Rare disease information input and gene mutation analysis method and system based on phenotype matching and storage medium

Abstract

The invention discloses a method and a system for assisting in inputting clinical information of rare diseases and analyzing gene mutation based on phenotypes. The method comprises the following steps: firstly, acquiring clinical information in voice, text and image forms of a patient through a multi-source data acquisition module, converting the clinical information into characters, and performing entity recognition and standardization processing to generate structured medical record data; secondly, extracting clinical phenotypes from the structured data; furthermore, a candidate gene list is obtained according to the gene-disease relationship, comprehensive scoring and sorting are carried out, and a concerned gene list is output. According to the method, efficient structured input and standardization of clinical information are realized, the accuracy and automation level of phenotype-gene matching are remarkably improved, the gene variation interpretation period is effectively shortened, and intelligent support is provided for precise diagnosis of genetic diseases.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, specifically to a method, system, and storage medium for rare disease information entry and gene mutation analysis based on phenotypic matching. Background Technology

[0002] Precise diagnosis of genetic diseases is one of the core challenges in modern medicine, particularly in precision medicine and rare disease treatment. Genetic testing technologies, such as whole-exome sequencing (WES) and whole-genome sequencing (WGS), can detect tens of thousands of gene variants in a single individual. However, accurately identifying pathogenic or suspected pathogenic variants truly relevant to a patient's clinical phenotype from this massive amount of variant data is the most time-consuming and complex step in interpreting clinical reports. This process relies not only on bioinformatics analysis of the variants themselves (such as frequency, functional prediction, conservation, etc.), but more importantly, on a deep correlation and comprehensive assessment of gene variant data with the patient's clinical presentation.

[0003] In interpreting genetic variations, the accurate extraction and structuring of clinical information is crucial. It provides indispensable phenotypic evidence for determining the pathogenicity of gene variations. However, current methods of acquiring and processing clinical information face significant bottlenecks. Patient clinical information often exists in unstructured forms, such as doctor-written medical records, patient-reported audio recordings, and imaging reports. This information is highly complex, diverse, and presented arbitrarily. Traditional rule-based or keyword-matching methods struggle to effectively handle the synonymy (e.g., "myocardial infarction" vs. "heart attack"), nesting (e.g., "acute anterior wall myocardial infarction" contains multiple entities), and contextual dependencies of medical terms, resulting in insufficient completeness, accuracy, and standardization of information extraction. This unstructured or semi-structured data state severely hinders the efficient integration and intelligent analysis of clinical information and genetic data.

[0004] Furthermore, to achieve precise matching between clinical phenotypes and gene variations, standardized phenotypic terminology systems such as Human Phenotype Ontology (HPO) have been developed internationally, and China has also established the Chinese Human Phenotype Standard Terminology Consortium (CHPO). These ontology libraries provide a foundation for the computer-readable representation of phenotypic information and cross-system interoperability. However, in practical applications, automatically and accurately mapping non-standardized clinical symptoms described by doctors or patients (such as "children are growing slowly" or "frequent nosebleeds") to standardized ontology terms like CHPO remains a significant challenge. This process typically relies on the expertise of interpreters and manual operation, which is not only inefficient but also prone to judgment bias due to differences in the experience of different personnel, affecting the consistency and positive rate of gene variation interpretation.

[0005] While some existing tools attempt phenotype-gene matching (such as Exomiser and Phenomizer), most rely on users manually inputting standardized HPO phenotypes, failing to fundamentally solve the problem of automated conversion from raw clinical information to standardized phenotypes. Furthermore, these tools often fail to fully utilize locally accumulated, manually reviewed and validated case data. Local databases contain a large number of "clinical phenotype-gene variation-interpretation conclusion" triplets for confirmed cases, data with extremely high reference value. If the clinical phenotype of new cases could be intelligently compared with historical cases in the local database, and highly similar phenotype patterns were found, candidate genes could be quickly located and existing interpretation conclusions reused, greatly improving analysis efficiency and reporting quality.

[0006] Therefore, existing technologies have significant shortcomings in areas such as automated and structured processing of multi-source heterogeneous clinical information, accurate mapping of non-standard clinical phenotypes to standard ontology terms (CHPO), and integration of local knowledge bases for intelligent gene screening and interpretation reuse. There is an urgent need for an automated and intelligent system that can span the entire process of "raw clinical information collection - structured input - standardized phenotype extraction - intelligent matching and sorting of gene variations" to address the pain points of low efficiency, poor consistency, and reliance on human experience in the current interpretation of genetic disease gene testing reports, thereby improving the efficiency and accuracy of rare disease diagnosis. Summary of the Invention

[0007] The purpose of this invention is to provide a method for rare disease information entry and gene mutation analysis based on phenotype matching, so as to solve the problems mentioned in the background art that the existing technology has obvious shortcomings in the automated and structured processing of multi-source heterogeneous clinical information, the accurate mapping of non-standard clinical phenotypes to standard ontology terms (CHPO), and the integration of local knowledge bases to achieve intelligent gene screening and interpretation reuse.

[0008] To achieve the above objectives, the present invention provides the following technical solution: Methods for rare disease information entry and gene mutation analysis based on phenotypic matching include: Step S1: Multi-source clinical information collection, acquiring the patient's clinical information, including data in the form of voice, text or images, and converting it into a structured electronic medical record; Step S2: Medical entity recognition, extracting clinical phenotype keywords from the structured electronic medical record, and performing synonym and hyponym enrichment to construct an expanded clinical phenotype set; Step S3: Phenotypic semantic enrichment and dual-track matching, matching the expanded clinical phenotype set with the CHPO standard phenotype database and / or the local database to obtain a standardized phenotype set; Step S4: Gene list generation, merging candidate genes corresponding to the two disease sets based on the OMIM gene-disease relationship; Step S5: Rank candidates by comprehensive scoring, taking into account at least phenotypic similarity, consistency of inheritance pattern, ACMG variant rating, and local interpretation records. Step S6: Output the sorted gene list. The gene list is output based on the scoring results to assist in the clinical interpretation of gene variations in hereditary diseases.

[0009] Furthermore, the method for generating structured electronic medical records in step S1 includes: Convert unstructured clinical information into text using at least one of the following methods: speech recognition and image text recognition; Use a BERT-CRF-based named entity recognition model to extract medical entities from text; The medical entities are categorized and archived according to preset medical record themes to generate structured medical records in JSON or XML format.

[0010] Furthermore, the BERT-CRF model is pre-trained using medical corpora and incorporates a dynamic transition matrix and hierarchical labeling strategy to identify nested medical entities, specifically including: BioBERT-zh, which was further pre-trained on the Chinese medical corpus, was used as the pre-training layer; A BERT encoder enhanced with a medical dictionary; CRF layer using dynamic transition matrix; Identify nested medical entities using a hierarchical labeling strategy; The extracted entities are populated into a preset medical record template to generate structured output in JSON or XML format.

[0011] Furthermore, in step S2, the synonym and hyponym enrichment step utilizes the medical terminology set to perform semantic enrichment on the clinical phenotype set M, specifically including: The CHPO standard phenotype set S was enriched with synonyms and hyponyms using the OMAHA terminology set and the UMLS medical dictionary to obtain the expanded CHPO phenotype set H. Synonym enrichment was performed on the clinical phenotype set M using the OMAHA and UMLS medical dictionaries; Hyposense enrichment was performed on the clinical phenotype set M using the OMAHA and UMLS medical dictionaries; The union of the above enrichment results yields the expanded clinical phenotype set M' and the CHPO phenotype set H, respectively.

[0012] Furthermore, the matching process in step S3 uses the Jaccard similarity coefficient to calculate the overlap between phenotypic sets and performs term alignment based on semantic relationships, specifically including: Perform intersection and union operations on the CHPO phenotype set corresponding to the candidate genes and the CHPO standard phenotype set obtained in step S3; The Jaccard similarity coefficient is calculated as intersection size / union size; and in the calculation logic of the intersection and union, synonyms and hyponyms are considered to be identical.

[0013] Furthermore, in step S5, when scoring comprehensively, priority is given to genes of existing matching cases in the local database, and their scores are directly inherited or superimposed on the scores of cases in the local database.

[0014] Furthermore, in step S3, if the clinical phenotype matches the local database, the existing gene interpretation content in the database is directly called to generate a report; if the match is unsuccessful, gene screening and sorting are performed through the CHPO and OMIM databases.

[0015] A rare disease information entry and gene mutation analysis system based on phenotypic matching, including

[0016] The multi-source data acquisition module is used to acquire clinical information in the form of voice, images, and text. A medical entity recognition engine is used to identify and extract feature keywords from clinical information; The phenotype enrichment and matching module is used to perform synonym / hypothesis expansion and mapping matching between clinical phenotypes and CHPO phenotypes; The gene screening and sorting module is used to score and sort genes based on factors such as phenotypic matching degree, inheritance mode, and ACMG rating. The local database interface module is used to call approved phenotype-gene-variation records; The structured medical record generation and display module is used to generate and visualize structured medical records and gene lists. Furthermore, it also includes a report generation module, which automatically generates draft gene testing reports based on a list of genes of interest, for review and publication by interpreters.

[0017] A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement steps of a method for rare disease information entry and gene mutation analysis based on phenotypic matching.

[0018] Compared with the prior art, the beneficial effects of the present invention are: (1) Improve the efficiency and standardization of clinical information processing: By integrating speech recognition, OCR and improved BERT-CRF model, the system automatically processes multi-source heterogeneous clinical information (speech, text and images), realizes the efficient conversion of unstructured medical records into structured data, significantly reduces manual entry and processing time, and improves information integrity and standardization. (2) Enhance the accuracy and comprehensiveness of phenotype-gene matching: Use multi-source medical ontology such as UMLS and SNOMED CT to enrich synonyms and hyponyms of clinical phenotypes and CHPO terms, effectively solve the problems of clinical expression diversity and terminology heterogeneity, expand the matching range, reduce the risk of missed detection due to expression differences, and improve the comprehensiveness and accuracy of matching. (3) Significantly shorten the gene variation interpretation cycle: Through automated phenotypic extraction and CHPO mapping, combined with Jaccard similarity, candidate genes are intelligently sorted and the interpretation cases that have been manually reviewed in the local database are reused first, which greatly reduces the workload of manually reviewing literature and duplicate rating, and can shorten the analysis time of a single case by more than one hour. (4) Improve diagnostic consistency and positive rate: reduce subjective bias caused by the experience difference of the interpreters in understanding the clinical phenotype, ensure the consistency of the analysis process through standardized process and intelligent algorithm, especially provide more potential genes of concern in negative results, provide more comprehensive reference for clinical practice, and reduce the need for post-sales re-analysis; (5) Realize the intelligent reuse of local knowledge assets: Deeply integrate the case database accumulated by the institution and reviewed by human review. When the phenotype of a new case is highly matched with that of a historical case, the existing interpretation conclusions can be directly called to realize the accumulation, sharing and efficient reuse of knowledge, and continuously improve the overall diagnostic capabilities of the institution. Attached Figure Description

[0019] Figure 1 This is a flowchart of the rare disease information entry and gene mutation analysis method based on phenotypic matching of the present invention; Figure 2 This is a flowchart illustrating the linking of clinical information to the CHPO standard phenotypic analysis of gene variations in this invention; Figure 3 This invention provides a flowchart for converting clinical information into CHPO phenotypes and automatically matching disease phenotypes with gene variants. Figure 4 This is a flowchart of the CHPO phenotypic terminology enrichment and standardization process of this invention; Figure 5 This is a schematic diagram of the Jaccard similarity coefficient calculation method of the present invention; Figure 6 This is a screenshot of the final structured medical record interface obtained by the rare disease information entry and gene mutation analysis system based on phenotypic matching of the present invention. Figure 7 This is a screenshot of the main interface of the rare disease information entry and gene mutation analysis system based on phenotypic matching of the present invention. Figure 8 This is a screenshot of the final result interface after the mutation sorting of the rare disease information entry and gene mutation analysis system based on phenotypic matching of the present invention is completed. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figures 1 to 4 This invention provides a technical solution: a method for rare disease information entry and gene mutation analysis based on phenotypic matching, including a system and a computer-readable storage medium. The method for gene mutation analysis based on phenotypic matching may include: data acquisition, acquiring data on the phenotype and gene mutations of the subject; phenotypic description, describing the subject's phenotype as standardized terms; assigning scores to the subject's gene mutations based on mutation pathogenicity, inheritance pattern, and phenotypic comparison; assigning scores based on a local database, searching the local database for cases with similar phenotypes to the subject and mutations in the same genes, adding the gene scores of these cases to the subject's gene scores to obtain the final score for each gene of the subject; and sorting, ranking the subject's gene mutations based on the scores. Specifically, it includes: Step 1: Obtain data on gene mutations in the subject.

[0022] Gene mutation data were obtained from the subjects' samples through standard gene sequencing analysis, as shown in the table below.

[0023]

[0024] Step 2: Obtain the patient's clinical phenotype.

[0025] Upon receiving medical record audio information, the system converts it into text information. The audio information is medical record data recorded in voice format. The audio information can be directly input via microphone, or pre-recorded and saved. The saved audio information can then be sent over a network or received via an external storage device. Converting the audio information into text can be done using the terminal's built-in speech-to-text software, or by sending the audio information to a server where speech-to-text software can recognize and convert it. The system then receives the returned text information from the server.

[0026] After obtaining the provided medical record images and application form photos, upload the medical record images and application form, and parse the images to convert them into text information.

[0027] Step 3: Generate the patient's structured medical record archive.

[0028] The clinical data in text form obtained in step 2 is used to identify key features and generate structured medical records according to preset themes. The multi-source data acquisition module integrates voice, text, and device data, and uses an improved BERT-CRF model to identify key features of clinical information and archive them according to preset medical record themes.

[0029] The structured medical record generation method of the present invention involves the configuration of a terminal, which may include: a memory, an operating system, a structured medical record generation program, a processor, and a display unit. The terminal is equipped with a rare disease information entry and gene mutation analysis system based on phenotypic matching. Figures 6 to 8 The main components of the system are: The multi-source data acquisition module is used to acquire clinical information in the form of voice, images, and text. A medical entity recognition engine is used to identify and extract feature keywords from clinical information; The phenotype enrichment and matching module is used to perform synonym / hypothesis expansion and mapping matching between clinical phenotypes and CHPO phenotypes; The gene screening and sorting module is used to score and sort genes based on factors such as phenotypic matching degree, inheritance mode, and ACMG rating. The local database interface module is used to call approved phenotype-gene-variation records; The structured medical record generation and display module is used to generate and visualize structured medical records and gene lists. It also includes a report generation module, which automatically generates draft gene testing reports based on a list of genes of interest, for review and publication by interpreters.

[0030] Step 4: Obtain the patient's CHPO standard phenotype and match gene variants.

[0031] Clinical phenotype keywords are extracted from the structured electronic medical records, and synonym and hyponym enrichment is performed to construct an expanded set of clinical phenotypes.

[0032] In the diagnosis of single-gene genetic diseases, the accuracy of clinical phenotype is crucial for the identification of pathogenic genes. Specific phenotypes help to accurately locate pathogenic genes. When there is a clear phenotype, the candidate genes are clear (e.g., café-au-lait spots generally correspond to NF1, and ectodermal dysplasia generally corresponds to EDA). When there is a general phenotype, there are many candidate genes. When the phenotype is complex and unclear, the gene is unknown.

[0033] For example, a patient presenting with short stature and developmental delay might be diagnosed as short stature by an intern, but a more experienced physician might diagnose Noonan syndrome. However, the number of genes associated with different diseases (phenotypes) varies greatly. There are approximately 788 genes associated with developmental delay, about 365 with short stature, about 57 with Noonan syndrome, and about 50 with brachydactyly, and so on. Therefore, accurate diagnosis during genetic testing can significantly narrow down the range of variations and reduce the workload of data analysis. Clinical practice is characterized by complexity and ambiguity. The long history of medical development has led to cross-classification of disease nomenclature, and many diseases exhibit phenotypic variability. Mutations in the same pathogenic gene may manifest differently in different patients. Furthermore, each physician's description and judgment of a patient's condition may differ, and the early, middle, and late stages of disease progression also vary. Therefore, connecting and analyzing the relationship between clinical disease phenotypic information and genetic data has become a pressing issue. CHPO (The Chinese Human Phenotype Ontology Consortium) plays an important role in linking disease phenotypes and pathogenic genes, but there is still a gap in the link with patients' clinical phenotypes. Interpreters need to be able to convert clinical phenotypes to CHPO phenotypes. This process requires a high level of experience and mastery of clinical phenotypes from the interpreters. Different people's understanding may be biased, which may affect the positive rate of the report.

[0034] This application starts with the patient's clinical phenotype and gene mutation information. It automatically classifies patient-provided case data and various clinical information, extracts keywords from the clinical information and converts them into standardized phenotypes. By analyzing the correlation between these phenotypes and the patient's gene mutation-related phenotypes, and combining this with existing cases in a local database, if the local database contains the same or similar phenotypes, and the patient detects a mutation already present in the local database, the pathogenic gene can be precisely located based on the local database. The interpretation content from the local database can then be directly extracted to generate a report. This not only simplifies the data analysis process and reduces the time spent searching for pathogenic mutations, but also eliminates the time required for literature searches and rating for mutation interpretation. When no relevant information is found in the local data, the patient's gene mutation-related phenotypes are mapped and matched with clinical phenotypes to find phenotype-related genes. Candidate genes are then ranked according to phenotypic correlation, ACMG rating, and inheritance pattern to determine the pathogenic gene.

[0035] Figure 2 This is a flowchart in this application that links clinical information to the CHPO standard phenotypic analysis of gene variants.

[0036] like Figure 2 As shown, the process of linking clinical information to CHPO standard phenotypic analysis of gene variants may include: 1) Information acquisition: acquiring the subject's clinical information and gene variant information; 2) Phenotypic linking: automatically classifying the subject's clinical information and linking it to CHPO standardized phenotypic entries; 3) Phenotypic and gene scoring: matching and scoring the overlap between the clinical phenotype and the corresponding CHPO standardized phenotype of the patient's genes and the corresponding entries in the local database; 4) sorting the subject's gene variants based on the subject's phenotypic and gene score, ultimately obtaining the genes or gene list that require clinical attention.

[0037] The following detailed description uses specific embodiments. In this embodiment, the synonyms and hyponyms involved in the expansion of CHPO standard phenotypes, clinical phenotypes, and local database phenotype entries can be added after being reviewed and approved by doctors, resulting in enriched phenotype databases. The phenotype and gene-related databases involved come from historically accumulated local databases and publicly available phenotype-gene databases, such as the CHPO database and the OMIM database.

[0038] In this embodiment, taking "intellectual disability" as an example, the expansion of synonyms and hyponyms can be achieved through the following steps: 4.1 Steps for expanding synonyms: 1) Multi-source terminology collection: A. Authoritative terminology set retrieval: Obtain CUI (C0021361) corresponding to "intellectual disability" from UMLS, and extract synonyms such as "mental developmental delay" (ICD-10 term), "cognitive impairment" (SNOMED CT description), and "intellectual disability" (common term in clinical genetics). Referencing the "neurodevelopmental disorders" category of the OMAHA medical terminology set, supplement Chinese synonyms such as "intellectual developmental delay" (commonly used in pediatrics) and "intelligent impairment" (standard term in psychiatry). B. Clinical scenario supplementation: Analyze electronic medical records (EMR) and patient interview records, and extract colloquial synonyms such as "slow-witted," "slow to react" (to be labeled as "patient self-reported terms"), and "learning difficulties" (to be limited to "intelligence-related learning difficulties"). 2) Screening and Standardization: A. Deduplication and Semantic Verification: Eliminate ambiguous terms (e.g., "cognitive impairment" should exclude types caused by non-intellectual factors) and retain core synonyms. B. Terminology Standardization: Refer to the "Chinese Classification and Diagnostic Criteria for Mental Disorders (CCMD-3)" and uniformly describe it as "intellectual disability (also known as mental retardation)", and indicate the source of the term (e.g., UMLS, OMAHA); 3) Expert Review and Integration: A. Submit the list of synonyms to pediatric neurology experts for review to confirm the applicability of the terms in clinical diagnosis (e.g., "learning difficulties" needs to be distinguished from "intellectual disability" and "specific learning disabilities"). B. Link the terms to the core term "intellectual disability" to form a thesaurus (e.g., intellectual disability = mental retardation = intellectual disability = intellectual developmental delay).

[0039] 4.2 Hypothesis expansion steps: 1) Hierarchical Relationship Extraction: A. Based on Ontology Terminology Sets: UMLS: Obtain hypothetical terms through the "has_subtype" relation, such as "mild intellectual disability" (IQ 50-70), "moderate intellectual disability" (IQ 35-49), "severe intellectual disability" (IQ 20-34), and "profound intellectual disability" (IQ < 20). OMAHA Terminology Set: Expand etiological subtyping hypothetical terms through the "SubclassOf" relation, such as "hereditary intellectual disability" → "Down syndrome-related intellectual disability" and "fragile X syndrome intellectual disability"; "acquired intellectual disability" → "post-brain injury intellectual disability" and "post-encephalitis sequelae intellectual disability". B. Supplementing Clinical Subtypes: Based on the Diagnostic and Statistical Manual of Mental Disorders (DSM-5), add clinical subtypes such as "unspecified intellectual disability" (degree or cause cannot be determined) and "intellectual disability with language impairment".

[0040] 2) Definition and Attribute Supplementation: A. Add operational definitions for each hyponym, such as: Mild intellectual disability → "Able to perform simple life skills, but requires guidance to complete complex tasks"; Down syndrome-related intellectual disability → "Intellectual disability caused by chromosome karyotype 47,+XX(XY), accompanied by distinctive facial features and growth and developmental delays." B. Associate with clinical characteristic attributes, such as IQ range, etiology, and comorbidities (e.g., "comorbid epilepsy" or "comorbid autism spectrum disorder").

[0041] 3) Dynamic updates: New subtypes of UMLS / OMAHA (such as "intellectual disability related to hereditary metabolic diseases") are regularly synchronized and included in the system after expert review.

[0042] By enriching clinical information, local database phenotypes, and the CHPO phenotype database, the system achieves the greatest possible match between clinical information and patient variation information corresponding to phenotypes. This avoids missed diagnoses due to information errors caused by heterogeneity in clinical manifestations or differences in experience among doctors. Synonym expansion improves the accuracy of electronic medical record retrieval (e.g., searching for "mental retardation" can match cases of "intellectual disability"), while subdividing hyponyms aids in precise diagnosis (e.g., distinguishing between "mild intellectual disability" and "Down syndrome intellectual disability"). Through this process, the semantics of the clinical phenotype of "intellectual disability" are enriched, balancing terminology standardization with clinical practicality.

[0043] After enriching clinical information, local database phenotypes, and the CHPO phenotype database, phenotype sets are obtained respectively. The overlap between pairs of sets is calculated using the Jaccard similarity coefficient. For example, a patient's clinical phenotype is thrombocytopenia, anemia, hyperbilirubinemia, and thrombocytopenia, with no related phenotypes in the parents. Enrichment of this clinical phenotype using OMAHA and UMLS yields the following phenotype sets: {thrombocytopenia, thrombocytopenia, aplastic anemia, myelodysplastic syndrome (MDS), leukemia with bone marrow infiltration, vitamin B12 or folic acid deficiency-related thrombocytopenia, May-Hegglin syndrome, macrothrombocytopenia syndrome, hereditary thrombocytopenia with radius absence, Wiskett-Aldrich syndrome, Fanconi syndrome, idiopathic thrombocytopenic purpura (ITP)}. Systemic lupus erythematosus, thrombotic thrombocytopenic purpura (TTP), hemolytic uremic syndrome (HUS), disseminated intravascular coagulation (DIC), sepsis, Gaucher disease, anemia, congenital dysplasia of erythropoiesis, iron deficiency anemia, megaloblastic anemia, sideroblastic anemia, hemolytic anemia, hereditary spherocytosis, G6PD deficiency, thalassemia, immune hemolysis, microangiopathic hemolysis, jaundice, elevated unconjugated bilirubin, Gilbert syndrome, Crigler-Najjar syndrome, elevated conjugated bilirubin, Dubin-Johnson syndrome, Rotor syndrome, Glanzmann syndrome, neonatal thrombocytopenia, GPIIb / IIIa abnormality. Based on the relationship between genes and diseases in the OMOM database, the CHPO phenotypes corresponding to the patient's gene variations were obtained. Using the same enrichment method, the enriched local database phenotype set and CHPO phenotype set were obtained.

[0044] The enriched clinical phenotypes were matched with local database phenotypes. Referring to the patient cases mentioned above, the local database phenotype library recorded phenotypes such as thrombocytopenia, anemia, hyperbilirubinemia, and thrombocytopenia. The system automatically labeled the relevant genes and sorted them according to the interpretation of the local database records as follows:

[0045]

[0046]

[0047]

[0048]

[0049] Step 5: Phenotypic semantic enrichment and dual-track matching. The expanded clinical phenotype set is matched with the CHPO standard phenotype database and / or the local database to obtain a standardized phenotype set.

[0050] The Jaccard similarity coefficient is used to calculate the similarity between pairwise sets of enriched clinical phenotype terms and patient gene corresponding CHPO phenotype terms. Specifically, it measures the number of overlapping terms between the two sets; a higher overlap indicates a greater similarity value, signifying a stronger correlation between the gene abnormality and the clinical phenotype. When using this similarity coefficient, it's crucial to input the calculation logic, where synonyms and hypotheses between phenotypes are considered identical. In other words, the intersection phenotype of set A and set B contains synonyms and hypotheses from both sets. This logic simultaneously achieves the goal of automatically enriching clinical and CHPO phenotypes based on semantics. For example, in the above case, the enriched clinical phenotype of the patient is set A, and the CHPO phenotype corresponding to the patient gene ITGA2B is set B. The similarity is calculated using the Jaccard similarity coefficient as follows: Figure 5 As shown.

[0051] Using the Jaccard similarity coefficient formula: Similarity = Intersection / Union, set A and set B have 20 intersections and 60 unions, so the similarity is approximately 20 / 60 = 33.3%. Similarly, other gene variation data and phenotypic similarities of patients were obtained in the same way, and then sorted according to the obtained similarity values. Because the local database categorizes gene variants into four types according to the laboratory's reporting standards: related suspected variants (phenotypically highly related, pathogenic or suspected pathogenic variants consistent with inheritance patterns), variants requiring further evaluation (phenotypically related, genetically consistent variants of unclear significance, or phenotypically unrelated, genetically consistent pathogenic or suspected pathogenic variants), minor discovered variants (pathogenic or suspected pathogenic variants consistent with inheritance patterns in the minor discovered gene list), and other suspected variants (phenotypically unrelated, genetically inconsistent pathogenic or suspected pathogenic variants). Except for other suspected variants, the first three types of variants are interpreted as important genes. Therefore, when sorting by phenotype, the system will comprehensively integrate and sort according to the gene's previously reported records. For example, the SCN5A gene has been interpreted as a minor discovered variant. Therefore, although it is unrelated to the phenotype and has a similarity value of 0, it will be automatically ranked before the two phenotypically related genes FTCD and AMACR that are not interpreted when sorting according to the local database. This is consistent with our reporting standards. Since ITGA2B has been previously reported as a suspected gene interpretation related to platelet dysfunction, it can be accurately located and interpreted in this embodiment. Furthermore, it can directly call the evidence level and other interpretation content recorded in the database to directly output the report, thus directly achieving the phenotypic input-gene localization-report output mode. When matched with the local database, the data analysis stage will be shortened by about half an hour, while the time for the variant rating interpretation stage will vary depending on the number of variants to be rated. In this embodiment, the time is shortened to about forty minutes, and the overall process time is reduced by at least one hour.

[0052] Step 6: Gene list is generated by merging candidate genes corresponding to the two disease sets based on the OMIM gene-disease relationship.

[0053] When the enriched clinical phenotypes of patients do not overlap with the phenotypes in the local database, the enriched clinical phenotype set is matched with the CHPO set. Natural language processing tools (Juno2Pheno or BGI Chaoyang system) are used to automatically match the clinical phenotypes to CHPO entries. Using the Phenomatcher tool, the system automatically outputs a gene list sorted in descending order of phenotype matching count. The Junoselect algorithm, combined with parameters such as genetic pattern and phenotype weight, prioritizes highly associated genes. The output gene list is as follows:

[0054]

[0055]

[0056]

[0057] Step S7: Output the sorted gene list. The gene list is output based on the scoring results to assist in the clinical interpretation of gene variations in hereditary diseases.

[0058] By ranking genes based on the similarity between clinical phenotypes and their corresponding CHPO phenotypes, and comprehensively considering factors such as inheritance patterns and ACMG ratings, clinically relevant genes can be quickly and accurately screened. This reduces the number of thousands of genes detected in a patient's test to just a few or a dozen, significantly decreasing the amount of data and the time spent by interpreters searching the CHPO database based on phenotypes. It also greatly reduces the discrepancies in interpretation results caused by insufficient clinical experience of physicians or interpreters' misunderstandings of clinical phenotypes. Furthermore, the gene list selected through this method displays all clinically relevant genes, providing clinicians with more reference points when needed, facilitating their identification of key genes requiring attention, especially in cases of negative reports. This can provide more reference for clinicians, meet their needs, and reduce the possibility of subsequent reanalysis.

[0059] The following is an illustration using a specific case.

[0060] Step 1: Text information from speech conversion; The patient has had thrombocytopenia for 4 days, and the condition is progressively worsening. A blood test at the local hospital confirmed the thrombocytopenia. Appetite is good, bowel movements are normal, and there has been no recent weight loss. The patient has previously been treated for anemia. No one in the family has had this condition. The older brother and younger sister are healthy.

[0061] Step 2: Recognize the text information in the application form image; Patient: Zhang San; Father: Zhang San's father; Mother: Zhang San's mother; Brother: Zhang San's brother; Sister: Zhang San's sister. Hospital: Wuhan Hospital; Gender: Male; Age: 5 years old; Department: ; Sample Type: Peripheral Blood; Collection Date: June 11, 2025; Receipt Date: June 14, 2025; Diagnosis: Thrombocytopenia, Anemia, Hyperbilirubinemia, Platelet Atherosclerosis; Parents showed no related phenotypes. Step 3: Text information from medical record images; Medical history characteristics: The patient, Zhang San, male, 5 years old, had a rapid onset and a short course of illness.

[0062] Chief complaint: Thrombocytopenia for 4 days. The child's platelet count was 62 four days prior to admission. 10^9 / L; no jaundice of the skin or sclera; no new ecchymosis or spots, no gingival bleeding, no bloody stools, no hematuria; no bone or joint pain. The child is prone to nosebleeds, which improved after taking traditional Chinese medicine one month ago. Since the illness, the child's mental state has been good, appetite is good, urination and bowel movements are normal, and there has been no significant change in weight.

[0063] Medical history: No special treatment was given outside the hospital; a complete blood count was performed at our outpatient clinic. White blood cell count 4.51 10^9 / L↓, Platelets 56 10^9 / L↓, Red blood cell count 3.16 10^12 / L, Hemoglobin 1138 / L. 4) Physical examination: Temperature: 36.6℃, Respiration: 21 breaths / min, Heart rate: 102 beats / min, Oxygen saturation: 98%, Normal development, good nutrition, good spirits, responsive, clear consciousness, poor complexion, no signs of dehydration. Skin jaundice, no edema, good skin elasticity, no marbled skin pattern, no skin rash, no purulent lesions, no pigmentation, no desquamation. Red, moist, no cyanosis around the lips, smooth oral mucosa, congested, bilateral tonsils grade 1, no exudate, no strawberry tongue. Symmetrical breath sounds in both lungs, clear breath sounds in both lungs, no dry or wet rales in both lungs, strong heart sounds, regular rhythm, no heart murmurs heard at the base of the heart or in the precordial area. Soft abdomen, no palpable enlargement of the liver and spleen, no masses in the abdomen, no tenderness in the abdomen, no muscle guarding, no rebound tenderness, circulation is normal. Auxiliary examination: Platelet count 62 10^9 / L↓, red blood cells 2.96 Blood glucose level decreased to 10^12 / L, hemoglobin level was 108 g / L. Absolute neutrophil count was 1.31. 10^9 / L.

[0064] Past medical history: Anemia, jaundice, severe picky eating habits, avoids green vegetables, and consumes little meat. The child has experienced recurrent fever in the past month. The patient denies any history of infectious diseases such as hepatitis or tuberculosis, denies any contact with such diseases, denies any food or drug allergies, denies any surgical history, denies any history of trauma, denies any history of blood transfusions, and denies any history of planned immunizations.

[0065] Family history: Parents are healthy, deny consanguineous marriage, deny any family history of the same or similar diseases, and have no family history of hereditary diseases. Preliminary clinical diagnosis: Anemia, thrombocytopenia, hyperbilirubinemia, jaundice, thrombocytopenia.

[0066] Step 4: Identify the characteristic keywords in the medical record text information and divide the characteristic keywords according to the preset medical record subject items.

[0067] Step 5: The identification information obtained from the application form's feature keywords is consistent with the segmented text information. For example, identifying "Name": Zhang San, the segmented text information is: Zhang San; identifying "Gender": Male, the segmented text information is: Male; identifying "Diagnosis": Thrombocytopenia, Anemia, Hyperbilirubinemia, Platelet Atherosclerosis, the segmented text information is: Thrombocytopenia, Anemia, Hyperbilirubinemia, Platelet Atherosclerosis. If the identification information is blank, the segmented text information is "-": for example, if the application form identifies "Department" and there is no text information, the segmented text information is "-". Subsequently, it is presented in the structured medical record according to the feature keywords.

[0068] Step 6: Clinical information from the audio source is analyzed using pre-defined keywords set according to a preset theme. For example, the preset theme "Chief Complaint" is identified as the chief complaint using keywords such as "discovery" and "hint," or through disease descriptions from a medical dictionary. The final chief complaint obtained is: Thrombocytopenia for 4 days with progressive worsening; a blood test at a local hospital showed thrombocytopenia; good appetite, normal bowel movements, and no recent weight loss.

[0069] Steps: Clinical information from medical record images is categorized into pre-set keywords to identify key features. If "chief complaint" is identified, the information following the chief complaint is grouped under the "chief complaint" topic, i.e., Chief complaint: Thrombocytopenia for 4 days. If "past medical history" or other related information is identified, the information following the chief complaint is grouped under the "past medical history" topic, i.e., Past medical history: Anemia, jaundice, severe picky eating habits, avoids green vegetables, and consumes little meat. This process is used to identify family history, clinical diagnosis, and other information.

[0070] Step 8: If there are duplicate calls to different preset topics, the priority of clinical information sources is: medical records > application forms > voice. A family tree is generated based on the entered family information. Feature keywords in the medical record text can be identified using a pre-stored medical record feature word library; that is, keywords matching the medical record text in the feature word library are used as feature keywords. Finally, a structured medical record illustration is obtained. Figure 5 As shown. After generating a structured medical record, the generated structured medical record can be modified using the edit command to improve and confirm it, thereby obtaining a complete and valid structured medical record and improving the quality of the structured medical record.

[0071] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for rare disease information entry and gene mutation analysis based on phenotypic matching, characterized in that, include: Step S1: Multi-source clinical information collection, acquiring the patient's clinical information, including data in the form of voice, text or images, and converting it into a structured electronic medical record; Step S2: Medical entity recognition, extracting clinical phenotype keywords from the structured electronic medical record, and performing synonym and hyponym enrichment to construct an expanded clinical phenotype set; Step S3: Phenotypic semantic enrichment and dual-track matching, matching the expanded clinical phenotype set with the CHPO standard phenotype database and / or the local database to obtain a standardized phenotype set; Step S4: Gene list generation, merging candidate genes corresponding to the two disease sets based on the OMIM gene-disease relationship; Step S5: Rank candidates by comprehensive scoring, taking into account at least phenotypic similarity, consistency of inheritance pattern, ACMG variant rating, and local interpretation records. Step S6: Output the sorted gene list. The gene list is output based on the scoring results to assist in the clinical interpretation of gene variations in hereditary diseases.

2. The method for rare disease information entry and gene mutation analysis based on phenotypic matching according to claim 1, characterized in that, The structured electronic medical record generation method in step S1 includes: Convert unstructured clinical information into text using at least one of the following methods: speech recognition and image text recognition; Use a BERT-CRF-based named entity recognition model to extract medical entities from text; The medical entities are categorized and archived according to preset medical record themes to generate structured medical records in JSON or XML format.

3. The method for rare disease information entry and gene mutation analysis based on phenotypic matching according to claim 2, characterized in that, The BERT-CRF model is pre-trained using medical corpora and incorporates a dynamic transition matrix and hierarchical labeling strategy to identify nested medical entities, specifically including: BioBERT-zh, which was further pre-trained on the Chinese medical corpus, was used as the pre-training layer; A BERT encoder enhanced with a medical dictionary; CRF layer using dynamic transition matrix; Identify nested medical entities using a hierarchical labeling strategy; The extracted entities are populated into a preset medical record template to generate structured output in JSON or XML format.

4. The method for rare disease information entry and gene mutation analysis based on phenotypic matching according to claim 1, characterized in that, The synonym and hyponym enrichment step in step S2 uses a medical terminology set to perform semantic enrichment on the clinical phenotype set M, specifically including: The CHPO standard phenotype set S was enriched with synonyms and hyponyms using the OMAHA terminology set and the UMLS medical dictionary to obtain the expanded CHPO phenotype set H. Synonym enrichment was performed on the clinical phenotype set M using the OMAHA and UMLS medical dictionaries; Hyposense enrichment was performed on the clinical phenotype set M using the OMAHA and UMLS medical dictionaries; The union of the above enrichment results yields the expanded clinical phenotype set M' and the CHPO phenotype set H, respectively.

5. The method for rare disease information entry and gene mutation analysis based on phenotypic matching according to claim 1, wherein the matching process in step S3 uses the Jaccard similarity coefficient to calculate the overlap between phenotypic sets and performs term alignment based on semantic relationships, specifically including: Perform intersection and union operations on the CHPO phenotype set corresponding to the candidate genes and the CHPO standard phenotype set obtained in step S3; The Jaccard similarity coefficient is calculated as intersection size / union size; and in the calculation logic of the intersection and union, synonyms and hyponyms are considered to be identical.

6. The method for rare disease information entry and gene mutation analysis based on phenotypic matching according to claim 1, characterized in that, In step S5, when scoring comprehensively, priority is given to the genes of existing matching cases in the local database, and their scores are directly inherited or superimposed on the scores of cases in the local database.

7. The method for rare disease information entry and gene mutation analysis based on phenotypic matching according to claim 1, characterized in that, In step S3, if the clinical phenotype matches the local database, the existing gene interpretation content in the database is directly used to generate a report; if the match is unsuccessful, gene screening and sorting are performed using the CHPO and OMIM databases.

8. A rare disease information entry and gene mutation analysis system based on phenotypic matching, characterized in that, include The multi-source data acquisition module is used to acquire clinical information in the form of voice, images, and text. A medical entity recognition engine is used to identify and extract feature keywords from clinical information; The phenotype enrichment and matching module is used to perform synonym / hypothesis expansion and mapping matching between clinical phenotypes and CHPO phenotypes; The gene screening and sorting module is used to score and sort genes based on factors such as phenotypic matching degree, inheritance mode, and ACMG rating. The local database interface module is used to call approved phenotype-gene-variation records; The structured medical record generation and display module is used to generate and visualize structured medical records and gene lists.

9. The rare disease information entry and gene mutation analysis system based on phenotypic matching according to claim 8, characterized in that, It also includes a report generation module, which automatically generates draft gene testing reports based on a list of genes of interest, for review and publication by interpreters.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images