehensive scoring and ranking method. This allows the final ranking results to combine the controllability of structured knowledge matching with the advantages of large language models in knowledge updating and semantic analysis, thereby improving the accuracy, recall, and robustness of candidate mutation site priority determination.

[0089] In one embodiment of this application, the method further includes: based on a mutation site screening model, semantically integrating and organizing the structured results, matching criteria, evidence summaries, and ranking results generated during the screening process into natural language, generating a gene interpretation report in a predetermined format, wherein the gene interpretation report is used to present the basis, evidence sources, and reasoning logic for forming the ranking results.

[0090] Specifically, this application transforms the screening and ranking results into directly usable outputs, which is an important component of this invention for clinical application and a key step in achieving interpretable and reportable output of results. During report generation, the large language model semantically integrates and organizes the structured results, matching criteria, evidence summaries, and ranking conclusions generated during the screening process into natural language. For example, it summarizes the correspondence between the subject's clinical phenotype and candidate genes or candidate mutation sites, summarizes supporting literature evidence and knowledge base evidence, explains the basis for the final ranking results, and generates report text that better suits clinical reading habits. In this stage, the large language model primarily plays a role in enhancing interpretation and organizing the report. Its generated content is constrained by the structured results, evidence levels, and traceability information output by the first and second pathways, thereby avoiding unconstrained generation that deviates from the screening criteria. By generating gene interpretation reports, this invention can not only output the ranking results of candidate genes or candidate mutation sites, but also simultaneously present the main basis, sources of evidence and reasoning logic for the results. Furthermore, by using a large language model, it can improve the coherence, readability and interpretability of the report content, thereby enhancing the interpretability, verifiability and clinical applicability of the results, and providing doctors, genetic counselors or researchers with more valuable auxiliary interpretation results.

[0091] Compared to traditional gene testing technologies such as PCR, gene chips, and high-throughput sequencing, this invention eliminates the need for additional large-scale experimental testing. Instead, it builds upon existing gene testing results by combining the subject's clinical phenotype, structured knowledge base, and the reasoning capabilities of a large language model to further screen and interpret candidate mutations. Therefore, it offers advantages in subsequent interpretation, including faster processing speed and lower additional costs, making it more suitable for rapid analysis and assisted interpretation of large-scale testing results. Compared to existing gene screening schemes that primarily rely on static knowledge bases, fixed rule matching, or single annotation scoring, this invention standardizes clinical phenotypes and utilizes a large language model to mine semantic and clinical relationships between different phenotypes, resulting in more complete, richer, and more realistic clinical contexts in the input information. Building upon this foundation, this invention constructs a clinical phenotype-driven dual-pathway screening mechanism. While preserving the stability, controllability, and verifiability of structured knowledge screening, it introduces a large language model for semantic reasoning and knowledge compensation. This effectively improves the accuracy, recall, and robustness of candidate gene and candidate mutation site screening, and alleviates the problems of lagging updates and insufficient coverage in existing static knowledge bases, enhancing the system's adaptability to new knowledge, rare genes, and complex clinical phenotype scenarios. Furthermore, this invention advances the screening granularity from the gene level to the mutation site level, and comprehensively scores and ranks the outputs of the first and second pathways at the site level, thereby more effectively identifying candidate mutation sites that are more relevant to the subject's clinical phenotype and disease state.

[0092] At the same time, this invention can not only output the ranking results of candidate genes or candidate mutation sites, but also simultaneously provide disease matching basis, phenotypic matching basis, large language model reasoning summary and supporting evidence, and further generate gene interpretation report, thereby significantly improving the interpretability, verifiability and clinical applicability of the results.

[0093] Therefore, this invention can reduce the reliance on manual screening and manual literature retrieval in the interpretation of gene detection results, improve interpretation efficiency, and enhance the product's promotional value and competitive advantage in precision medicine, genetic disease auxiliary diagnosis, bioinformatics analysis platforms, and related application scenarios.

[0094] Furthermore, such as Figure 4 As shown, based on the above-described clinical phenotype-based gene mutation screening method, the present invention also provides a clinical phenotype-based gene mutation screening device, comprising: The first weight determination module 100 is used to obtain the gene detection results and the clinical text to be analyzed, determine candidate genes based on the gene detection results, the clinical text to be analyzed and the pre-constructed multi-source database, and calculate the first pathway basic weight of each candidate gene. The extraction module 200 is used to acquire gene sequencing data corresponding to the clinical text to be analyzed, and based on the gene sequencing data, extract candidate mutation sites in the genomic region corresponding to the candidate gene, and acquire external evidence corresponding to the candidate mutation sites. The input module 300 is used to extract the phenotype of the clinical text to be analyzed to obtain a structured phenotype. The structured phenotype, the context information of the candidate mutation sites, and external evidence are input into the trained mutation site screening model to obtain intermediate semantic analysis results. The second weight determination module 400 is used to determine the basic weights of the second path based on the intermediate semantic analysis results. The sorting module 500 is used to obtain the comprehensive weight of each candidate mutation site based on the basic weight of the first pathway and the basic weight of the second pathway, and to screen the candidate mutation sites based on the comprehensive weight to obtain the screening results.

[0095] It should be noted that the foregoing explanation of the embodiment of the gene mutation point screening method based on clinical phenotype also applies to the gene mutation point screening device based on clinical phenotype in this embodiment, and will not be repeated here.

[0096] This invention discloses a gene mutation screening device based on clinical phenotype. It acquires gene detection results and clinical text to be analyzed, identifies candidate genes based on these factors, and calculates the basic weight of a first pathway for each candidate gene. It then acquires gene sequencing data corresponding to the clinical text, extracts candidate mutation sites within the genomic regions corresponding to the candidate genes, and obtains external evidence for these sites. Phenotypic extraction is performed on the clinical text to obtain a structured phenotype. This structured phenotype, contextual information of the candidate mutation sites, and external evidence are input into a trained mutation site screening model to obtain intermediate semantic analysis results. The basic weight of a second pathway is determined based on these intermediate semantic analysis results. A comprehensive weight for each candidate mutation site is obtained based on the first and second pathway weights. The candidate mutation sites are then screened based on this comprehensive weight to obtain the screening results. This application achieves site-level screening by screening candidate mutation sites, supporting refined determination of pathogenic mutations and thus improving the accuracy of gene mutation screening.

[0097] Figure 5 A schematic diagram of the structure of a terminal provided in an embodiment of this application. The terminal may include: The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0098] When the processor 502 executes the program, it implements the gene mutation point screening method based on clinical phenotype provided in the above embodiments.

[0099] Furthermore, the terminal also includes: Communication interface 503 is used for communication between memory 501 and processor 502.

[0100] The memory 501 is used to store computer programs that can run on the processor 502.

[0101] The memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0102] If the memory 501, processor 502, and communication interface 503 are implemented independently, they can be interconnected via a bus to communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, only one line is used in the diagram, but this does not imply that there is only one bus or one type of bus.

[0103] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0104] Processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0105] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for screening gene mutation points based on clinical phenotypes.

[0106] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0107] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0108] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0109] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can read and execute instructions from or in conjunction with such an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). In addition, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically by optically scanning paper or other media, then editing, interpreting or otherwise processing them as necessary, and then storing them in computer memory.

[0110] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0111] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0112] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0113] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for screening gene mutation points based on clinical phenotype, characterized in that, include: The system acquires the gene detection results and clinical text to be analyzed, identifies candidate genes based on the gene detection results, clinical text to be analyzed, and a pre-constructed multi-source database, and calculates the basic weight of the first pathway for each candidate gene. Obtain gene sequencing data corresponding to the clinical text to be analyzed; based on the gene sequencing data, extract candidate mutation sites within the genomic region corresponding to the candidate gene, and obtain external evidence corresponding to the candidate mutation sites. Phenotypic extraction is performed on the clinical text to be analyzed to obtain a structured phenotype. The structured phenotype, the contextual information of the candidate mutation sites, and external evidence are input into a trained mutation site screening model to obtain intermediate semantic analysis results. The basic weights of the second pathway are determined based on the intermediate semantic analysis results. The comprehensive weight of each candidate mutation site is obtained based on the basic weight of the first pathway and the basic weight of the second pathway. The candidate mutation sites are then screened based on the comprehensive weight to obtain the screening results.

2. The gene mutation point screening method based on clinical phenotype according to claim 1, characterized in that, Obtain the gene testing results and clinical text to be analyzed. Based on the gene testing results, the clinical text to be analyzed, and a pre-constructed multi-source database, identify candidate genes and calculate the basic weight of the first pathway for each candidate gene, including: The system acquires the gene detection results and clinical text to be analyzed, and loads a pre-constructed multi-source database to obtain the gene-disease correspondence and gene-phenotype correspondence. Extract the disease information and phenotype information to be analyzed from the clinical text to be analyzed; Each mutated gene in the gene detection results is traversed, and the associated disease information and associated phenotype information corresponding to each mutated gene are obtained from the gene-disease correspondence and gene-phenotype correspondence. If the phenotypic information to be analyzed matches the associated phenotypic information, the mutated gene will be used as a candidate gene. The phenotypic information associated with each candidate gene is compared with the phenotypic information to be analyzed by calculating the string similarity to obtain a phenotypic similarity score. The disease information associated with each candidate gene is compared with the disease information to be analyzed by calculating the string similarity to obtain a disease similarity score. The phenotypic similarity score and the disease similarity score are weighted and fused to obtain the first pathway basic weight for each candidate gene.

3. The gene mutation point screening method based on clinical phenotype according to claim 2, characterized in that, The phenotypic similarity score and the disease similarity score are weighted and fused to obtain the first pathway base weight for each candidate gene, including: Based on the relationship between disease similarity scores and preset scoring thresholds, the first weight corresponding to the disease similarity score is determined; If the associated phenotypic information and associated disease information are found to be related based on the preset phenotypic-disease correspondence, then a second weight is assigned to the phenotypic similarity score. If the associated phenotypic information and associated disease information obtained based on the preset phenotypic-disease correspondence are not related, then a third weight is assigned to the phenotypic similarity score. The disease similarity score is weighted based on the first weight to obtain a first weighted value, and the phenotypic similarity score is weighted based on the second weight or the third weight to obtain a second weighted value. The sum of the first weighted value and the second weighted value is used as the first pathway base weight for each candidate gene; The second weight is greater than the third weight.

4. The gene mutation point screening method based on clinical phenotype according to claim 1, characterized in that, The basic weights of the second pathway are determined based on the intermediate semantic analysis results, including: Based on preset semantic matching rules and evidence level evaluation mechanism, the intermediate semantic analysis results are mapped into semantic support scores and evidence levels. The semantic support score and the evidence level are multiplied together to obtain the basic weight of the second pathway.

5. The gene mutation point screening method based on clinical phenotype according to claim 4, characterized in that, Based on preset semantic matching rules and evidence level evaluation mechanisms, the intermediate semantic analysis results are mapped to semantic support scores and evidence levels, including: Based on preset semantic matching rules, the current semantic support level in the intermediate semantic analysis results is determined; Find the correspondence between the preset semantic support level and the semantic support score to obtain the semantic support score corresponding to the current semantic support level; The evidence level is determined based on the evidence level evaluation mechanism to determine the support strength and verifiability of external evidence in the intermediate semantic analysis results, and the evidence level is obtained based on the support strength and verifiability.

6. The gene mutation point screening method based on clinical phenotype according to claim 1, characterized in that, The training steps for the mutation site screening model include: Standardized gene mutation points and corresponding standardized phenotypic terms are collected. Healthy individuals' whole genome sequences are selected from publicly available population genome data as genetic background templates. Target gene mutation points are then inserted into the corresponding genomic coordinates to construct virtual patient whole genome samples. Based on the virtual patient whole genome sample, the pre-trained large language model is fine-tuned to obtain a trained mutation site screening model. The training tasks of the mutation site screening model include: gene-level association learning tasks and site-level training tasks.

7. The gene mutation point screening method based on clinical phenotype according to claim 1, characterized in that, The contextual information of the candidate mutation site includes at least one of the following: the gene where the candidate mutation site is located, genomic coordinates, mutation type, amino acid change, transcript information, protein functional domain location, information related to the predetermined genetic pattern, population frequency information, functional impact prediction results, and database annotation information. The external evidence includes at least one of the following: literature evidence fragments, case summaries, database entries, descriptions of functional experiments, descriptions of disease mechanisms, and family information; The intermediate semantic analysis results include at least one of the following: candidate mutation site identification, core phenotype support, auxiliary phenotype support, consistency of disease mechanism or genetic pattern, semantic contradiction information, summary of supporting reasons, explanation of evidence sources, and preliminary support level.

8. The gene mutation point screening method based on clinical phenotype according to claim 1, characterized in that, The comprehensive weight of each candidate mutation site is obtained based on the basic weights of the first and second pathways, including: The first pathway base weight of each candidate gene is assigned to all candidate mutation sites located on the candidate gene, so that the candidate mutation sites have the same first pathway base weight as their respective genes. The basic weights of the first and second pathways corresponding to each candidate mutation site are added together to obtain the comprehensive weight of each candidate mutation site.

9. The gene mutation point screening method based on clinical phenotype according to claim 1, characterized in that, The method further includes: Based on the mutation site screening model, the structured results, matching criteria, evidence summaries and ranking results generated during the screening process are semantically integrated and organized into natural language to generate a gene interpretation report in a predetermined format. The gene interpretation report is used to present the basis, evidence sources and reasoning logic for the ranking results.

10. A gene mutation point screening device based on clinical phenotype, characterized in that, include: The first weight determination module is used to obtain the gene detection results and clinical text to be analyzed, determine candidate genes based on the gene detection results, clinical text to be analyzed and a pre-constructed multi-source database, and calculate the first pathway basic weight of each candidate gene. The extraction module is used to acquire gene sequencing data corresponding to the clinical text to be analyzed, and based on the gene sequencing data, extract candidate mutation sites in the genomic region corresponding to the candidate gene, and obtain external evidence corresponding to the candidate mutation sites. The input module is used to extract the phenotype of the clinical text to be analyzed to obtain a structured phenotype. The structured phenotype, the contextual information of the candidate mutation sites, and external evidence are input into a trained mutation site screening model to obtain intermediate semantic analysis results. The second weight determination module is used to determine the basic weights of the second path based on the intermediate semantic analysis results. The sorting module is used to obtain a comprehensive weight for each candidate mutation site based on the basic weights of the first and second pathways, and to filter the candidate mutation sites based on the comprehensive weights to obtain the screening results.

11. A terminal, characterized in that, include: The system includes a memory, a processor, and a clinical phenotype-based gene mutation screening program stored in the memory and executable on the processor, wherein the clinical phenotype-based gene mutation screening program, when executed by the processor, implements the steps of the clinical phenotype-based gene mutation screening method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that can be executed to implement the steps of the clinical phenotype-based gene mutation screening method as described in any one of claims 1 to 9.

Images