Enzyme activity parameter extraction method and system based on large language model

Abstract

The invention provides an enzyme activity parameter extraction method and system based on a large language model, and belongs to the field of text information processing and data mining. According to the method, the document in the PDF format is converted into the Markdown format through the OCR technology, and then key data are automatically extracted by combining a large language model with an optimized cue word; the accuracy and consistency of data are ensured through strict cue word optimization and post-processing operation in the automatic extraction process; and verifying the effectiveness of automatic extraction through accuracy and recall rate, wherein the finally generated enzyme database can be used for subsequent research and analysis. By combining the OCR technology and the large language model, the limitation of existing manual data extraction is broken through, and the automation degree of literature analysis and data extraction is remarkably improved; through cue word engineering and optimized cue word design, automatic extraction of structured data in complex literatures is achieved, and especially accurate recognition and extraction of enzyme kinetic parameters are achieved.

Description

Technical Field

[0001] The invention relates to the field of text information processing and data mining, and in particular to a method and system for extracting enzyme activity parameters based on a large language model. Background Art

[0002] In biomedical research, enzyme activity parameters are one of the important indicators for studying enzyme catalytic reactions. With the exponential growth of literature data, how to quickly, efficiently and accurately extract enzyme activity parameters from massive biomedical literature has become a major challenge. Traditional manual extraction methods have the problems of large workload, low efficiency, and prone to errors, and cannot adapt to the growing demand for literature.

[0003] Most of the existing automated information extraction technologies rely on rule matching or keyword-based methods, which perform poorly when dealing with complex literature content. In recent years, with the rapid development of large language models, natural language processing (NLP) technology has made significant progress, especially in understanding complex language patterns and processing large amounts of unstructured data. Therefore, a method and system for extracting enzyme activity parameters based on a large language model is proposed, which can effectively improve the accuracy and efficiency of enzyme activity parameter extraction. Summary of the invention

[0004] The present invention aims to solve the problems of low efficiency and insufficient accuracy in the existing enzyme activity parameter extraction process. By combining the natural language processing capabilities of a large language model, the optimization technology of prompt word engineering, and the standard operation of manual extraction, a method and system for extracting enzyme activity parameters based on a large language model are provided. The present invention can significantly reduce the manual workload and improve the accuracy and reliability of extracting enzyme activity parameters from biomedical literature.

[0005] To achieve the above object, the present invention provides a method for extracting enzyme activity parameters based on a large language model, comprising the following steps: (1) Screen out literature related to enzyme activity from the biomedical literature database; (2) Perform optical character recognition on the selected documents to generate text data; (3) splitting the text data into sample data containing different features to generate difficult samples and random samples; (4) Extracting artificial enzyme activity parameters for the difficult samples to generate golden benchmark data; (5) using a large language model driven by prompt words to automatically extract enzyme activity parameters from the random samples through prompt word optimization, and post-processing the automatic extraction results to generate structured enzyme activity parameter data; (6) Combining the manual extraction result in step (4) with the automatic extraction result in step (5), the enzyme activity parameter data automatically extracted are corrected and optimized using the golden benchmark data to obtain the final enzyme activity parameter data.

[0006] Furthermore, in step (3), the difficult sample is the document containing the most enzyme function parameter data; and the random sample is the document randomly selected from the document pool.

[0007] Furthermore, in step (5), the prompt words are: "Please extract the enzyme name, source organism, kinetic constant, pH and temperature in the table"; "Please identify and extract enzyme mutant information and experimental data on their effects on enzyme activity in the literature"; in the prompt word optimization process, random samples and difficult samples are used to evaluate the prompt words, and samples with a recall rate of less than 50% are manually analyzed and optimized. In each optimization, the text gradient is gradually increased to solve specific problems.

[0008] Furthermore, the recall rate is obtained by dividing the amount of correctly extracted data by the total amount of data in the golden benchmark data.

[0009] Furthermore, in step (5), the post-processing operation includes: cleaning invalid fields and empty values, standardizing digital formats, and unifying units.

[0010] To achieve the above object, the present invention also provides an enzyme activity parameter extraction system based on a large language model, comprising: Literature filtering module, used to filter literature related to enzyme activity from biomedical literature; An optical character recognition module is used to perform optical character recognition on the screened documents to generate text data; Data splitting module, used to split text data into sample data containing different features and generate difficult samples and random samples; Manual extraction module, used to manually extract enzyme activity parameters from difficult samples to generate golden benchmark data; Automatic extraction module, used to automatically extract enzyme activity parameters from random samples based on a large language model; The optimized extraction module is used to integrate the outputs of the manual extraction module and the automatic extraction module to build an optimized enzyme activity parameter extraction framework; The database module is used to store the extracted enzyme activity parameter data.

[0011] Furthermore, the document filtering module performs screening based on keyword matching or a document screening algorithm based on domain knowledge.

[0012] Furthermore, the automatic extraction module includes: The prompt word engineering unit is used to design and optimize prompt words to drive large language models to extract enzyme activity parameters; The original output unit is used to generate the initial extraction results through the large language model; The post-processing unit is used to further process and format the initial extraction results to generate structured enzyme activity parameter data.

[0013] To achieve the above-mentioned purpose, the present invention also provides an electronic device, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the above-mentioned enzyme activity parameter extraction method based on a large language model.

[0014] To achieve the above objectives, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for extracting enzyme activity parameters based on a large language model.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention proposes a method and system combining manual extraction with automatic extraction based on a large language model, which realizes the function of efficiently extracting enzyme activity parameters from complex biomedical literature. The system significantly improves the accuracy of literature parsing and data extraction through prompt word optimization and post-processing technology, and can provide efficient data support for researchers in the field of life sciences. Through performance evaluation and feedback mechanism, the system can continuously optimize prompt word design and model parameters to ensure continuous improvement in extraction effect. The present invention can quickly and batch process a large amount of biomedical literature, greatly reducing manual workload, and improving the efficiency and accuracy of data collection. It is particularly suitable for database construction and information collation in the field of enzymology research; it is also suitable for applications such as literature management, data mining, and knowledge extraction in the field of life sciences. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of the enzyme activity parameter extraction method based on a large language model of the present invention; Figure 2 It is an architecture diagram of the enzyme activity parameter extraction system based on a large language model of the present invention; Figure 3 It is a structural schematic diagram of the enzyme activity parameter extraction system based on a large language model of the present invention; Figure 4 It is a schematic diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0017] The present invention is described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the features of the following embodiments and implementations can be combined with each other.

[0018] The present invention proposes an enzyme activity parameter extraction method and system based on optical character recognition (OCR), large language models (LLMs) and prompt engineering, which is used to extract key data from enzyme activity research literature. Specifically, it includes the steps of literature preparation, OCR processing, data splitting, manual and automatic extraction, prompt optimization, data post-processing, performance evaluation and final database construction.

[0019] See also Figure 1 and Figure 2 The present invention provides a method for extracting enzyme activity parameters based on a large language model, comprising the following steps: (1) Biomedical literature filtering: Filter out literature related to enzyme activity from the biomedical literature database.

[0020] Over the years, enzymology research has accumulated a large amount of data on the classification, properties, applications, and physiological and biochemical information of enzymes. These functional parameters and other characteristics are integrated and stored in the BRENDA database (BRaunschweig ENzymeDAtabase), providing a valuable reference resource for the scientific community. BRENDA is currently the world's most comprehensive online information system for functional, biochemical and molecular biological data related to enzymes, metabolites and metabolic pathways. Its core information is manually extracted from original scientific literature in databases such as PubMed and Scopus. It currently contains more than 137,000 documents and about 3 million pieces of information, covering 84,000 enzymes. Through the BRENDA service, the present invention obtained a set of biomedical literature identifiers and screened out 3,450 documents from them, which recorded the experimental results of enzyme functional parameters. This document pool is used for subsequent document optical character recognition (OCR) processing and data extraction.

[0021] (2) OCR processing of biomedical literature: Perform optical character recognition on the screened literature to generate text data.

[0022] In order to make the content in the document readable and understandable by large language models (LLMs), the present invention uses the Mathpix OCR tool to convert PDF documents into data samples in Markdown format. Mathpix Convert API (Mathpix PDF to Markdown Application Programming Interface) is selected for its advanced performance in recognizing mathematical formulas, texts and tables, and it can convert the recognized content into multiple formats, such as LaTeX, Markdown and plain text. In the present invention, the advantage of using the Markdown format is that it uses a consistent markup language and standardizes elements such as titles, lists and links, which makes it easier for both humans and machines to parse the content in subsequent steps.

[0023] The OCR conversion formula is expressed as:

[0024] Among them, MD ij Represents the Markdown format content of the jth unit on page i of the document, PDF ij Represents the PDF of the jth unit on page i of the document; Mathpix OCR (Optical Character Recognition) is used to convert the text and tables of the PDF document into an easily parsable format.

[0025] (3) Data segmentation: Split the text data obtained in step (2) into sample data containing different features to generate difficult samples and random samples.

[0026] Specifically, after OCR processing of the documents, the system divides the Markdown data samples into three parts: "random samples", "difficult samples" and "remaining samples". Among them, "difficult samples" correspond to the documents containing the most enzyme function parameter data, while "random samples" are documents randomly selected from the document pool. These two types of samples are used in the evaluation of subsequent manual extraction and automatic extraction experiments. In this way, the system can evaluate the performance of the extraction process in documents of different complexity.

[0027] The formula is:

[0028] Among them, S represents the Markdown data sample, MD random represents randomly selected literature, MD hard Indicates a high-difficulty sample, MD rest represents the remaining samples.

[0029] (4) Construction of golden benchmark data: Artificial enzyme activity parameter extraction is performed on difficult samples to generate golden benchmark data.

[0030] The gold benchmark was established by the research team manually extracting useful data from 36 difficult sample papers, with the main goal of being able to systematically collect and organize key enzymatic parameters such as K m , K cat and K cat / K m The conditions under which these parameters were measured are also recorded, including enzyme name, source organism, substrate, cosubstrate, temperature, pH, and any mutations or variations. m , K cat and K cat / K m When searching for a table with the keyword “kinetic” in the title, first check for tables that contain the keyword “kinetic”, since enzyme activity data are often presented in tabular form. If no relevant table is found (which is a rare case), search for “K cat " to locate the relevant description.

[0031] In addition, enzyme names usually end with "ase" and are usually spelled out in the title or text of the article. Mutants are recorded with the gene mutation code, and the wild-type enzyme is recorded as WT. Substrates and co-substrates should be recorded exactly as given in the text, and temperature and pH values ​​should be taken from the table or the "Materials and Methods" section.

[0032] (5) Using a large language model driven by prompt words, the random samples are automatically extracted for enzyme activity parameters through prompt word optimization, and the automatic extraction results are post-processed to generate structured enzyme activity parameter data.

[0033] For "difficult samples" and "random samples", the present invention develops an automatic extraction pipeline, which includes prompt engineering and post-processing procedures. Prompt engineering uses large language models (such as GPT-4 and Claude 3.5 Sonnet) for design and testing to help extract key information from the literature. The post-processing program is responsible for organizing and cleaning the output data generated by the large language model, and finally converting it into a standardized data framework format.

[0034] Prompt word engineering: The optimization process of prompt words is carried out through multiple iterations of improvement. The evaluation of each prompt word adjustment is not based on the entire data set, but focuses on the problem sample. Prompt words are tested using random samples and difficult samples, samples with a recall rate below 50% are manually analyzed, and targeted text prompts are added for each error. The optimized prompt words include task descriptions, operating instructions, and format references.

[0035] Examples of prompt words: "Please extract the enzyme name, source organism, kinetic constant (kcat , K m , K cat / K m ), pH and temperature”; "Please identify and extract information on enzyme mutants from the literature and experimental data on their effects on enzyme activity".

[0036] Automatic extraction process: The automatic extraction process continuously improves the extraction effect of LLMs on target literature through prompt word optimization technology. The system can automatically identify and extract enzyme names, source organisms, substrates, kinetic constants (such as k cat , K m , K cat / K m ), pH value, temperature, and mutant variation. The final automatic extraction system after prompt word adjustment can be applied to all data samples to generate an enzyme activity database.

[0037] The formula for data extraction is expressed as:

[0038] Among them, E is the extracted enzyme-related information set, including enzyme name E mame , Source: BioE species , kinetic parameter E kcat 、E km , Experimental conditions E pH , E temp and mutant information E mut .

[0039] The present invention uses large language models such as GPT-4 and Claude 3.5 in the extraction process. The design of prompt words is the key to achieving efficient automatic extraction. The extraction process is divided into the following steps: A. Prompt word optimization: Use random samples and difficult samples to evaluate prompt words, and manually analyze and optimize prompt words for samples with a recall rate below 50%; the recall rate is obtained by the amount of correctly extracted data / the total amount of data in the golden benchmark data. In each optimization, the text gradient is gradually increased to solve specific problems. Although the entire dataset is not evaluated for each prompt word modification, this gradual optimization process effectively improves the extraction performance.

[0040] B. Task description: The task part of the prompt word describes the extracted fields, such as enzyme name, source organism, substrate, K m , k cat , k cat / K m , temperature and pH etc.

[0041] C. Instruction part: The instructions include all matters needing attention in the process of prompt word optimization, such as data format, identification of specific terms, etc.

[0042] D. Format reference: A standard output format template is provided in the prompt word to ensure that the data structure output by the large language model is consistent and convenient for subsequent processing.

[0043] Data post-processing: The post-processing program is responsible for organizing and cleaning the output data generated by LLMs into a standardized data frame format to support subsequent numerical comparison and evaluation. The program uses the Pandas library and mainly processes tabular output with 13 columns. The specific processing steps are as follows: 1. Space cleaning: remove extra spaces and blank lines, and ensure that all missing values ​​are marked as NA (not available); 2. Field validity cleaning: clean up the values ​​with wrong units in some columns and mark them as NA; 3. Numerical format cleaning: Use regular expressions to identify scientific notation and error expressions and convert them to standard formats; 4. Unit conversion: Extracted K m , K cat and K cat / K m The values ​​are standardized by unit, K m Unified as mM, K cat For -1 , K cat / K m for .

[0044] Through these steps, the raw extracted data are converted into a consistent and comparable standard format to obtain structured enzyme activity parameter data.

[0045] (6) Performance evaluation, prompt word optimization and final automatic extraction: Combine the manual extraction results with the automatic extraction results, and use the golden benchmark data to calibrate and optimize the enzyme activity parameter data obtained by automatic extraction to obtain the final enzyme activity parameter data.

[0046] The performance evaluation procedure was used to verify the accuracy of the model, especially when evaluating the precision and recall of data extracted from LLMs. The evaluation procedure compared the output of the automated extraction pipeline with the enzyme data entries in the BRENDA database to quantify the extraction performance against the Golden Benchmark.

[0047] Performance indicators include: Precision: The amount of data correctly extracted / the total amount of data extracted by the model; Recall: the amount of data correctly extracted / the total amount of data in the golden benchmark data; F1 score: The harmonic mean of precision and recall.

[0048] The F1 score calculation formula is:

[0049] Among them, P is the precision and R is the recall.

[0050] Based on the performance evaluation results, the system will continuously optimize the prompt words and solve various problems to obtain more accurate output. After the prompt word optimization is completed, the final automatic extraction process will be applied to all data samples to generate the final enzyme database.

[0051] The optimized prompt words are as follows: “Please make sure to extract the complete kinetic constants from all tables in the literature”; “Please focus on identifying and extracting experimental conditions with mutant information and their effects on k cat and K m The impact of value".

[0052] With this optimization, the automated extraction system is able to automatically extract complete and accurate enzyme function data from the literature without the need for complex initial setup.

[0053] During the performance evaluation process, the system compares the output of the automated extraction pipeline with the results of manual extraction and verifies it using a golden benchmark. Compared with traditional manual data extraction methods, the automated extraction system of the present invention demonstrates higher speed and accuracy, reduces human errors, and significantly reduces the time and cost required to extract data from the literature.

[0054] Through the automatic extraction pipeline, researchers can quickly obtain structured enzyme activity data, greatly improving scientific research efficiency.

[0055] See also Figure 2 and Figure 3 The present invention provides an enzyme activity parameter extraction system based on a large language model, which mainly includes a document filtering module, an optical character recognition (OCR) module, a data splitting module, a manual extraction module, an automatic extraction module, an optimization extraction module and a database module.

[0056] Specifically, the literature filtering module selects literature related to enzyme activity from massive biomedical literature based on keyword matching or literature screening algorithms based on domain knowledge. Literature screening adopts a rule-based keyword filtering method combined with a rule base built with domain knowledge. The screening formula is as follows:

[0057] Among them, D represents literature, K i (D) indicates whether document D contains keywords i, w i is the weight of the corresponding keyword, and S(D) represents the score of the literature related to enzyme activity. When S(D) is greater than a certain preset threshold T, the literature is considered relevant and passes the screening.

[0058] The optical character recognition (OCR) module is used to perform optical character recognition on the screened documents and generate text data.

[0059] Specifically, the selected documents are converted from PDF format to processable Markdown text format through optical character recognition (OCR) technology. This module can recognize charts and formulas in the documents and convert them into structured data for subsequent processing.

[0060] The present invention first uses the Mathpix OCR tool to perform optical character recognition (OCR) on the enzyme activity research literature in PDF format and converts it into Markdown format. This conversion process is achieved by the following steps: Select OCR tool: Mathpix OCR tool was selected based on its advantages in recognition accuracy and efficiency of mathematical formulas, texts and tables; Mathpix supports output in multiple formats, ensuring that the converted files have a consistent structure for subsequent processing; File upload and processing: Upload the PDF file to be processed to Mathpix, and generate a Markdown file after OCR processing; the Markdown format can retain the structure of the text and facilitate machine parsing.

[0061] The data splitting module is used to split the text data into sample data with different characteristics, which are divided into difficult samples and random samples. The text data after document conversion will be split into different data samples for different extraction methods. The system divides the samples into "difficult samples", "random samples" and "other samples", where: Difficult samples: Literature samples with high complexity that are difficult for the model to process directly, requiring manual annotation and extraction; Random sample: a sample of common documents, automatically processed by the system; Other samples: Data samples used to supplement and ensure the comprehensiveness of the assessment.

[0062] The manual extraction module is used to manually extract enzyme activity parameters from difficult samples according to the standard operating procedure to generate golden benchmark data. For difficult samples, the manual extraction module extracts enzyme activity parameters such as k according to the pre-set standard operating procedure (SOP).cat , K m The extraction of these parameters mainly depends on the analysis and calculation of the experimental results. The manually extracted golden benchmark data is used to calibrate the output of the automatic extraction module.

[0063] Typical formulas for manual calculations are:

[0064] Among them, V max is the maximum reaction rate, E is the enzyme concentration, k cat is the catalytic rate constant of the enzyme, K m is the Michaelis constant and [S] is the substrate concentration.

[0065] Manual extraction process: When creating the golden benchmark, the research team manually extracted key enzyme activity parameters from 166 papers. This process followed a detailed standard operating procedure, including enzyme name extraction, biological source extraction, substrate and co-substrate extraction, kinetic constant extraction, and reaction condition recording. Among them: Enzyme name extraction: By searching for keywords (e.g., “ase”) in the article title, abstract, and text, the full name and abbreviation of the enzyme were recorded; Extracted from biological sources: usually appears with the enzyme name, record the full Latin name, and note the variants; Substrate and co-substrate extraction: Based on the reactions described in the literature, the names of substrates and co-substrates and their reaction relationships are recorded in detail; Kinetic constant extraction: Focus on the contents marked with the keyword "kinetic" in the table and extract K m , K cat and K cat / K m value; Reaction condition records: including pH value, temperature, etc., obtained by consulting the Materials and Methods section.

[0066] The automatic extraction module performs automatic extraction based on a large language model, including a prompt word engineering unit, a raw output unit, and a post-processing unit; among them: The prompt word engineering unit is used to design and optimize prompt words to drive large language models to extract enzyme activity parameters; The original output unit generates the initial extraction result through a large language model; The post-processing unit further processes and formats the raw output to generate structured enzyme activity parameter data.

[0067] The automatic extraction module uses a large language model (such as the GPT series model) to process random samples. This module interacts with the large language model through prompts to automatically extract enzyme activity parameters from the literature. The design of prompts must fully consider the structure and language characteristics of the literature to guide the model to output accurate extraction results.

[0068] The prompt word engineering unit iteratively optimizes the prompt words to improve the extraction effect of the model. The specific formula is:

[0069] Among them, P t represents the prompt word in the tth iteration, T represents the current input document text, f(T) is the prompt word generation function, Represents the random adjustment amount in the prompt word optimization. As the prompt words are optimized, the model gradually converges to the best extraction effect.

[0070] Automatic extraction process: Develop an automated extraction pipeline to extract data based on the capabilities of Large Language Model (LLM). The main steps include: Prompt Engineering: Design and optimize prompts for specific extraction tasks to improve the extraction accuracy of the model. By gradually adjusting the prompt content, more targeted processing of problem samples is increased.

[0071] Post-processing procedures: The initial extraction results generated by large language models are often unstructured text. In order to facilitate subsequent data storage and analysis, they need to be post-processed. The post-processing module formats the initial results into structured data, such as CSV or a format acceptable to the database. The output results of the model are cleaned and organized, converted into a standardized data frame format, and the data is processed using the Pandas library. This includes cleaning invalid fields and null values; standardizing number formats (such as scientific notation); and unifying units to ensure consistency of kinetic parameters.

[0072] The optimized extraction module is used to integrate the outputs of the manual extraction and automatic extraction modules to construct an optimized enzyme activity parameter extraction framework. In order to further improve the system performance, the present invention proposes an optimized extraction framework, which combines the manual and automatic extraction results and uses the manually extracted golden benchmark data to correct and optimize the output of the automatic extraction module. The optimization process includes iterative optimization of prompt words and tuning of model parameters.

[0073] The optimization process can be expressed as an optimization problem:

[0074] in, are the prompt words and model parameters, is the loss function, The gold standard for artificial extraction. is the enzyme activity parameter predicted by the model. By minimizing the loss function, the system achieves continuous optimization of the prompt words and the model.

[0075] Performance evaluation: To measure the extraction effect of the system, the performance evaluation module compares the automatic extraction results with the golden benchmark data and uses the following indicators for evaluation: Accuracy P (Precision):

[0076] Recall R (Recall):

[0077] F1-score:

[0078] Among them, TP represents the number of correctly extracted parameters, FP represents the number of incorrectly extracted parameters, and FN represents the number of missed parameters. Through these evaluation indicators, the system can continuously adjust the prompt words and model parameters to achieve the best performance.

[0079] The database module is used to store the extracted enzyme activity parameter data.

[0080] The database module will store the extracted enzyme activity parameters in a structured database to support subsequent data analysis and use. The database module uses standard CSV format or other database-compatible formats so that researchers can efficiently query and analyze the extracted enzyme activity data.

[0081] The final form of the system includes a user-friendly interface that allows biomedical researchers to easily operate the system, including: Document upload: Users upload target documents to the system, and the system automatically performs OCR processing and format conversion; Data selection: Users can select the specific parameter type to be extracted (such as K cat , K m wait); Result presentation: The extraction results are presented in a standardized data frame format, and users can choose to export them to multiple formats (such as CSV, Excel).

[0082] The system of the present invention has good scalability and can integrate new OCR tools or updated LLM to cope with the ever-changing document formats and data extraction requirements. In addition, users can set up the system through a simple interface without complex technical background.

[0083] Corresponding to the above-mentioned embodiment of the method for extracting enzyme activity parameters based on a large language model, the embodiment of the present application also provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for extracting enzyme activity parameters based on a large language model. Figure 4 As shown, it is a hardware structure diagram of any device with data processing capability where the method for extracting enzyme activity parameters based on a large language model provided in the embodiment of the present application is located, except Figure 4 In addition to the processor, memory, DMA controller, disk, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiment is located may also include other hardware, which will not be described in detail, generally based on the actual functions of the device with data processing capabilities.

[0084] Corresponding to the aforementioned embodiment of the enzyme activity parameter extraction method based on a large language model, an embodiment of the present invention also provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the enzyme activity parameter extraction method based on a large language model in the above embodiment is implemented.

[0085] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capability and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0086] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0087] The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A method for extracting enzyme activity parameters based on a large language model, characterized in that: The following steps are involved: (1) Screen out literature related to enzyme activity from the biomedical literature database; (2) Perform optical character recognition on the selected documents to generate text data; (3) splitting the text data into sample data containing different features to generate difficult samples and random samples; (4) Extracting artificial enzyme activity parameters for the difficult samples to generate golden benchmark data; (5) using a large language model driven by prompt words to automatically extract enzyme activity parameters from the random samples through prompt word optimization, and post-processing the automatic extraction results to generate structured enzyme activity parameter data; (6) Combining the manual extraction result in step (4) with the automatic extraction result in step (5), the enzyme activity parameter data automatically extracted are corrected and optimized using the golden benchmark data to obtain the final enzyme activity parameter data.

2. The method for extracting enzyme activity parameters based on a large language model according to claim 1, characterized in that: In step (3), the difficult sample is the literature containing the most enzyme function parameter data; and the random sample is the literature randomly selected from the literature pool.

3. The method for extracting enzyme activity parameters based on a large language model according to claim 2, characterized in that: In step (5), the prompt words are: "Please extract the enzyme name, source organism, kinetic constant, pH and temperature in the table"; "Please identify and extract enzyme mutant information and experimental data on their effects on enzyme activity in the literature"; in the prompt word optimization process, random samples and difficult samples are used to evaluate the prompt words, and samples with a recall rate of less than 50% are manually analyzed and optimized. In each optimization, the text gradient is gradually increased to solve specific problems.

4. The method for extracting enzyme activity parameters based on a large language model according to claim 3, characterized in that: The recall rate is obtained by dividing the amount of correctly extracted data by the total amount of data in the golden benchmark data.

5. The method for extracting enzyme activity parameters based on a large language model according to claim 3, characterized in that: In step (5), the post-processing operations include: cleaning invalid fields and empty values, standardizing digital formats, and unifying units.

6. An enzyme activity parameter extraction system based on a large language model, characterized in that: include: Literature filtering module, used to filter literature related to enzyme activity from biomedical literature; An optical character recognition module is used to perform optical character recognition on the screened documents to generate text data; Data splitting module, used to split text data into sample data containing different features and generate difficult samples and random samples; Manual extraction module, used to manually extract enzyme activity parameters from difficult samples to generate golden benchmark data; Automatic extraction module, used to automatically extract enzyme activity parameters from random samples based on a large language model; The optimized extraction module is used to integrate the outputs of the manual extraction module and the automatic extraction module to build an optimized enzyme activity parameter extraction framework; The database module is used to store the extracted enzyme activity parameter data.

7. The enzyme activity parameter extraction system based on a large language model according to claim 6, characterized in that: The document filtering module performs screening based on keyword matching or a document screening algorithm based on domain knowledge.

8. The enzyme activity parameter extraction system based on a large language model according to claim 6, characterized in that: The automatic extraction module comprises: The prompt word engineering unit is used to design and optimize prompt words to drive large language models to extract enzyme activity parameters; The original output unit is used to generate the initial extraction results through the large language model; The post-processing unit is used to further process and format the initial extraction results to generate structured enzyme activity parameter data.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the enzyme activity parameter extraction method based on a large language model as described in any one of claims 1-5 above.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for extracting enzyme activity parameters based on a large language model as described in any one of claims 1 to 5 is implemented.

Images