Interpretation Method, Device, Computer Equipment, Readable Storage Medium and Program Product for Genetic Variation

Abstract

The present application relates to a method, apparatus, computer device, readable storage medium, and program product for interpreting genetic variations. By obtaining genetic data to be interpreted, based on a pre-constructed search database, multimodal data of reference documents corresponding to at least one variant site in the genetic data to be interpreted is determined. According to the at least one variant site and the corresponding multimodal data of the reference documents, corresponding variant rating evidence and the basis for supporting the variant rating evidence are determined. Based on the upgrade / downgrade strategy corresponding to the variant rating evidence, the variant rating evidence is processed for upgrade / downgrade according to the basis for supporting the variant rating evidence, and a variant rating result is obtained. It realizes the full automation of the ACMG literature evidence rating by automatically reading the literature content, which can significantly improve the work efficiency of genetic variation interpretation; and by processing the variant rating evidence for upgrade / downgrade, the accuracy and reliability of the final variant rating result can be significantly improved.

Description

Technical Field

[0001] The present application relates to the field of bioinformatics technology, and in particular, to a method, apparatus, computer device, readable storage medium, and program product for interpreting genetic variations. Background Art

[0002] In genetic diagnosis and genomics research, the assessment of variant pathogenicity is a crucial step. According to the guidelines of the American College of Medical Genetics and Genomics (ACMG), the assessment of genetic variant pathogenicity requires extensive literature support, especially evidence related to de novo variants (PS2, PM6), functional experiments (PS3, BS3), variants with high prevalence in the patient population (PS4), and family co-segregation (PP1, BS4), etc.

[0003] In traditional techniques, genetic counselors must screen variant evidence by carefully reading and analyzing information such as text, pictures, and tables in a large number of medical literature. The traditional ACMG literature-based evidence method relies on manual screening of literature, and analyzes the text descriptions, tables, and picture contents related to variants in the literature one by one. Although this method has provided assistance to genetic counselors in the early stage, with the surge in the number of genetic report interpretations, the method of obtaining evidence by manually reading literature can no longer meet the growing diagnostic needs, and it is not only time-consuming and laborious, but also easily affected by human errors, thus affecting the accuracy. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method, apparatus, computer device, computer-readable storage medium, and computer program product for interpreting genetic variations that can be efficient and accurate.

[0005] In a first aspect, the present application provides a method for interpreting genetic variations, including:

[0006] Obtain the genetic data to be interpreted, where the genetic data to be interpreted includes at least one variant site;

[0007] Based on a pre-constructed search database, determine the multi-modal data of the reference literature corresponding to at least one variant site in the genetic data to be interpreted;

[0008] According to the at least one variant site and the corresponding multi-modal data of the reference literature, determine the corresponding variant rating evidence and the support basis of the variant rating evidence;

[0009] Based on the promotion and demotion strategy corresponding to the variant rating evidence, perform promotion and demotion processing on the variant rating evidence according to the support basis of the variant rating evidence to obtain a variant rating result.

[0010] In one embodiment, the method further includes: extracting multi-modal data related to mutations from medical literature, where the multi-modal data includes at least one of text, images, and tables; constructing a search database based on the extracted multi-modal data related to mutations, where the search database includes the association relationship between mutation information and the multi-modal data.

[0011] In one embodiment, a corresponding mutation rating evidence and the supporting basis of the mutation rating evidence are determined by using a trained multi-modal literature evidence recognition model; the multi-modal literature evidence recognition model is trained by the following method: extracting multi-modal data related to mutations from medical literature, where the multi-modal data includes at least one of text, images, and tables; performing evidence annotation on the multi-modal data based on the literature-based evidence evaluation criteria defined by ACMG to obtain a multi-modal evidence data set; training a selected target multi-modal large language model by using the LoRA fine-tuning method based on the multi-modal evidence data set to obtain the multi-modal literature evidence recognition model.

[0012] In one embodiment, the determining of the corresponding mutation rating evidence and the supporting basis of the mutation rating evidence by using a trained multi-modal literature evidence recognition model includes: generating evidence recognition prompt information according to the at least one mutation site and the multi-modal data of the corresponding reference; inputting the evidence recognition prompt information into the multi-modal literature evidence recognition model to obtain an evidence recognition result output by the multi-modal literature evidence recognition model, where the evidence recognition result includes the mutation rating evidence and the supporting basis of the mutation rating evidence.

[0013] In one embodiment, the upgrading and downgrading processing of the mutation rating evidence according to the supporting basis of the mutation rating evidence based on the upgrading and downgrading strategy corresponding to the mutation rating evidence to obtain a mutation rating result includes: generating upgrading and downgrading processing prompt information based on the mutation rating evidence, the supporting basis of the mutation rating evidence, and the upgrading and downgrading strategy corresponding to the mutation rating evidence; inputting the upgrading and downgrading processing prompt information into the multi-modal literature evidence recognition model to obtain a mutation rating result output by the multi-modal literature evidence recognition model.

[0014] In one embodiment, the extracting of multi-modal data related to mutations from medical literature includes: parsing the medical literature and converting the medical literature into a structured multi-modal data format; extracting multi-modal data related to mutations from the structured multi-modal data.

[0015] In a second aspect, the present application further provides an apparatus for interpreting genetic mutations, where the apparatus includes:

[0016] An acquisition module for acquiring genetic data to be interpreted, where the genetic data to be interpreted includes at least one variant site;

[0017] A first determination module for determining multimodal data of references corresponding to at least one variant site in the genetic data to be interpreted based on a pre - constructed search database;

[0018] A second determination module for determining corresponding variant rating evidence and the basis for supporting the variant rating evidence according to the at least one variant site and the multimodal data of the corresponding references;

[0019] A result output module for performing an upgrade / downgrade process on the variant rating evidence according to the basis for supporting the variant rating evidence based on an upgrade / downgrade strategy corresponding to the variant rating evidence to obtain a variant rating result.

[0020] In a third aspect, the present application also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the method described in the first aspect above are implemented.

[0021] In a fourth aspect, the present application also provides a computer - readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in the first aspect above are implemented.

[0022] In a fifth aspect, the present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the method described in the first aspect above are implemented.

[0023] The above - mentioned method, device, computer device, computer - readable storage medium and computer program product for interpreting genetic variations, by acquiring genetic data to be interpreted, determining multimodal data of references corresponding to at least one variant site in the genetic data to be interpreted based on a pre - constructed search database, determining corresponding variant rating evidence and the basis for supporting the variant rating evidence according to the at least one variant site and the multimodal data of the corresponding references, and performing an upgrade / downgrade process on the variant rating evidence according to the basis for supporting the variant rating evidence based on an upgrade / downgrade strategy corresponding to the variant rating evidence to obtain a variant rating result. It realizes the full automation of ACMG literature evidence rating by automatically reading the literature content, which can significantly improve the work efficiency of genetic variation interpretation; and by performing an upgrade / downgrade process on the variant rating evidence, it can significantly improve the accuracy and reliability of the final variant rating result. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0025] Figure 1 It is a schematic flowchart of a method for interpreting genetic variations in an embodiment;

[0026] Figure 2 It is a schematic flowchart of the model training step in an embodiment;

[0027] Figure 3 It is a schematic flowchart of the upgrade and downgrade processing step in an embodiment;

[0028] Figure 4 It is a structural block diagram of an apparatus for interpreting genetic variations in an embodiment;

[0029] Figure 5 It is an internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0030] In order to make the purpose, technical solutions and advantages of the present application clearer, the following further details the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0031] Since the traditional method of obtaining evidence for genetic report interpretation by manually reading literature can no longer meet the growing diagnostic needs, and it is time-consuming and laborious, and is also prone to affecting accuracy due to human errors.

[0032] Although there are related tools such as InterVar (a bioinformatics software tool) and VarSome (a powerful genetic variation annotation and analysis tool) that can achieve partial automation of non-literature evidence analysis, manual judgment is still required for literature evidence. Most tools that can achieve automatic rating of ACMG evidence items focus on the perspective of bioinformatics analysis and automate the prediction of partial non-literature evidence. For the evaluation of literature evidence, it still mainly relies on manual reading and interpretation. Although some studies have proposed the idea of automatic extraction of literature evidence, these methods are usually limited to building a local literature evidence library for extracting literature evidence of some already interpreted variations and have not been able to fully achieve automatic rating of literature evidence. At the same time, traditional methods lack in-depth understanding and analysis of multi-modal data such as text and charts. Therefore, traditional tools are still difficult to accurately achieve full automatic recognition of literature evidence.

[0033] Based on this, the embodiments of the present application provide a method for interpreting genetic variations. By making full use of the technical advantages of multi-modal large models and combining with the ACMG literature evidence rules, it automatically identifies multi-modal evidence information such as text, pictures, and tables from medical literature. It can also identify and analyze pictures (such as pedigree charts, functional experiment result charts) and tables (such as case data, statistical result tables) in the literature, so as to comprehensively integrate various forms of evidence information in the literature and improve the accuracy and efficiency of ACMG literature evidence interpretation in genetic interpretation.

[0034] In one embodiment, as Figure 1 shown, a method for interpreting genetic variations is provided, and the method includes the following steps:

[0035] Step 102, obtain the genetic data to be interpreted.

[0036] Among them, genetic data is an important basis for studying biological genetic variations, gene functions, and evolution, covering biomedical data with multi-level information such as genomes, transcriptomes, and epigenomes. In this embodiment, the genetic data to be interpreted includes at least one variant site. Specifically, a variant site refers to the specific position where a variation occurs in the genome, usually one or more bases in the genomic sequence have changed. These variations can occur at any position in the DNA (DeoxyriboNucleic Acid) sequence, including coding regions (such as inside genes) and non-coding regions (such as intergenic regions, regulatory elements, etc.). Variant sites are important concepts in genetics and genomics research because they may affect gene functions, expressions, and regulations, and thus affect the traits of organisms or cause diseases.

[0037] Specifically, when interpreting genetic variations, it is first necessary to obtain the genetic data to be interpreted, and the genetic data to be interpreted includes at least one variant site. Then, through subsequent steps of analysis and interpretation, the dynamic changes of the genome and its effects on biological traits can be revealed. For example, certain variant sites may be related to genetic diseases or complex diseases. By analyzing the variant sites, personalized treatment plans can be developed.

[0038] Step 104, based on the pre-constructed search database, determine the multi-modal data of the reference corresponding to at least one variant site in the genetic data to be interpreted.

[0039] Among them, the search database is pre-constructed based on medical references and includes the association relationship between variant information and multi-modal data in the literature. Variant information is entities such as genes, variations, and diseases that are valuable for judging ACMG literature evidence. Multi-modal data includes at least one of text, images, and tables in the literature related to the variant information.

[0040] In this embodiment, by pre - constructing a search database, after obtaining the genetic data to be interpreted based on the above steps, the multimodal data of the reference literature corresponding to at least one variant site in the genetic data to be interpreted can be determined based on the pre - constructed search database. Specifically, the multimodal data of the reference literature associated with the at least one variant site can be searched in the search database to query the multimodal data of the reference literature associated with the at least one variant site.

[0041] Step 106: Determine the corresponding variant rating evidence and the supporting basis of the variant rating evidence according to the at least one variant site and the multimodal data of the corresponding reference literature.

[0042] Among them, the variant rating evidence refers to the literature - type evidence that conforms to the ACMG guidelines, including but not limited to benign evidence items (such as categories BS3, BS4, BP2, BP5, etc.) and pathogenic evidence items (such as categories PS2, PS3, PS4, PM3, PM6, PP1, PP4, etc.). The supporting basis can be key information such as the number of probands, functional experimental results, and de novo mutation situations related to the variant rating evidence extracted and summarized from the literature.

[0043] Specifically, according to the at least one variant site and the multimodal data of the corresponding reference literature, the corresponding variant rating evidence and the supporting basis of the variant rating evidence can be determined through analysis and evaluation.

[0044] Step 108: Based on the up - grading and down - grading strategy corresponding to the variant rating evidence, perform up - grading and down - grading processing on the variant rating evidence according to the supporting basis of the variant rating evidence to obtain the variant rating result.

[0045] Among them, the up - grading and down - grading strategy can be the up - grading and down - grading judgment criteria of the literature evidence items formulated based on the characteristics of different variant rating evidences following the ACMG guidelines. The variant rating result refers to the final evaluation result for the genetic data to be interpreted obtained after performing up - grading and down - grading processing on the variant rating evidence.

[0046] In this embodiment, based on the up - grading and down - grading strategy corresponding to the variant rating evidence, perform up - grading and down - grading processing on the variant rating evidence determined in the above steps according to the supporting basis of the variant rating evidence, so as to obtain a more accurate variant rating result.

[0047] In the above method for interpreting genetic variations, by obtaining the genetic data to be interpreted, based on a pre-constructed search database, multimodal data of references corresponding to at least one variant site in the genetic data to be interpreted is determined. According to the at least one variant site and the multimodal data of the corresponding references, the corresponding variant rating evidence and the supporting basis for the variant rating evidence are determined. Based on the promotion and demotion strategy corresponding to the variant rating evidence, the variant rating evidence is promoted or demoted according to the supporting basis of the variant rating evidence to obtain the variant rating result. It realizes the full automation of the ACMG literature evidence rating by automatically reading the literature content, which can significantly improve the work efficiency of genetic variation interpretation; and by promoting or demoting the variant rating evidence, the accuracy and reliability of the final variant rating result can be significantly improved.

[0048] In one embodiment, in step 106, determining the corresponding variant rating evidence and the supporting basis for the variant rating evidence may specifically include: determining the corresponding variant rating evidence and the supporting basis for the variant rating evidence by using a trained multimodal literature evidence recognition model. Among them, the multimodal literature evidence recognition model may be an artificial intelligence model trained to accurately identify ACMG literature evidence items related to variants (including variant rating evidence and corresponding supporting bases).

[0049] In this embodiment, evidence recognition prompt information may be generated according to at least one variant site in the genetic data to be interpreted and the multimodal data of the corresponding references; and the evidence recognition prompt information is input into the multimodal literature evidence recognition model, so as to obtain the evidence recognition result output by the multimodal literature evidence recognition model, where the evidence recognition result includes variant rating evidence and the supporting basis for the variant rating evidence.

[0050] Specifically, the evidence recognition prompt information may be relevant prompts for instructing the multimodal literature evidence recognition model to perform literature evidence recognition, and may specifically include task objective prompts. For example, the task objective prompt may be in a similar form of "According to the input variant information and relevant literature information, quickly and accurately identify the literature evidence types supported by the variant following the ACMG guidelines". Among them, the specific content of the "variant information" may be at least one variant site in the input genetic data to be interpreted, and at least one of the HGVS (Human Genome Variation Society) expression of the variant site, gene alias, and gene-related disease information. The specific content of the "relevant literature information" may be the multimodal data of the references determined to correspond to at least one variant site, such as relevant text fragments and charts.

[0051] In one scenario, the evidence recognition prompt information may also include a defined output format prompt. For example, the output format prompt may specifically be in a similar form such as "If the literature supports multiple evidence types, return a list of evidence rating types that conform to the ACMG standard and their reasons in JSON format; if there is no supported evidence, return an empty JSON {}." This enables the model to output evidence recognition results that meet the requirements.

[0052] In addition, the evidence recognition prompt information may also include a defined task prompt. For example, the task prompt may specifically be in a similar form such as "Since the PS2 priority is high, if it is rated as PS2, there is no need to rate PM6. If there is not enough evidence in the literature to support any evidence type of the ACMG standard, return an empty JSON {}. For parts involving pedigree information, mutation frequency, etc., data such as the number of probands and the number of patients in the pedigree need to be counted." This can enhance the reasoning ability and understanding ability of the model, and thus improve the processing efficiency and accuracy of the model.

[0053] In an exemplary embodiment, as Figure 2 shown, the above multi-modal literature evidence recognition model can be trained through the following method:

[0054] Step 202, extract multi-modal data related to mutations from medical literature.

[0055] Among them, medical literature is valuable reference materials related to medicine. The multi-modal data includes at least one of text, images, and tables in the literature.

[0056] In this embodiment, the medical literature can be parsed and processed to convert the medical literature into a structured multi-modal data format; then, extract multi-modal data related to mutations from the structured multi-modal data.

[0057] Specifically, the medical literature (such as a PDF-format literature) can be deeply parsed, and the text content such as the title, abstract, keywords, text, references, supplementary materials, etc. in the literature, as well as non-text elements such as tables and pictures, can be converted into a data structure convenient for subsequent analysis. During this process, the original structure of the literature is retained, and the tables are also saved in the form of pictures, and an accurate relationship index between the text and the charts (pictures and tables) is established. It is also possible to accurately identify genes, mutations, and disease entities in the literature based on natural language processing entity extraction technology, and perform standardized processing on these entities through entity alignment technology, so as to obtain the extracted multi-modal data related to mutations.

[0058] In one scenario, in view of the huge amount of information in medical literature and the possible problem of information overload during storage and use, a binary classification model can also be used to screen out text fragments valuable for evidence judgment, so as to eliminate article fragments that cannot be used as ACMG literature evidence items and only retain text fragments valuable for evidence judgment.

[0059] In another scenario, the association relationships between variant information such as genes, variants, and disease entities contained in each text fragment, as well as the index information of related multimodal data such as text fragments, pictures, and tables, can be stored in a local knowledge graph or an Elasticsearch (search and analysis engine) database, so as to obtain a constructed search database, laying a solid foundation for subsequent evidence rating work.

[0060] Step 204: Based on the literature evidence evaluation criteria defined by ACMG, perform evidence annotation on the multimodal data to obtain a multimodal evidence data set.

[0061] Since finding literature evidence items that meet the ACMG guidelines is a complex and meticulous task. Evidence is not only hidden in text fragments, but sometimes it is also necessary to comprehensively analyze multimodal information such as tables and pictures (such as experimental results, pedigree charts) to accurately extract specific literature evidence items. If only relying on the text content itself may not be sufficient to fully confirm specific evidence items in the literature, and other relevant evidence items may also be missed. Therefore, in this embodiment, based on the literature evidence evaluation criteria defined by ACMG, evidence annotation can be performed on the multimodal data, so as to obtain a multimodal evidence data set, and then the multimodal evidence data set can be used for model training, thereby enhancing the comprehensiveness and accuracy of the model in the process of literature evidence rating.

[0062] Among them, the literature evidence evaluation criteria defined by ACMG can be in the form shown in Table 1 below, including pathogenic evidence items such as PS2, PS3, PS4, PM3, PM6, PP1, PP4, and benign evidence items such as BS3, BS4, BP2, BP5. By annotating the multimodal evidence data, more accurate support can be provided for the evaluation of literature evidence. Table 1 is as follows:

[0063]

[0064] In this embodiment, for each variant that meets the ACMG literature evidence rating, evidence items (such as PS2, PM3, PP1, etc. and their supporting bases) can be marked for the corresponding literature fragments and charts. Specifically, the following two methods can be used to collect and annotate evidence data:

[0065] (1) Annotations are made based on the accumulation of the internal report interpretation system. Since the internal report interpretation system records the literature fragments, tables, and pictures that genetic counselors have previously interpreted as references for variant evidence ratings, it is possible to annotate the corresponding evidence items for these literature fragments, tables, and pictures based on this record.

[0066] (2) Annotations are made with the assistance of large models. For example, by designing prompts that conform to the ACMG evidence rating criteria and leveraging the language understanding ability of large models, the text and chart content of variants and related literature can be analyzed to obtain possible evidence fragments and charts for annotation.

[0067] In one scenario, the above-mentioned annotated content can also be reviewed and corrected to further improve the accuracy of the annotated content.

[0068] In one scenario, to better manage and annotate multimodal data, the annotated multimodal evidence data can also be centrally managed and stored. The stored content can include the mapping relationships between variant information, evidence items (i.e., variant rating evidence and supporting bases), reference literature IDs, literature fragments (i.e., text fragments in the corresponding literature), reference table links, and reference picture links, etc. Among them, the table content can be stored in picture format and saved to the local database together with the relevant pictures.

[0069] In one scenario, for the situation where the same literature content or the same chart may involve multiple evidence items, a multi-label method can also be used for annotation. For example, for the literature fragment with the literature ID "24664454", since it involves multiple evidence items (i.e., variant rating evidence and supporting bases) such as PS2 and PS4, its evidence items can be annotated based on multiple labels such as PS2 and PS4.

[0070] In an exemplary embodiment, in addition to annotating the evidence items in the literature that conform to the ACMG guidelines, in order to enhance the model's understanding of the chart information in the evidence rating data, some descriptions corresponding to the charts can also be annotated. Especially the relevant information of pedigree charts, functional experiment charts, and variant tables, to enhance the model's in-depth understanding of the charts that meet the evidence rating criteria. For example, the understanding of the filial relationship, inheritance pattern, and disease status of variant carriers in the pedigree chart can be annotated; the understanding of changes in protein expression levels and enzyme activities in the functional experiment chart can be annotated, etc. It can be understood that this part annotates the descriptive statements corresponding to the charts that are meaningful for evidence rating in the literature, rather than evidence items (such as PS2, PP1, etc.).

[0071] Step 206: Based on the multimodal evidence dataset, use the LoRA fine-tuning method to train the selected target multimodal large language model to obtain a multimodal literature evidence recognition model.

[0072] In this embodiment, the model training process may include multi-modal instruction fine-tuning data design, screening of the target multi-modal large language model during the fine-tuning training process, fine-tuning strategy selection, etc.

[0073] Among them, for the multi-modal instruction fine-tuning data design, the rating criteria of evidence items such as PS2, PS3, PS4, PM3, PM6, PP1, PP4, BS3, BS4, BP2, and BP5 in the ACMG guidelines can be combined, and the multi-modal evidence instructions (i.e., evidence recognition prompt information) can be written in a chain of thought manner. Since the diseases associated with genes are of great significance in the rating process, for example, for the PP4 evidence item, it is necessary to judge the high conformity between the patient's phenotype and the disease phenotype. Therefore, the disease information related to gene mutations should be included in the prompt to determine whether the phenotype or family history of the cases in the literature highly conforms to the phenotype of the disease.

[0074] During the model training process, two types of data, namely multi-modal evidence data sets and pure text evidence data sets, can be used as training data and test data, that is, both the training data and the test data include these two types of data. Among them, the multi-modal evidence data includes text evidence data and relevant chart information, while the pure text evidence data does not include chart information. Since in the above data collection and annotation process, in addition to collecting multi-modal literature evidence data, the chart description data is also specifically annotated. And the chart description data can be sorted in a format similar to the multi-modal evidence instruction data, thus ensuring the consistency of the data structure.

[0075] During the model training process, first, according to the local hardware environment and computing resources, multiple pre-trained multi-modal large language models can be selected as the base models. These models can simultaneously analyze natural language text information and pictures, can understand multi-modal data integrating text and pictures, and have leading pure text understanding capabilities. For example, specifically, the Qwen-VL series models (a large-scale vision language model) or the multi-modal Pixtral series models of Mistral AI (a company focusing on the field of artificial intelligence, which is a multi-modal language model released by Mistral AI) can be used.

[0076] Secondly, the base models are screened. Specifically, the test data of the above-organized instruction data can be input into multiple base models for prediction respectively, and the prediction results are statistically analyzed. The base models that perform well in the ACMG literature evidence rating task are selected as the target multi-modal large language models to ensure a better understanding of the task. This not only helps to avoid poor fine-tuning effects caused by large differences between the pre-training and fine-tuning data, but also can accelerate the convergence speed of model fine-tuning.

[0077] Then, an efficient fine-tuning strategy is selected. Since full-parameter fine-tuning requires high computing resources and takes a long time to train, the LoRA (Low-Rank Adaptation, an efficient fine-tuning technique for pre-trained models) fine-tuning method is adopted in this embodiment. By using the LoRA fine-tuning method, only a small part of the parameters need to be updated, which can reduce the consumption of computing resources while maintaining most of the performance of the pre-trained model.

[0078] Finally, based on the test data of multi-modal instructions including multi-modal evidence instructions data and chart description instructions data, multi-task fine-tuning training is performed on the above-selected target multi-modal large language model to obtain a multi-modal literature evidence recognition model. Through this multi-modal multi-task fine-tuning method, not only the model's understanding ability of various types of charts in the literature (especially pedigree charts, variant function experimental charts, and variant-related tables) is significantly improved, but also it is ensured that the model can accurately identify ACMG literature evidence items related to variants. Therefore, the obtained multi-modal literature evidence recognition model after training can be used to interpret actual genetic data, providing an efficient and accurate solution for genetic variant interpretation.

[0079] In an exemplary embodiment, as Figure 3 shown, in step 108, based on the upgrade / downgrade strategy corresponding to the variant rating evidence, the variant rating evidence is processed for upgrade / downgrade according to the supporting basis of the variant rating evidence to obtain a variant rating result. Specifically, it may further include:

[0080] Step 302, generating upgrade / downgrade processing prompt information based on the variant rating evidence, the supporting basis of the variant rating evidence, and the upgrade / downgrade strategy corresponding to the variant rating evidence.

[0081] Among them, the promotion / demotion processing prompt information is similar to the above-mentioned evidence recognition prompt information, and is a relevant prompt for instructing the multi-modal literature evidence recognition model to perform promotion / demotion processing. Specifically, it may include promotion / demotion instruction commands. For example, the promotion / demotion instruction command may include a similar form such as "You are an expert in ACMG variant interpretation, and can perform promotion / demotion processing on the evidence type of the variant according to the ACMG evidence promotion / demotion strategy and the evidence basis, and output the promotion / demotion result". Among them, the "evidence basis" includes variant rating evidence and the corresponding supporting basis, which can specifically be obtained by the fine-tuned multi-modal literature evidence recognition model. For example, the evidence basis may specifically be in a similar form such as "Evidence basis: PS4 literature PMID 22464770 reports 1 proband with cardiomyopathy. PS4 literature PMID 24503780 reports 1 proband with cardiomyopathy. PS4 literature PMID 27532257 reports 1 case with cardiomyopathy. PS4 literature PMID 30996762 reports 3 siblings with isolated biventricular non-compaction cardiomyopathy. PS4 literature PMID 28855170 reports 1 case with cardiomyopathy. PS4 literature PMID 24664454 reports 2 probands with cardiomyopathy."

[0082] The "promotion / demotion strategy" refers to the promotion / demotion description of the corresponding evidence item stipulated by ACMG. For example, for the evidence item PS4 (the frequency of this variant in the affected population is significantly higher than that in the control population. Extremely rare variants may have no statistical significance in case-control studies. If this variant is observed in multiple probands with the same phenotype, different levels of PS4 evidence grades can be given according to the number of probands), the promotion / demotion strategy may specifically be in a similar form such as "If there are 15 or more probands carrying the variant to be evaluated, assign PS4; if there are 6 or more probands carrying the variant to be evaluated and less than 15, assign PS4_Moderate; if there are 2 or more probands carrying the variant to be evaluated and less than 6, assign PS4_Supporting. Note: Only count the number of probands, and the number of probands in the literature is not counted repeatedly."

[0083] In this embodiment, the corresponding promotion / demotion processing prompt information can be generated based on the variant rating evidence, the supporting basis of the variant rating evidence, and the promotion / demotion strategy corresponding to the variant rating evidence. Usually, based on internal business requirements, the rating results for benign evidence (including BS3, BS4, BP2, BP5) and PP4 can maintain the original rating unchanged, while the rating results for these evidence items such as PS2, PS3, PS4, PM3, PM6, and PP1 can be promoted / demoted based on the corresponding promotion / demotion strategy, so as to achieve a comprehensive and accurate assessment of the pathogenicity of the variant.

[0084] Step 304: Input the upgrade / downgrade processing prompt information into the multi-modal literature evidence recognition model to obtain the variant rating result output by the multi-modal literature evidence recognition model.

[0085] Among them, the multi-modal literature evidence recognition model is obtained through the above-mentioned fine-tuning training. This model can not only accurately extract evidence items related to variants, but also intelligently analyze and summarize the supporting basis of the evidence items. For example, for the PS4 evidence item, the model can automatically extract and summarize key information such as the number of probands. Based on the input variant information, the variant rating evidence output by the model and its supporting basis, combined with the ACMG guidelines, the upgrade / downgrade strategies for the corresponding literature evidence items formulated for the characteristics of each evidence item of PS2, PS3, PS4, PM3, PM6, and PP1 can generate corresponding upgrade / downgrade processing prompt information. By inputting this upgrade / downgrade processing prompt information into the multi-modal literature evidence recognition model, the variant rating result output by the model can be obtained. It can be understood that the upgrade / downgrade processing prompt information can drive the multi-modal literature evidence recognition model to achieve the upgrade / downgrade processing of literature-based evidence, thereby significantly improving the accuracy and reliability of variant pathogenicity assessment.

[0086] It should be understood that although each step in the flowcharts involved in the above-described embodiments is displayed sequentially as indicated by the arrows, these steps do not necessarily need to be executed sequentially in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages do not necessarily need to be executed at the same moment, but can be executed at different moments. The execution order of these steps or stages does not necessarily need to be sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0087] Based on the same inventive concept, the embodiments of the present application also provide a genetic variant interpretation device for implementing the genetic variant interpretation method described above. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the genetic variant interpretation device provided below can refer to the limitations on the genetic variant interpretation method in the above text, and will not be repeated here.

[0088] In an exemplary embodiment, as Figure 4 shown, a genetic variant interpretation device is provided, including: an acquisition module 402, a first determination module 404, a second determination module 406, and a result output module 408, where:

[0089] An acquisition module 402, configured to acquire genetic data to be interpreted, where the genetic data to be interpreted includes at least one variant site;

[0090] A first determination module 404, configured to determine multimodal data of reference documents corresponding to at least one variant site in the genetic data to be interpreted based on a pre-constructed search database;

[0091] A second determination module 406, configured to determine corresponding variant rating evidence and a support basis for the variant rating evidence according to the at least one variant site and the multimodal data of the corresponding reference document;

[0092] A result output module 408, configured to perform an upgrade / downgrade process on the variant rating evidence according to the support basis of the variant rating evidence based on an upgrade / downgrade strategy corresponding to the variant rating evidence, and obtain a variant rating result.

[0093] In an exemplary embodiment, the apparatus further includes a search database construction module, configured to extract multimodal data related to variants from medical literature, where the multimodal data includes at least one of text, images, and tables; and construct a search database based on the extracted multimodal data related to variants, where the search database includes an association relationship between variant information and the multimodal data.

[0094] In an exemplary embodiment, the second determination module may specifically be configured to determine corresponding variant rating evidence and a support basis for the variant rating evidence by using a trained multimodal literature evidence recognition model.

[0095] In an exemplary embodiment, the multimodal literature evidence recognition model is trained by the following method: extracting multimodal data related to variants from medical literature, where the multimodal data includes at least one of text, images, and tables; performing evidence annotation on the multimodal data based on the literature evidence evaluation criteria defined by ACMG to obtain a multimodal evidence data set; and training a selected target multimodal large language model by using the LoRA fine-tuning method based on the multimodal evidence data set to obtain the multimodal literature evidence recognition model.

[0096] In an exemplary embodiment, the second determination module may further be configured to: generate evidence recognition prompt information according to the at least one variant site and the multimodal data of the corresponding reference document; and input the evidence recognition prompt information into the multimodal literature evidence recognition model to obtain an evidence recognition result output by the multimodal literature evidence recognition model, where the evidence recognition result includes the variant rating evidence and the support basis for the variant rating evidence.

[0097] In an exemplary embodiment, the result output module is further configured to: generate an upgrade / downgrade processing prompt message based on the mutation rating evidence, the basis for supporting the mutation rating evidence, and the upgrade / downgrade strategy corresponding to the mutation rating evidence; input the upgrade / downgrade processing prompt message into the multi-modal literature evidence recognition model to obtain the mutation rating result output by the multi-modal literature evidence recognition model.

[0098] In an exemplary embodiment, the apparatus further includes a data extraction module, configured to: parse the medical literature, convert the medical literature into a structured multi-modal data format; extract multi-modal data related to mutations from the structured multi-modal data.

[0099] Each module in the above genetic mutation interpretation apparatus can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor in the computer device in the form of hardware or be independent of it, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above respective modules.

[0100] In an exemplary embodiment, a computer device is provided, and its internal structure diagram can be as Figure 5 shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC), or other technologies. The computer program, when executed by the processor, implements a genetic mutation interpretation method. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0101] Those skilled in the art can understand that Figure 5The structure shown is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0102] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0103] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0104] In an embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0105] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0106] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0107] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope recorded in this application.

[0108] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for interpreting genetic variations, characterized in that, The method includes: Obtaining genetic data to be interpreted, where the genetic data to be interpreted includes at least one variant site; Based on a pre-constructed search database, determining multi-modal data of reference documents corresponding to at least one variant site in the genetic data to be interpreted; the search database is pre-constructed based on medical reference documents and includes the association relationship between variant information and multi-modal data in the documents. The variant information is an entity valuable for judging ACMG literature evidence, and the multi-modal data includes at least one of text, images, and tables in medical reference documents; According to the at least one variant site and the corresponding multi-modal data of the reference document, using a multi-modal literature evidence recognition model to determine the corresponding variant rating evidence and the supporting basis for the variant rating evidence; the multi-modal literature evidence recognition model is an artificial intelligence model obtained through training and capable of recognizing ACMG literature evidence related to variants. The ACMG literature evidence includes variant rating evidence and the corresponding supporting basis; Based on the upgrade / downgrade strategy corresponding to the variant rating evidence, performing upgrade / downgrade processing on the variant rating evidence according to the supporting basis of the variant rating evidence to obtain a variant rating result; The performing upgrade / downgrade processing on the variant rating evidence according to the supporting basis of the variant rating evidence based on the upgrade / downgrade strategy corresponding to the variant rating evidence to obtain a variant rating result includes: Obtaining the upgrade / downgrade strategy corresponding to the variant rating evidence as defined by ACMG based on the variant rating evidence, extracting key information from the supporting basis of the variant rating evidence according to the upgrade / downgrade strategy, and performing upgrade / downgrade processing on the variant rating evidence based on the key information to obtain a variant rating result.

2. The method according to claim 1, characterized in that The method further includes: Extracting multi-modal data related to variants from medical literature; Based on the extracted multi-modal data related to variants, constructing a search database, where the search database includes the association relationship between variant information and the multi-modal data.

3. The method according to claim 1, wherein The multi-modal literature evidence recognition model is obtained through the following method: Extracting multi-modal data related to variants from medical literature, where the multi-modal data includes at least one of text, images, and tables; Based on the literature evidence evaluation criteria defined by ACMG, performing evidence annotation on the multi-modal data to obtain a multi-modal evidence data set; Based on the multi-modal evidence data set, using the LoRA fine-tuning method to train a selected target multi-modal large language model to obtain the multi-modal literature evidence recognition model.

4. The method according to claim 3, wherein The using the multi-modal literature evidence recognition model to determine the corresponding variant rating evidence and the supporting basis for the variant rating evidence includes: Generating evidence recognition prompt information according to the at least one variant site and the corresponding multi-modal data of the reference document; Inputting the evidence recognition prompt information into the multi-modal literature evidence recognition model to obtain an evidence recognition result output by the multi-modal literature evidence recognition model, where the evidence recognition result includes the variant rating evidence and the supporting basis for the variant rating evidence.

5. The method according to claim 3, characterized in that, Based on the promotion and demotion strategy corresponding to the variant rating evidence, the variant rating evidence is promoted or demoted according to the supporting basis of the variant rating evidence to obtain a variant rating result, including: Generating a promotion and demotion processing prompt message based on the variant rating evidence, the supporting basis of the variant rating evidence, and the promotion and demotion strategy corresponding to the variant rating evidence; Inputting the promotion and demotion processing prompt message into the multi-modal literature evidence recognition model to obtain the variant rating result output by the multi-modal literature evidence recognition model.

6. The method according to any one of claims 2 to 5, characterized in that Extracting multi-modal data related to variants from medical literature, including: Performing parsing processing on the medical literature to convert the medical literature into a structured multi-modal data format; Extracting multi-modal data related to variants from the structured multi-modal data.

7. An apparatus for interpreting genetic variations, characterized in that, The device includes: An acquisition module, configured to acquire genetic data to be interpreted, where the genetic data to be interpreted includes at least one variant site; A first determination module, configured to determine multi-modal data of a reference document corresponding to at least one variant site in the genetic data to be interpreted based on a pre-constructed search database; the search database is pre-constructed based on medical reference documents and includes an association relationship between variant information and multi-modal data in the documents, the variant information is an entity valuable for judging ACMG literature evidence, and the multi-modal data includes at least one of text, images, and tables in medical reference documents related to the variant information; A second determination module, configured to determine corresponding variant rating evidence and the supporting basis of the variant rating evidence by using a multi-modal literature evidence recognition model according to the at least one variant site and the corresponding multi-modal data of the reference document; the multi-modal literature evidence recognition model is an artificial intelligence model obtained through training and capable of recognizing ACMG literature evidence related to variants, and the ACMG literature evidence includes variant rating evidence and corresponding supporting basis; A result output module, configured to perform promotion and demotion processing on the variant rating evidence according to the supporting basis of the variant rating evidence based on the promotion and demotion strategy corresponding to the variant rating evidence to obtain a variant rating result; Based on the promotion and demotion strategy corresponding to the variant rating evidence, the variant rating evidence is promoted or demoted according to the supporting basis of the variant rating evidence to obtain a variant rating result, including: Obtaining the promotion and demotion strategy corresponding to the variant rating evidence specified by ACMG based on the variant rating evidence, extracting key information from the supporting basis of the variant rating evidence according to the promotion and demotion strategy, and performing promotion and demotion processing on the variant rating evidence based on the key information to obtain a variant rating result.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images